Recently, motion analysis really helps athletes improve their performance. Since the systems of motion analysis are expensive and expertise in biomechanics or kinematics is necessary, it is difficult for athletes themselves to use them. In this work, we propose a motion analysis system with acceleration sensor. The proposed system is inexpensive since the system consists of inexpensive components. Besides, only two devices for analysis are required by concentrating on analyzing the motions of arm swing and foot strike. Experimental results demonstrate that the proposed system is useful for athletes to analyze running form.

Internet of Things (IoT) has facilitated the connection of many smart devices via internet. Recent cyberattacks have shown that resource constrained IoT nodes are easy prey that lead towards compromising the secrecy of the data and vulnerabilities could be exploited remotely to take control of safety-critical systems. Photoresistor sensors have applications in IoT systems, such as smart street lighting, intelligent cameras, light activated smart consumer electronics, smart home, smart healthcare, etc. Building hardware security primitives, such as True Random Number Generator (TRNG), based on the intrinsic properties of photoresistor would be a novel direction to develop cost-savvy IoT security primitives. Therefore, this paper proposes a TRNG prototype that is devised from uncertainty presents in photoresistor sensors. The proposed TRNG prototype does not require any complex interfacing for preprocessing the weak signal, thereby reducing the unnecessary delay and the recurring hardware cost. The proposed prototype employs the novel approach of additive scrambling that aids to sample sensors at a higher rate. The proposed TRNG has an average random bit generation rate of 8 kbps that is better than the recent work in the literature. The quality of randomness was validated by 15 test batteries of NIST STS test.

Approximate arithmetic circuitry refers to a class of arithmetic circuits that achieve power and area efficiency through the intentional introduction of inaccuracies into the output behavior of the circuits. Further, low-power adiabatic computing techniques have been proposed as a method for reducing power dissipation. To demonstrate the potential benefit of hybridizing adiabatic and approximate computing, we selected a group of approximate adders and redesigned them using Clocked CMOS Adiabatic Logic (CCAL).We simulated our novel class of adiabatic approximate adders using a 45nm process and Cadence Spectre. The results of our simulations show that the hybridization of approximate and adiabatic techniques reduces power dissipation by 40-56% at 50 MHz compared to approximate CMOS implementations. We conclude that the hybridization of adiabatic and approximate computing would be an excellent design choice for applications with rigid power and area constraints that also boast a tolerance for noise.

Cable television operators are considering the introduction of IP broadcasts that distribute video signals by IP multicast. We have studied 4K・8K multi-channel IP broadcast on the FTTH transmission line. But, it is a issue that the in-building transmission system in an apartment has not been FTTH. Therefore, we have prototyped an in-building transmission device based on the DOCSIS standard, which enables IP communication on a coaxial cable. Then, we confirm 4K・8K multi-channel distribution and evaluate packet loss rate using AL-FEC.

In this paper, we propose a new demapping approach for the multiple-input multiple-output aided layered division multiplexing (MIMO-LDM) system. The new demapping approach demaps the core layer (CL) signal using a part of enhanced layer (EL) signal and processes the other part of EL signal as noise. The new demapping approach has a near optimal performance and acceptable demapping complexity that is independent with the modulation order of EL signal. The simulation results also shows that the new demapper outperforms the original Gaussian approximation (GA) demapper on CL performance at low injection level (IL) and high code rate.

In the IEEE802.11 WLAN, the TCP throughput unfairness will appear when multiple hosts concurrently communicate with a single access-point (AP). In this paper, we propose a TCP fairness control method using the transmission delay in WLAN. The target throughput is introduced as the equal throughput, which will be dynamically updated with measured throughputs. Then, the delay is controlled by the PI feedback control such that the measured throughput is equal to the target. The effectiveness is confirmed through experiments using our elastic WLAN system testbed, where the hosts have achieved the similar throughputs.

We have been conducting research on an advanced digital terrestrial TV broadcasting system (advanced ISDB-T) that can transmit UHDTV content. The advanced ISDB-T can provide both mobile service and fixed service using hierarchical transmission similar to ISDB-T. Furthermore, the transmission capacity is increased by using a large-sized FFT. Although large-sized FFT degrades speed tolerance in mobile reception, the advanced ISDB-T can mitigate this influence by selecting an appropriate scattered pilot (SP) pattern. We analyze the influence of large-size FFT in mobile reception and evaluate the reception performance caused by differences in SP patterns through computer simulation and field experiments.

The cost for deploying wireless power transfer (WPT) technology into products still remains high due to the order-made design process. To alleviate the engineering cost to adapt to each use-case, a magnetic resonance coupling based reconfigurable 2-D WPT system was recently proposed. In this system, modules communicate with each other via the Bluetooth Mesh protocol. However, this approach typically suffers from errors due to radio interference. To improve robustness, We proposed a topology construction protocol based on ``in-band'' communication. Through simulations, it was discovered that the topology construction can be executed in 2 seconds for a 30 module configuration.

The power sensor driven without a battery using the energy harvesting is developed. The power monitoring system handled in this paper uses the developed home appliance power sensor. However, the power monitoring system does not take measures against hardware failures of gateway. Therefore, home appliances are unable to be monitored when failures occur in the gateway. We propose a method of redundancy. Our proposed method is achieved by communication in which the power sensor does not distinguish the gateway and alive confirmation messages between gateways. We confirmed that failover is executed when a failure occurs.

Today there is exponential growth and convergence in technologies like Internet of Things (IoT), machine learning and other forms of Artificial Intelligence leveraging the way we collect, and analyze data from smart sensors. A Digital Twin (DT) Technology is the creation of the virtual replicas of physical IoT devices that helps to monitor the applications, analyze data collected and predict future behavior and performance of devices. In this paper, we propose a semantic DT based IoT model with support at the edge with container technology to mimic the IoT devices and enhance the interoperability of the heterogeneous IoT devices.

While more consumer devices become smarter with voice-based natural interfaces, this trend poses such a problem that a user might frequently experience undesirable responses from nearby devices, i.e., seemingly uncoordinated and often redundant responses. In this paper, we present the lightweight adhoc group management leveraging inaudible sound and RF local-connectivity in room area, thereby enabling to efficiently suppress the uncoordinated interactions of devices upon a user's voice command via instant group communication.

This paper proposes a new kind of cybersecurity systems, named Botnet Defense System (BDS), which defends a network system against malicious botnets. The unique feature is that a BDS uses white-hat botnets to fight malicious botnets. The BDS monitors a network system. If detecting malicious botnets, it launches white-hat worms and constructs white-hat botnets in order to destroy the malicious botnets. A basic strategy is also proposed. The primary factor is the secondary infectivity of the white-hat worms. If it is high, the BDS only have to launch a few white-hat worms. Otherwise, it should launch as many white-hat worms as possible. The validity of the strategy was confirmed through simulation with agent-oriented Petri nets.

This paper introduces a practical problem on railway systems and shows an application of IoT wireless temperature sensors as a solution to the problem. The paper shows the results of monitoring temperatures inside housings storing equipment for Kyushu Shinkansen operation in a Signal and Communication House of Kyushu Railway Company.

This paper presents a technique that uses generative adversarial networks (GANs) to remove unnecessary patterns from the captured image of the object. These patterns are superimposed onto B-ch images when an image is captured, for the purpose of acquiring a depth map. They become unnecessary after the depth map has been acquired. Our experiment showed that the patterns can be removed more effectively by using a G-ch image in which the pattern is not superimposed, in addition to a B-ch image. From the results, we demonstrate the effectiveness of the technique that uses GANs

In this paper, we provide a new HDR video system using a dual exposure camera. In order to prevent ghost artifacts caused by misalignment of motions, multi-stage convolutional neural network (CNN)-based motion estimation is proposed. In the proposed method, true motions and occlusion maps are estimated and then a HDR video is synthesized with motion compensated images. It provides not only robustness against exposure levels but also flexibility to extend easily search ranges by stacking stages with same CNN coefficients. Experimental results indicate that the proposed method is effective to estimate true motions so that ghost artifacts are eliminated well.

Shape from Focus (SFF) is one of passive optical methods for estimating 3D shape of an object. In SFF, a large number of 2D images with different focus levels are required. The number of images may affect the complexity and the accuracy of the results. In this manuscript, a Gaussian process regression (GPR) method is proposed to get 3D shape from the minimum number of 2D images. The proposed method (SFF.GPR) is applied to fit focus curves, which are obtained by applying one of focus measure operators. Experimental results demonstrate the effectiveness of the proposed method.

It is common in color imaging to use a single-sensor system with a color filter array (CFA), especially a Bayer CFA. We propose a new imaging system with a pastel-color CFA. In this system, the color plane reconstruction is realized with an abstraction of the white-like component from RAW images. In this paper, we evaluate the effect of sensor noise on various imaging systems. In our experiment, the optical color filter and sensor noise are simulated using computer calculations. With these results, it is recognized that the proposed method is less affected by sensor noise.

Whether you are the CEO of a Fortune 500 company in Manhattan, the chief administrator of a metropolitan trauma center, or the parents of web savvy teenagers in the heartland, your important information needs protection. When it comes to safeguarding highly sensitive government information, the National Security Agency protects the United States' most critical information and systems against cyber-attacks through hardening and defending the cyber infrastructure. NSA has a proud history of designing and fielding secure cybersecurity solutions. Commercial Solutions for Classified (CSfC) is an extremely important component in NSA's commercial cybersecurity strategy. US national security customers need secure access to data anytime and anywhere. CSfC solutions leverage current commercial technology in accordance with NSA-approved architectures to design solutions for these sensitive missions. This presentation will highlight the many activities involving CSfC while also serving as a foundational introduction for those who are unfamiliar with the initiative. The target audience is everyone interested in learning how they can better protect their sensitive information.

A key frame extraction algorithm combining image entropy and perceptual hashing is proposed. Firstly, one frame with the largest image entropy is extracted from every three frames, and a set of candidate frame sequences whose data is one-third of the original frame is obtained. Then, the key frame sequence is determined by using the entropy difference of the candidate frames. Finally, the perceptual hashing algorithm is used to remove redundancy and gets the final key frame sequence. Through the experimental,the extracted key frames can represent the video content more comprehensively and less redundant frames.

In recent years, software in devices such as smartphones and tablets has become increasingly multifunctional, and the use of OSS has become essential. In software development using large-scale OSS, it is important to report defects to appropriate personnel promptly. In this paper, we propose a method to classifying defect reports into appropriate categories using fine-tuned BERT and visualize self-attention information. In the evaluation, category classification was performed using defect reports of the actual OSS project. The F1 score was 0.87, which indicated that high-accuracy classification was possible. Also, the visualization results show that category-specific words can be extracted.

In this work, we present the development of a companion robot that can provide company to a person watching television. To establish guidelines for developing the utterance function for the robot, we analyze the utterances of pairs of people watching TV together, which were then classified into eight categories. We observe that human-human dialogue while watching TV most often starts with "Disclosures" utterances that express one's own feelings. Moreover, we found that utterances with high response rates could be classified as "Questions," "Edifications," and "Confirmations". Based on the analysis, implemented an utterance function corresponding to "disclosure" related to TV program.

Surveillance cameras are commonly used for security purpose. However, most of them are used for verification after incidents happen. In this paper, we propose a method for proactive video surveillance system using human pose estimation. Our method estimates anomaly scores of human actions using the video descriptors (human pose and bounding box) extracted by pose estimation and tracking methods. We estimate human poses, apply PCA, estimate GMM parameters, and finally calculate anomaly scores based on the GMM. We evaluate our method and compare with higher-level recognition-based method. Experimental results demonstrate effectiveness of human pose for video anomaly detection.

Heterostructures have become essential constituents of most advanced electronic devices. Heterostructures are of great interest because the motion of charge carriers possibly controlled by modifying energy band profiles of constituent materials. In this paper, we are discussing an upright fabrication structure of Heterojunction device. We have used SILVACO TCAD to demonstrate 2D vertical structure design and simulation of Binary semiconducting material Silicon Carbide (SiC) Heterojunction diode. We have used Silicon Carbide (SiC) and Silicon to form n-n type heterojunction diode. IV characteristics simulation results were conducted over -10V to +10V range.

In this paper, we propose a novel hardware friendly image dehazing method to improve the image quality and save the execution time simultaneously. To obtain estimated airlight in real time, we select the local patch with the current densest haze to estimate the airlight. With TDM search strategy and fixed-point guided filter, the gate counts of hardware architecture are greatly reduced. In experiments, our proposed method outperforms other dehazing algorithms in terms of SSIM and CIEDE2000. The hardware architecture trades off dehazing accuracy against throughput, which can obtain high-quality restored images with a high-throughput of 166.67Mpixels/s at 500MHz frequency.

We propose a new traffic-sign detection method based on a weakly-supervised multi-purpose single convolutional neural architecture. The base classification network used is the very light convolutional architecture, MobileNetv2, which is used as a region proposal network and a classification network. The method attained few milliseconds for single image testing and averaged about 55 milliseconds on 800x1300 resolution, while maintaining an acceptable accuracy. This method takes advantage of weak supervision which completely eliminates the time required for dataset annotation. We trained and tested our technique on two datasets: the German Traffic Signs Recognition Benchmark(GTSRB) and the German Traffic Signs Detection Benchmark(GTSDB).

This paper is concerned with the implementation of a monitoring system for multi-channel/multipoint broadcasting signals transmitted through a terrestrial UHD Headend system based on ATSC 3.0. Implemented monitoring system measures the quality of jitter and delay of stream transmitted from the Headend, monitors additional services such as ESG and subtitles, and performs remote web monitoring function through various smart devices.

How to input characters without mistakes is one of very important points for the interface, when we consider usability of electronic devices such as personal computers and smartphones. Therefore, in this paper, we examine whether we can classify mistypes for Japanese input and inform the users how they make mistakes using "Attention" and "LSTM". We made the Neural Network to classify the mistype data and inform how they make mistakes. As a result, the correct answer rate of the Neural Network was 91% and we were able to know where point the Neural Network focus on by putting in "Attention".

We propose a method for measuring the flow of people using information from smartphones because if it is possible to collect information from smartphones, it is possible to realize a more detailed measurement the flow of people as to what kind of action each person takes. When the experiment by the prototype was done, it could measure to some extent such as time which there are many people or few.

In Korea, ATSC 3.0 based terrestrial UHD broadcasting service is on the air and emergency alert service is also being developed. In addition, the Ministry of Science and ICT (MSIT) plans to expand its emergency alert service for public media, such as digital signage, public transportation and public facilities by utilizing terrestrial UHD broadcasting. This paper presents the design and implementation of emergency alert gateway that can provide emergency alerts and disaster information via TV, companion devices and digital signage.

The Self-driving vehicle is an important issue. One of the biggest advantages of self-driving vehicles is to reduce traffic accidents. The goal of our efforts is to provide more convenient and safe transportation. Therefore, we present a "high-accuracy mapping design based on Multiview images and three-dimensional LiDAR point clouds" in this paper. The proposed system for 3D LiDAR combines color camera and three-dimensional space coordinate corresponding to the two-dimensional space coordinates technology.

AVS2 is the latest generation of video coding standard independently developed by China. We propose a parallel hardware structure of ASIC for AVS2 intra prediction encoding. This architecture by analyzing software algorithm, according to the characteristics of the ASIC implementation of parallelization, designs the DC, a planar, horizontal and vertical Angle prediction hardware structure of the parallel processing, and finished the calculation of SATD. After logic synthesis using the SMIC 28nm standard cell library, simulation results show that the proposed architecture of intra prediction for 300MHz, real-time processing 1920x1080@60fps sequence of images, extremely suit for VLSI HD encoder.

This paper presents a low-power fully digital multiplying delay-locked loop (MDLL) based frequency synthesizer for GPS-L4 band. In the proposed MDLL, the supply noise sensitivity and RMS jitters are mitigated by implementing the core logic of frequency-operated digital low-dropout regulator (DLDO) in the main loop. The proposed MDLL in a 40 nm CMOS process exhibits an in-band phase noise of -107 dBc/Hz at 100 kHz offset with < 1ps of RMS jitter while consuming maximum power of 0.96 mW at nominal supply voltage of 1.0 V.

In IEEE 802.11ax networks, multiple uplink wireless-connectivity services are supported by an uplink multi-user multiple-input and multiple-output (MU-MIMO) technique. However, inappropriate multi-user selection for uplink transmissions may degrade the network throughput performance. In this study, we propose a method for uplink multi-user (UL-MU) transmissions in IEEE 802.11ax networks. In the proposed method, stations are clustered on the basis of their transmission times, to efficiently utilize channels for UL-MU transmissions. From the performance evaluation, we demonstrate that the proposed method improves the network throughput performance by improving the channel utilization.

5G services have already been launched by more than 50 telecom operators in 28 countries in 2019. Four operators in Japan will also be ready for starting 5G commercial services in 1Q/2020. Superior wireless features or 5G; (1) eMBB: enhanced mobile broadband up to 20Gbit/s, (2) URLLC: ultra-reliable and low latency communications less than 1ms, and (3) mMTC: massive machine type communication over 1 million devices/km2, open a door to realize a new world, that we had never experienced before. 'Augment' is a key word for 5G services. In this talk, some examples of 'Augment's will be given; how sports, city walks, and entertainments will be augmented and innovated by 5G technologies will be presented, and the audiences will be experienced such a new world by watching several demonstration videos. In addition to a new real experience for personal services, 5G is also strongly expected to solve variety of social issues, such as autonomous driving, remote control of construction machine, and remote surveillance with 4K/8K video. Low latency feature of 5G is essential for such critical missions. KDDI research inc. has been studying various advanced technologies, which supports the basis of the 5G services. In our optical access technology using the analog transmission technique, we have recently succeeded in a 1024QAM transmission for mobile fronthaul links for the first time. In our post-quantum cryptography technology, breaking several world records in cryptanalysis were reviewed. We have recently proposed the fastest algorithm for fully homomorphic encryption, which might be utilized for privacy-preserving machine learning as a service. Finally, tele-existence as a rather future oriented demonstration was presented, which is a robotic system that expands the presence of human beings. An experience of remote traveling to Ogasawara island 1500 miles away from Tokyo, where the traveler touched a sea turtle with a real sense of touch through an avatar robot, was demonstrated. A new world explored by 5G is widely forecasted. 5G is just in front of us. KDDI implements various initiatives and aims to provide an unlimited world in the coming 5G era.

In this paper, a deep-learning based pupil tracking technique is developed for application of visible-light wearable eye trackers. By the YOLO based deep learning detection, the proposed pupil tracking method predicts the centers of pupils effectively in the visible-light mode. By testing the pupil tracking performance with the developed inference model, the precision is up to 80%, and the recall is close to 83%. Besides, the average horizontal and vertical pupil tracking errors of the deep-learning based design are only 4 pixels, which are much less than those of the previous ellipse fitting based design at the same visible-light conditions.

Although the market demand of displays bright enough to allow the diffusion of readable information against a very bright outside world is important, in particular in the cockpit (including avionics and automotive application), existing technologies still do not supply the desired brightness combined with very low power consumption, wide field of view, large eyebox and long lifetime. This paper introduces a new study aiming to develop a high resolution microled microdisplay able to reach 1 Mcd/m².

To realize a recommendation system that combines the benefits of a human recommendation (field experts and salesclerks) and a computer recommendation (recommendation algorithms, such as in e-commerce), we constructed an emotion model that appropriately represents customers' emotions during a recommendation by a human. For a typical scene of a recommendation by a human, we focused on a bespoke scene of tailoring clothes via interaction with a tailor. An interview and evaluation experiment allowed us to construct an emotion model that appropriately expressed the emotions evoked in the recommendation process.

Three-dimensional human pose estimation can be effectively applied to teleoperation of humanoids, thanks to its intuitive and easy usability. However, the human pose wrongly inferred can cause critical and dangerous situations for robotic operation. In this paper, a reliable human pose estimation method is presented. Initially, 3-D joint positions are inferred by a conventional pose estimation method using stereo images. The joint positions are refined by Kalman filtering followed by median filtering. The motor control value for each joint is smoothed with an IIR filter to perform stable and flexible operations.

Electronic low-cost devices are an important part of the market nowadays, for example, the RTL-SDR dongles. These electronic devices are an option for getting an RF signal on a computer for around $20 USD. However, the manufacturers of RTL-SDR dongles do not provide a complete data sheet for their products, causing uncertainty among the consumers. In this work, we make a comparative study of some RTL-SDR dongles to provide insightful information about the real behavior of these devices. Our results show how some dongles are more sensible for perceiving signals on some bands than others, especially those with metal enclosures.

Recently, the next-generation DTTB(Digital Terrestrial Television Broadcasting) for UHDTV(Ultra-High Definition Television) broadcasting is being researched and developed in the world. For the Japanese next-generation DTTB for UHDTV broadcasting, LDM-BST-OFDM(Layered Division Multiplexing - Band Segmented Transmission - Orthogonal Frequency Division Multiplexing) transmission scheme is proposed. In this paper, maximum ratio combining method which is performed after sub-carrier demodulation is proposed for LDM-BST-OFDM transmission scheme. In this paper, the reception characteristics of the proposed scheme is evaluated by computer simulations and the required CNR(Carrier to Noise Ratio) of fixed reception can be improved by the proposed scheme.

In this paper, an energy consumption minimization scheme that adjusts both the task offloading ratio to MEC server and the CPU operating frequency of smart device based on a Dynamic Voltage and Frequency Scaling (DVFS) technique under application delay constraint is proposed.

The scheme that combines ISDB-T (Integrated Service Digital Broadcasting - Terrestrial) with LDM (Layered Division Multiplexing) has been proposed as the way to shift to the next-generation DTTB (Digital Terrestrial Television Broadcasting). However, when LDM is applied, the receiver is complicated. In this paper, the performance is evaluated by computer simulations when the receiver was simplified and the modulation of the upper layer was changed. As the results of computer simulations, the receiver can be simplified without deteriorating the performance. The required CNR (Carrier to Noise Ratio) of the lower layer can be improved by the proposed scheme.

We have conducted IP multicast distribution experiments using MPEG-H MPEG Media Transport in managed CATV networks to promote 4K/8K UHDTV. In this paper, we report an experiment in which we uplinked 4K/8K UHDTV content to an internet exchange through existing dedicated lines and facilities, and then conducted multi-channel distribution simultaneously through commercial fiber to the home (FTTH) lines provided by CATV managed networks. We confirmed that the UHDTV contents could be received by prototype STBs through the FTTH environment. The experimental result shows the possibility of commercializing nationwide simultaneous and multi-channel distribution of UHDTV contents through existing IP network facilities.

Network embedded systems including Internet of Things (IoT) systems oblige developers to conduct frequent tests for various configurations, which require a significant amount of time and money. Moreover, the tests might not be performed due to the impossibility to acquire target devices produced by other factories or manufacturers. We propose a testing environment on a computer composed of target device emulators, and a repository of system configurations and test cases. Therefore, the environment performs tests for various configurations automatically and drastically improves the test efficiency.

Internet of Things devices are the largest, fastest growing, longest-lived, and least secure category of Internet hosts. Insecure consumer IoT devices threaten other home devices and distributed denial of service attacks from IoT botnets pose a growing threat to Internet services. We examine threats to, and by, IoT devices and develop a framework to protect consumer IoT devices from compromise and limit the threat they can pose in-turn to other home and Internet hosts. We indicate prioritized steps that ISPs and manufacturers can take to improve home IoT security. Framework components are hosted at https://github.com/CIRALabs

Although machine-to-machine (M2M) frameworks, e.g., OneM2M and LWM2M, supports to interwork together, they may only provide uni-directional data convergence, i.e., from LWM2M to OneM2M. This paper studies and investigates the overhead of achieve bi-directional data exchange between two M2M frameworks using MQTT. In most conditions, the synchronization between two M2M frameworks may not add too much extra burden except the resource write procedure.

Smart speakers and compatible IoT devices are emerging. We have developed an iHAC Hub realizes coordinated operation of IoT devices can be controlled by communication protocols. This paper proposes a cooperation method for M2M devices by using iHAC Hub and smart speaker. In the proposed method, we provide a framework for cooperating environment sensing IoT devices and smart speaker compatible IoT devices. The iHAC Hub generates voice commands for controlling IoT devices by using the cloud service and reproduces them toward the smart speaker to realize cooperative control of the smart speaker compatible IoT device based on the environment information.

In this paper, we consider a delay sensitive multimedia data dissemination in V2I network based on Systematic Network Coding (SNC). We propose an algorithm that improves SNC based data transmission by transmitting network coded data at the idle time of packet transmission between handovers. The experiment results using actual traffic data show that the proposed approach can reduce not only the overall decoding error rate but also decoding failure rates.

It is vital that video encoders remove redundant information present in the spherically projected 360◦ video frames. To this end, this paper proposes a novel weighted residual technique and an integrated technique between the proposed weighted residual technique and a state-of-the-art adaptive quantization technique. The proposed methods adapt to the spherical characteristics in order to reduce the magnitude of the redundant information. The tests of the weighted residual technique and the integrated technique with All-Intra configuration using Versatile Video Coding (VVC) test model, produce average bit rate savings between 0.91% and 1.35% with regards to various spherical objective quality metrics.

This paper proposes a wind noise reduction method based on a U-shaped neural network. While the U-Net is developed for image segmentation, it is constructed by using the spectrograms of noisy signals as the input feature, and it is trained to estimate the ideal ratio mask between a pair of noisy and clean signals. The performance of the proposed method is measured in terms of SDR, SIR, and SAR. As a result, it is shown that the proposed method provides a higher average SDR, SIR, and SAR than conventional methods such as MS-based and NMF-based methods under various SNR conditions.

The amount of video data is so large that efficient video compression is required for the storage and transmission. Intra prediction is one of important components in video compression. In this paper, we examine various Convolutional Neural Network (CNN) structures to estimate optimal intra prediction mode as Most Probable Modes (MPMs). Moreover, we investigate several combinations of the MPMs obtained by the CNN and MPMs derived from High Efficiency Video Coding Test Model (HM). From these experimental results, we find that using 6 MPMs from both CNN and HM with moderate number of channels or kernel size is preferred.

This paper presents a novel image fusion-based TM method. We use Gaussian mixture model clustering algorithm to estimate the dark and bright distributions in the luminance histogram of the input HDR image. Then, we generate two LDR images using two locally-adaptive TFs obtained by the components of each distribution. Finally, the output image is produced by the image fusion technique employing a brightness weight and a local contrast weight. The experimental results show that the proposed algorithm achieves high performance compared to state-of-the-art methods in terms of detail preservation and brightness adjustment.

It is vital that video encoders understand and remove the redundant information present in spherically projected 360ᵒ videos. To this end, this paper formulates a spherically adaptive objective function that incorporates novel adaptive quantization and weighted residual techniques to reduce the magnitude of redundant information while minimizing the distortion. Furthermore, this paper identifies the residual weighting function and the most optimal quantization parameters that are used to encode 360ᵒ videos. The proposed method is tested with test model of Versatile Video Coding (VVC) with All-Intra configurations. The obtained results exhibit an average bit rate saving of 3.18%.

Due to the non-linear behaviour of the power amplifier (PA), the amplification of signals with fluctuating envelopes leads to distortion inevitably. These nonlinear effects can be counteracted by the digital pre-distortion (DPD). In this paper, a method of quantization noise and narrow-band dithering signal for improving feedback signal distortion of the DPD system is presented. Before quantization, the technique of injecting a narrow-band dithering signal to reduce these effects and be filtered by the digital narrow-band filter after analog to digital converter (ADC). Experiment shows this method can be used to obtain more accurate data to effective DPD.

UAVs were developed for military purposes such as unmanned reconnaissance and surveillance. Due to recent advances in technology, it has been diversified into civilian and special mission products. In the UAV flight control system, a flight control engine, a signal processing engine and a wireless communication modem are installed separately. Such drone electronic modules are inefficient in terms of weight, cost and power consumption. In this paper, we implement SoC for FCS for ultra-lightweight and low power consumption of flight control system and present the results of integrated Command/Control and Down Link modem implementation.

Bootstrap is the first transmitted signal in ATSC 3.0 transmission frame. The first OFDM symbol of bootstrap signal is identical for all transmission frames, and has repetition patterns in time domain. The first bootstrap symbol is used for time/frequency synchronization and channel estimation. And the last three symbols contain the transmission frame parameters. In this paper, the machine learning-based bootstrap signal detection technique is proposed. For the machine learning technology, deep neural network (DNN) structure is considered. The proposed technique can detect the first bootstrap symbol with highly large time offset.

The Non-Metallic Connector (NMC) was introduced several years ago. In this new iteration, the NMC has been redesigned to accomplish data transfer with an optical interface and wireless power transfer with the use of micro-inductors.

This paper proposes an inaudible communication method with high reception ratio which is resistant to ambient noise using selective dual frequency. In addition, the proposed system measures the noise frequency band of the surrounding environment, and can increase the reception rate by selectively using frequencies that do not occur. This communication method enables short-range communication in place of Bluetooth, NFC, and Wi-Fi in the speakers and microphones internal to the smart devices and home appliances. Experiments with selective dual frequency show that the reception ratio is 95% within 7.5 m distance.

In this paper, we propose a novel decoding strategy for superposition modulation (SM) for cooperative IoT system. Unlike the conventional method where the SIC (successive interference cancellation) decoding is applied, whose performance degrades when the decoder fails to detect the main signal, we propose a novel decoding algorithm which derive the LLR (log likelihood ratio) directly from the received signal. The new decoding scheme performs well even when the main signal detection fails and outperforms conventional SIC based decoding method by more than 2 dB in the fading environment.

NTMobile (Network Traversal with Mobility) is our original technology capable of providing the NAT traversal and secure end-to-end communication.this paper proposes a system to set the WoT (Web of Things) server in a private network, and the server is accessed from different private networks by using NTMobile.According to this method, it is possible to construct a secure system preventing server attacks such as DDoS attacks, and also almost all Web programming technologies can be used as they are.We have built the web server in RaspberryPi and set in the private network, and confirmed remote controls via the Internet.

In this study, we propose a new data management platform using blockchain that contributes to provide reliable IoT services by guaranteeing both integrity and authenticity of the real-world data. In the proposed platform, the origin attempts to guarantee authenticity of the data by requesting the near terminals to generate the same types of data. By comparing the data and the update history of the data among the terminals, the proposed platform attempts to verify authenticity of the data because the tendency of the update history of the generator should be almost the same as that of the cooperators.

This paper proposes an efficient and secure data management system by using the blockchain technology to issue certificates for IoT devices. Certified IoT devices ensure user privacy and data integrity while providing efficient data storage and retrieval through blockchain-based certificate management for IoT. The certificates issued to the IoT devices hold data generated by the device and store it on the blockchain. We evaluate our proposed system on Ethereum to measure transaction cost and speed according to the amount of data stored by issuing and verifying certificates.

This work proposes a framework for integrating middleware Ginga implementations into Internet of things (IoT) networks, through support to the message queuing telemetry transport (MQTT) protocol. The Ginga's API has been extended and some reference applications were developed, in order to validate the proposed framework.

It has been shown that a loudspeaker coupled to a long narrow pipe, through which sound is radiated through a series of small spaced apertures can have narrow directivity. This paper describes a process by which this concept is practically applied for use in a soundbar for delivering sound to a listener indirectly by reflection from room surfaces. A model was developed and used to guide the process which revealed various sound quality issues for a basic design. With modifications a successful approach with narrow directional sound of an acceptable quality was found. Its architecture and performance are reported.

High Efficiency Video Coding(HEVC) employs both past or future frames when encoding the current frame in a video sequence. This paper proposes a framework for using virtual reference frames, to achieve increased coding gains in the longterm for repetitive scenes in static camera scenarios.

This paper proposes a new method that integrates a deep learning based object detection network into traditional background modeling to detect moving objects. The proposed method allows us to efficiently identify candidates that contain moving objects while only setting a small number of anchors in the moving area of the image through guidance from the traditional background modeling method. This paper overcomes the disadvantages of conventional background modeling methods and conventional deep learning based object detection methods in terms of dynamic backgrounds and objects' motion states.

Recently, the convolutional neural network (CNN)-based denoising methods adopt single-scale features to separate image structures from the noisy observation. Single-scale features, however, have limitation in covering the full characteristics of image structures at different scales. In this paper, we propose a novel denoising network that makes use of the multi-scale feature pyramid where each feature map represents the characteristics of image structure at different scales. We then combine these multi-scale features to obtain the contextual information and utilize it to effectively generate clear denoised results. Experimental results show that our network achieves superior performance to other conventional methods.

Augmented and mixed reality (AR/MR) contents need 3D data in various models to interact with human users. However, creating a 3D model is a complicated and expensive process involving 3D acquisition, 3D reconstruction, and rendering with computer graphics techniques. To solve that problem, we use an autoencoder to extract local information. We then train the latent space in a generative adversarial network (GAN). The GAN takes local context from the latent variable and then generates a point cloud of various robust shapes. Our method can generate a novel 3D model that can significantly save computational load to render AR/MR contents.

Emergency Services' (ES) vehicles objective is to attend accident scenes in a fast and safe manner. This task is becoming challenging due to the increasing population, drivers' distraction, emergency cases and subsequent road traffic. Drivers' distraction and low situational awareness could cause major delays to ERS. This paper presents the results of 50 drivers' investigation related to their activities during traffic congestion aiming to inform the design of a prototype Head-Up Display (HUD) which could provide ERS drivers with real-time manoeuvring information and proposed routing options superimposed directly to the vehicle's windscreen, improving ERS drivers' response times and safety.

This paper describes a multi-timescale waveform descriptive grammar that assists researchers and engineers in syntactically labeling signals and developing test and control signals for consumer electronics design, diagnosis, and validation. The proposed waveform description and synthesis framework can accommodate the working knowledge of electrical and computer engineers engaged in consumer electronics design and validation projects, by offering seamless integrations of the proposed framework into their existing workflows.

The development of display technology and semiconductor technology enables mass production of large, high-definition OLED panels. Meanwhile, there was previous research to improve the sound quality, and the exciter speaker was attached to the panel, acting as the diaphragm of the dynamic speaker. This provides viewers with the clear, direct sound from the panel, maximizing immersion. However, the strong directivity of OLED panel excitation is also an advantage and a disadvantage. In this study, acoustic emission characteristics were improved by using multi-channel OLED panel speakers.

The iris region matching would the core of the visible-spectrum gaze tracker. The design challenge is much more difficult than the one for the traditional infra-ray gaze tracker due to the imperfection of the eye images. In practical application environments, the iris regions on the eye images may be corrupted by the reflection light, and the matching process is also affected by the eyelids and wrinkles near the eyes. In this paper, we further improve the fitness function used for iris region matching. The experimental results show that the new one can effectively prevent the matching results from being dominated by the wrong textures near the eyes.

In this paper, we developed a tour mining system "Tour Miner" as web application based on the assumption of actual use by applying the methods and results proposed by our research team in the previous research. The web application can select travel records which match individual preferences or interests. We then released our web application to the public at the special exhibition at Yamaguchi Prefectural Museum for 1 month. According to the feedback from 425 users who used our Web Application, the selected travel records matched with their interests, and those travel records were useful in their travel plans preparation.

In recent years, following the development of automated driving (AD) and advanced driver assistance systems (ADAS), stereo cameras which are able to measure distance are advancing accordingly by employing wide-angle field of view to detect various objects. At the same time, demand for low cost stereo cameras which can detect far objects also exists. Also, detection distance performance degradation due to low resolution CMOS image sensors and distortion from wide-angle lenses remain as problems when increasing the stereo camera field of view (FOV). In order to solve such problems, we developed a multi-shift stereo camera. In the multi-shift method, the left and right CMOS image sensors are shifted their center with respect to each lens optical axis center. By using this method, far object detection performance can be preserved while obtaining a wider FOV without the use of an expensive image sensor. We also developed a monocular vision detection employing top view image subtraction method, enabling detection of pedestrians and cyclists in the monocular vision region which is created at the outer region of the camera FOV. Furthermore, by using the disparity information from the center region to estimate the camera posture, we can achieve high accuracy for detection and distance measurement at the monocular region.

The gaze tracker has become an important human machine interface, while it remains challenging for the users to move their head randomly. That is, the system performance appears significantly inferior due to the motion of a human head. However, the head movement compensation cannot be simply formulated as an eye detection problem. The minor error of the eye corner detection algorithm leads to an unacceptable result for limbus circle matching since the incorrect eyeball model is adopted. In this paper, we further explore the 3-D eyeball model anchored by the inner eye corner point. The relative location between the eyeball center and the inner eye corner is analyzed. This feature is used to guide the eyeball model construction as well as the limbus circle matching. The experimental results show the proposed approach allows a range of the estimation error for eye detection and the final limbus circle matching performance has been remarkably improved.

We have applied various kinds of 3D CG technologies to increase the Field Of View (FOV) and visibility of displayed images for AR HUDs. As a results, we successfully developed impressive 3D CG technologies for HUDs with wide FOV and high visibility

3GPP determined mmWave as the primary technology to achieve a high data rate in 5G-NR. Millimeter-wave channel suffers from high pass loss due to high frequency. As a result, accurate beamforming is required to overcome the high path loss effect leading to a sparse channel. Because of its sparse nature, mmWave channel estimation is challenging. In this paper, we are studying the spatial consistency of the non-stationary mmWave channel using the Correlation Distance Matrix (CDM) for channel estimation. Our observation shows that spatial consistency can be used to decrease channel estimation complexity provided that the update distance kept small.

In an orthogonal frequency division multiplexing (OFDM), the pilot signal is utilized for the channel estimation (CE). The scattered pilot (SP) reduces the number of pilot signals. However, the interpolation is required. The linear interpolation (LI) is achieved simply, but many errors are occurred in a multipath fading. The combined complex time frequency interferometry (CTFI) achieves the good CE by averaging the channel impulse response (CIR) of the different time window. However, a large number of pilot signals is required. To solve these problems, in this paper, we propose the CE based on the LI-CTFI for a SP-OFDM.

This project is aimed to develop an indoor locating system using Bluetooth low energy (BLE) that able to track personal asset in indoor venues, where the GPS signal strength is uncovered. Furthermore, the performance of the system is evaluated in term of signal speed, energy consumption, and sufficient distance of deploying both transmitter and receiver. For the system development, ESP32 Wi-Fi, BLE supported modules, MQTT and React application are used as the front-end and back-end of proposed system. In summary, the indoor locating system is capable of showing the tracking device using a 2D map on the browser.

The growing adoption of Internet of Things technologies leverages the development of complex IoT applications. This situation is particularly visible in the consumer electronics realm, where integrating products from different manufacturers requires considerable effort. An increasing demand for rapid IoT application development arises, as it is time consuming and demands a multiple competences. This work aims to simplify the creation of ready-to-run semantic services from a high-level description. We proposed and implemented an architecture, and successfully tested the results in an existing configuration, with a reduced development effort.

we present a quick response time of smart lights in the local control method without cloud. Second, when multiple home IoT devices are used collaboratively like IFTTT or smart speakers, we can still observe the slow response time because of the cloud.For the fast and private control of home IoT trigger-action services, we demonstrate how a local control method accelerates the response time for the camera triggering action service.

We proposed a process-based anomaly detection for Cyber-Physical System. We focus on the relationship of the execution order of message passing between the devices in the MQTT network. We performed a `bird's eye' monitoring technique to message passing in the network with process mining. Using a process model called as a process tree, we check the deviations of behavior. CPS is a closed-loop system, we can easily detect the anomaly. However, it is hard to perform forensic analysis. Therefore, we proposed an analysis method which comes as an advantage of this approach.

Currently there are few studies on performance evaluation of containerized services under constrained resources. To enable efficient service provisioning on IoT gateways, this paper presents the empirical evaluation of Docker Swarm (DS), a container solution, on resource constrained devices such as Raspberry Pi3 boards. We have used multiple open source Intrusion Detection System (IDS) and Deep Learning (DL) based data analytic solutions in our experiments to evaluate creation time, CPU utilization, and memory usage. Our results reveal that creation time and memory usage of service container are critical factors for dynamic provisioning in constrained environments.

NHK has developed a high-resolution virtual reality (VR) system for highly immersive visual experiences. Eight 4K laser projectors cast 12K×4K-resolution VR images onto a cylindrically shaped screen for 180-degree public viewing. Three-dimensional audio is reproduced using 15 loudspeakers and 2 woofers. We shot images with three synchronized 8K cameras and stitched them together to produce high-resolution VR images. We featured our images using this system at our annual event open to the public, confirming that the system will provide viewers a highly realistic immersion for next-generation television viewing styles.

Light field imaging is the best given the amount of information it provides when compared to conventional photography, it captures angular and spatial information from all directions. Despite that, its limited resolution poses great difficulty in the use of these enormous capabilities. In this paper, we tried to lessen the impact of this drawback by using a deep-learning algorithm. We adopted the idea of dividing the process into disparity estimation and color prediction. Our system was trained to double the angular resolution fast and accurately. Experimental results demonstrate that our system can reconstruct high-quality images faster than the state-of-the-art techniques.

This paper proposes a fast convergence algorithm for adaptive noise cancellers with SNR-based stepsize control. SNR-based stepsize control reduces interference by the noise-cancelled signal in adaptation. A second SNR initially controls the stepsize, followed by a first SNR to promote coefficient growth. The numerator of the second SNR is the noise-cancelled signal obtained as the subtraction result different from the primary signal in the conventional algorithm to accelerate initial convergence. Evaluations with clean speech and noise recorded at a busy station demonstrate that the time until SNR switchover is reduced by 91% compared to the conventional algorithm.

With developed CTU bit allocation algorithms, the video compression's performance improved significantly. However, without the consideration of the human visual system characteristics, they fail to obtain subjectively optimal results. In order to improve the image's subjective quality, we propose a perception-aware CTU level bit allocation algorithm. A perceptual distortion metric with spatial and temporal masking effects is presented firstly, and the rate-perceptual distortion constraint function is then established to obtain the optimal quantization step for each CTU. Experiments show that the proposed algorithm achieves better subjective quality with negligible affection on rate control on RD17.0 of AVS2.

Recently, surveillance cameras are ubiquitous for both real-time monitoring and recording important moments. Temporarily seamless surveillance using many cameras requires increasing amount of human efforts and enormous size of storage. To solve this problem, we proposed a metadata extraction method for object search and a person re-identification. The proposed metadata extraction method can be applied to a wide range of surveillance systems such as search for missing children in public space and crowd monitoring system.

We developed an 8K 119.88-Hz (120-Hz) the high efficiency video coding (HEVC)/H.265 encoder to transmit smoother videos. To ensure backward compatibility with 59.94-Hz (60-Hz) video decoders, the encoder supports temporal scalable coding to partially decode 60-Hz video frames from compressed 120-Hz bit-streams. To evaluate the image quality of our encoder, we conducted subjective evaluation experiments for 8K 60-Hz and 120-Hz videos. From the results, we find that the required bitrate for 8K 120-Hz videos is 85 Mbps.

This study aimed to figure out users' physical condition through a gait classification and the optimized positions on the human body to attach IMU sensors by using the IMU-based spectrogram approach with deep CNN models. Our experimental results show that a single IMU sensor data can successfully predict the subject groups even without requiring hand-craft extraction and selection of features. The classification accuracy can be much improved by a proper choice of sensor combinations and input layout. This allows us to reduce the number of attached sensors on the human body while achieving very high accuracy on the gait classification.

In this paper, we present a computer vision framework that controls robots to auto type in on a mobile device such as an Android phone or an I-pad. The framework consists of three parts: (i) an image undistortion and segmentation algorithm that supports images captured by a top mounted camera or a side mounted camera, (ii) a deep neural network~(DNN) algorithm that can detect the keyboard region and recognize its isolated characters an input image, and (iii) a grid click algorithm to construct data points to compute the image to robot coordinate translation matrix for mobile devices of different sizes.

Speaker de-identification can be used to protect the privacy of a speaker. One class of de-identification schemes uses voice conversion techniques to transform utterances of a source speaker into utterances of a target speaker. This paper presents two new generic attacks on such schemes. The attacks have been implemented for Gaussian mixture model-based schemes. Experimental results show that the first new attack achieves success rates that are considerably higher than the rates of existing attacks. The second new attack shows that these rates can be increased further if the attacker has access to an utterance of the target speaker.

Wireless indoor localization is critical for autonomous agents in modern and future smart warehouses. Millimeter-wave (mmWave) frequencies have been investigated for high-precision localization in recent years for indoor as well as outdoor positioning. We propose machine learning (ML) techniques over a radio map to estimate the location of an autonomous material handling agent used in warehouses. Based on our experimental results we demonstrate that a Multilayer Perceptron (MLP) based positioning achieves centimeter level accuracy with Root Mean Square Error (RMSE) of 0.84m. The proposed localization technique achieves up to 80% lower positioning error compared to state-of-the-art mmWave wireless localization techniques.

We present ThuMouse, a novel interaction paradigm aimed to create a gesture-based and touch-free cursor interaction that accurately tracks the motion of fingers in real-time, allowing users to move the cursor using frequency-modulated continuous-wave (FMCW) radar. ThuMouse regressively tracks the position of a finger, which enables finer-grained interaction. This paper presents the gesture sensing pipeline, with regressive tracking through deep neural networks, data augmentation, and computer vision as a training base. This work builds a foundation for designing finer micro-gesture-based interactions, and we report on a proof-of-concept demonstration showing how our system can function as a mouse.

Automatic content placement is desired for augmented reality applications. Conventional approaches, such as marker deposition or three-dimensional modeling of all user-selected objects, are time-intensive. We herein propose a framework that is cost-efficient when the same content is superposed on the same multiple objects. The framework consists of steps: 1) scanning a user-selected object with an RGB-D SLAM, 2) cropping the object from the scanned images, and 3) training a convolutional neural network on the cropped images and various backgrounds. Our CNN outputs the absolute position and tilt of a device. An important aspect here is data augmentation.

This paper provides the design and implementation of a VR gun controller with haptic feedback for the HTC Vive with recoil adjustability. The primary focus of this design is to provide realistic haptic feedback for games utilizing projectile launching weapons to improve gaming immersion on a level further than simply audio and visual.

Crowdsourcing became increasing popular in recent years, thanks to the explosive growth of online users. However, most crowdsourcing platforms ceased to operate due to the untrustworthy data or lack of motivated participants. The blockchain technologies bring decentralization as the potential solution. This paper introduces reciprocal crowdsourcing, a novel decentralized cooperative crowdsourcing model powered by the blockchain to improve the trustworthy among crowdsourcing participants, who perform transparent collaborative work in the system thereafter. To validate our proposal, we implemented `` Cell Evolution'', a blockchain game, in which the players can build cooperative game worlds on the blockchain with reciprocal crowdsourcing.

In this paper, we present a reliable vertex normal estimation method from sparse point clouds that improves the accuracy of plane-based frame-to-frame registration. We define a face normal reliability measure. The vertex normals are calculated by weighted averaging adjacent face normals based on the reliability. Through the experiments, it is confirmed that the proposed method produces consistent and reliable vertex normals.

A bi-focal geometric phase (GP) lens can be used to improve the depth range of an integral floating display to form switchable depth planes. However, due to a chromatic aberration of a GP lens to form R/G/B sub-images at different locations, a color breaking of floated image can occur. In this paper, we propose a novel method to compensate that chromatic aberration by integrating R/G/B pixels at depths.

In this paper, we propose The Multilayer Perceptron Vector Quantized Variational Autoencoder (MLP-VQ-VAE) to manage the flexibility of controlling the number of z-latent vectors to quantize and embed space sizes efficiently. In the experiments, the MLP-VQ-VAE is applied to quantize spectral envelope parameters from the 48 kHz high-quality vocoder named WORLD. The MLP-VQ-VAE reduces the memory sizes of the representation of z-latent or the length of vectors to quantize and embed space size or codebook size, decreases the Log Spectral Distortion, compares the conventional VQ and the VQ-VAE.

In this paper, approximate Versatile Video Coding (VVC) fractional interpolation filters are proposed. They significantly reduce computational complexity of VVC fractional interpolation with a negligible PSNR loss and bit rate increase. In this paper, an approximate VVC fractional interpolation hardware implementing the proposed approximate filters is designed and implemented. The proposed approximate hardware has less area and up to 40% less power consumption than exact VVC fractional interpolation hardware. Therefore, it can be used in consumer electronics products that require low power VVC encoder hardware.

This paper proposes a novel Structural Similarity Index (SSIM) assisted pseudo-sequence-based (PSB) coding order and prediction structure for improved compression and random-access efficiency for light field videos (LFV). The experimental analysis compares the performance of the proposed method against two recent PSB LFV compression methods in terms of PSNR gain. The results demonstrate the proposed technique yields improved bitrate and better random-access to views compared to existing methods.

Visual positioning is a critical function in applications such as navigation and extended reality experiences. Recently, deep learning technologies, especially classification, had been implemented on the positioning task. However, to acquire a comprehensive positioning dataset and produce a high-performance neural network model is challenging. In this article, we propose to solve the issue by projecting training images from auto-generated 3D point cloud maps. By utilizing branch convolutional neural network (B-CNN) model, the "Zoom-in" equivalent property results in favorable positioning accuracy and successful real-time implementation.

Whiteleg shrimp accounts for the highest proportion in the shrimp export of Vietnam. Yet, in hatcheries, shrimp larvae quantity is still estimated manually. Several approaches were proposed to address this issue but overlapping problem reduced accuracy significantly. In this paper, this problem is addressed by implementing two-phase Mask R-CNN based instance segmentation to segment shrimp larvae for counting purpose. Compared to one-phase Mask R-CNN, the accuracy of counting by applying two-phase Mask R-CNN increased by a maximum margin of 16.1%. Our model had remarkable results, with accuracy ranging from 92.2% to 95.4% for moderate overlapping images.

The big data collected from IoT (Internet of Things) devices and equipment has difficulty sending to the cloud server due to the finite resources. To get over this limitation, edge computing is suggested. In this paper, we propose the data acquisition device (DAQ), integrating collecting and computing functions, programmed to conduct the target works such as sending message to the other DAQ device and receiving data from the several kind of PLC(Programmable Logic Controller). DAQ communicates with gateway using LoRa or WiFi connectivity.

In this paper, we present a novel traffic accident detection method using a spatio-temporal three-dimensional (3D) convolutional neural network. The proposed method consists of pre-activation ResNet and feature pyramid network (FPN) structure. To reduce computational load with preserving the detection accuracy, we propose interval input method. Experimental results show that the proposed network outperforms existing methods in the sense of both accuracy and recall measures.

In this paper a Stress Detection System based on Machine Learning Algorithms (MLAs), keyboard and mouse data is presented. The development of this system is composed by three steps. Firstly, each user performs some tasks while a web application framework collects data from keyboard and mouse. At the end of each task, he/she communicates the stress level in order to create the stress class. Secondly, from collected data, features extraction and features selection procedures through a Neighborhood Component Analysis (NCA) are implemented. Lastly, three MLAs, trained with features as input and stress classes as output, are implemented to detect stress.

The one of initial symptom of dementia is "Decrease of Living Motivation". In this study, the authors developed a system based on internet of things platform for monitoring activities of daily living for elderly people and verified three hypotheses on a relationship between decrease of living motivation and change in activities of daily living. We made experiment with one subject for 10 weeks to monitor activities of daily living and conducted a questionnaire to measure subject's living motivation. From this result, we consider that the decrease of living motivation can be detected by monitoring the change in going out time.

The proposed research aims to help older people maintain their emotional well-being. This research is focused on developing a wearable affective framework, which can help in detecting the emotions of the user in addition to monitoring their physiological signals. The proposed framework is validated using a fall detection algorithm deployed in a custom-built watch wearable, and an emotion detection framework. A dataset of 21700 samples acquired using the proposed framework yielded a maximum efficiency of 97.25 %, 96 %, and 94 %, in classifying the state and emotion classes into Alert, Active and Normal classes respectively, using multi-class SVM model.

Tourism is one of the key revenue generators in communities worldwide. In the present day, traveling has become a lot easier with all the information available through the Internet. However, there are still challenges in identifying the right medical resources while traveling to a new city for the first time. In this research, we propose a privacy-assured framework that can help travelers reach the medical services on their trip on-time during the emergency. Through this research, we have developed a cost-effective tour wearable, We-Tour, that can notify the user if they need to visit a hospital service as required.

With increasing energy problems, energy management becomes more important around the world. As Smart Home Technology is applied to the home, many methods of reducing energy consumption are being studied. We focused on the Energy Storage System in the home for efficient energy management. It is important to efficiently manage the Energy Storage System. Therefore, we propose the Energy Storage System Management Method based on Deep Learning for energy-efficient Smart Home.

Smart City offers a variety of services. Among the various services, energy services play a role in balancing urban energy consumption and supply. However, as energy prosumers emerge, two problems arise. The fluctuation of the energy market can be greatly disturbed, and the forecast of energy demand due to large fluctuation becomes impossible. In this paper, energy data blocks reflecting the energy data elements are created in real time to make stable energy market. In addition, energy data blocks stored in real-time enable forecasting of urban energy demand.

As the fourth industrial revolution and information and communication technology base are growing and the Internet of Things is distributed, a newly set goal is to use the energy of various industries efficiently. This paper is a study to solve the energy consumption problem and to use it efficiently. Specifically, sensors store energy usage in the cloud in real time, analyzing energy efficiency and showing the user. It also provides intelligent services for people to use energy more effectively than traditional methods. This process prevents and monitors energy consumption that is not efficient in real life by mobile and computer.

We propose a calibration-free method for localizing mobile robots using a stereo camera installed on the ceiling. We use both optical images and depth data obtained from the stereo camera to localize the robots. This method achieves 2.3 mm precision at the center of the field of view and 30 mm near the edge when the stereo camera is installed on an approximately 3.2-meter high ceiling. We also show experimentally that it is possible to perform automatic control of a mobile robot based on the location information measured by the proposed method.

When natural disasters occur, there is a possibility of having many injured people in the disaster area. In this study, we proposed a recognition system of individual status to help rescue teams. Employing Unmanned Aerial Vehicles (UAVs) system after disaster occurrence gives many advantages. This study aims to classify whether an individual status is standing, sitting, or lying on the ground by using supervised machine learning. Experiments revealed that the system is able to recognize all three types of individual status with an accuracy of 95.6%.

Digital television (DTV) receivers also suffer from software and hardware errors, which result in malfunction in user premises. This work tackles this problem and proposes a self-diagnosis and behavior correction methodology, which has the potential to keep DTV receivers working, under such errors or wrong configuration data.

Compensation circuit design has been extensively studied in literatures to achieve constant voltage for single load in inductive power transfer (IPT) systems. In this work, we propose secondary-side compensation circuit for IPT system which can obtain load-independent constant voltages for multiple loads. This topology can be beneficial to applications that require difference voltages for multiple loads. In this paper, as a primitive investigation, we demonstrate how to design compensation circuit to achieve constant voltages for two loads. Simulation and experiment results are provided to confirm effectiveness of the proposed scheme.

The rapid advance of Blockchain has been used for the Internet-of-Things to provide reliable distributed infrastructure and applications without a centralized authority. This paper aims to create a new consensus protocol that can be efficiently and resiliently used for the Internet-of-Things based on the Practical Byzantine Fault Tolerance algorithm (PBFT), which is critical for IoT scalability and reliability. This paper analyzes the current implementation of PBFT consensus protocol and proposes a new secure approach for achieving consensus by adding the properties of random hash generation and threshold comparison in order to defend against DoS attacks.

In this paper, we propose a voice interactive home system using CG characters that can freely change its appearance according to the user's request. Specifically, when a human sensor detects a person, the CG character begins talking to increase the chance of conversation with the user. The system can also remotely operate various home appliances in the house with ECHONET-Lite through the CG character by voice dialogue, and also provide life information provision service such as weather and time. In this study, we considered a spoken dialogue system that is easier to talk to than a humanoid type.

To protect critical infrastructure on digital networks, appropriate security controls need to be placed using either purpose-built cybersecurity technologies or embedded security controls in command and control center protocols. Few vendors offer devices that use cryptography to secure the communications power systems communications. There is a need to research and develop a suitable cryptographic module that is capable of mitigating the risk of unforeseeable threats, specifically to DERs and bulk power systems. This paper presents research to improve the security of an electric grid that employs millions of DERs with the goal of securing DER communications

Currently, the most commonly used scheme for identity authentication on the Internet is based on asymmetric cryptography and the use of a centralized model. Centralized model needs a Certificate Authority (CA) as a trusted third party and a trust chain of CA. However, CA-based PKI is weak in the single point of failure and certificate transparency. Our system, called SS-DPKI, propose public and decentralized PKI system model. We describe a detailed scheme as well as application to use decentralized PKI based secure communication. Our proposal prevents storage overhead on the data size of transactions and provide reasonable certificate verification time.

Binary counters are widely used as essential building blocks for varieties of circuit operations, especially for fast multipliers. In this paper, a 7:3 counter design is proposed, which uses 3- and 4-bit stacker to shift the "1" from one side to the other and then finds the "1" in the specified position by accurate sorting. Compared with previous designs, this counter reduces the number of XOR gates and curtails the critical path.

Consumer electronics and embedded systems must ensure trusted authentication and secure communications. Communication session encryption keys are negotiated in a key agreement protocol enabled by a public key/private key pair for authentication. However, embedded systems are limited in their capability to implement public key encryption and client-side authentication. In this paper, we introduce an authentication and session key generation mechanism using Physical Unclonable Functions (PUFs) that extract randomness from device physical characteristics. The method uses PUF models to help with key exchange and creates periodically refreshable. The approach mitigates tampering attacks as no key is stored on device.

The paper presents a smart home energy management system (HEMS) with an optimal source selection algorithm. The proposed system comprising of two input sources, a PV-battery duo source and the main electricity grid, takes into consideration the electricity price information, user load profile, irradiance profile and, the battery power information. The smart HEMS ensures the right energy mix by optimizing the power consumption from the local resources and the main grid in order to improve efficiency and reduce the overall cost of the system. Hardware implementation of the proposed HEMS was successfully tested on a small-scale prototype in the laboratory.

In an Energy Management (EM) scenario, photovoltaic (PV) generation systems could lead to an important cost-saving and "shiftable loads" (e.g., dishwasher, washing machine, cooker hood) play an important role. Among all "shiftable loads", electric boiler has considerable importance since it can be considered as a thermal storage. In this perspective, it is crucial to know the typical usage patterns and the state of charge of this appliance. In this paper, a methodology to identify the electric boiler usage patterns and to estimate its state of charge is presented.

The biggest difference of mobile devices over AC powered is that their power budget is very limited. That is, the CPU resource management solution for mobile devices makes a greater contribution to product quality. Their main algorithm was to choose the minimum energy or the best-fit performance to accommodate a given job. However, in real world, performance and energy are essential to each other and cannot be treated in these independent and consecutive referencing methods. In this document, we would like to give you new ideas for how to consider performance efficiency in CPU resource management solution.

This paper proposes a building energy management model considering prediction errors in load consumption and renewable energy generation. Demand side management has a problem because consumers has difficulty to control power consumption due to the lack of consideration of actual use environment. Therefore, we propose penalty factor to minimize the gap between scheduling power and usual power usage patterns. Since the prediction error causes difficulty in power operation, this paper proposes energy storage system operation method based on reserve portion. The simulation results show the electricity rate of the building is reduced by more than 21% per month.

Low-density parity-check (LDPC) code decoding have never been converted into a quadratic unconstrained binary optimization (QUBO) problem. Thus, a new method is created: one that converts the LDPC code to a QUBO problem, inputs the QUBO problem into Ising machines (computers based on the Ising model that are designed to solve the QUBO problem), obtains the QUBO solution and converts it to a LDPC solution. In formulating this method, one expands the currently scope of studies involving Ising machines, helping current and future researchers unlock its full range of capabilities and possibilities.

The rising demand of computing power in mobile applications leads to the question how suitable FPGAs might be for usage in environments with limited energy resources. One of the most important benefits of FPGAs is their ability for reconfiguration and therefore adding a certain degree of flexibility in the field. In this paper, different, low-power optimized blocks are integrated into a partial slice and investigated upon the related static power consumption. Key components like LUTs, D-FFs, IO units and basic combinational circuitry are subject to power reduction and depict the baseline for further improvements in terms of extended battery runtime.

This session will have an integrated panel on smart healthcare followed by technical presentations.

The increasing trend of using wearable sensors for monitoring various physiological signals has helped researchers to advance the research in the development and application of wearable computing systems. The emotional well-being of a person is as important as an active lifestyle to maintain a healthier balanced life. Affective computing, also known as artificial emotional intelligence, can recognize, interpret and simulate human emotions. In this research, we propose a cost-effective semi-immersive affective monitoring system that can help in analyzing emotion features for monitoring anxiety disorders by monitoring physiological signals of the user and aid in emotion elicitation through a semi-immersive environment.

In this paper, an internet of medical things (IoMT) based unified drug delivery system (iDDS) has been proposed for automatic seizure detection and control. iDDS has mainly two units: seizure detection unit and drug delivery unit. Seizure detection is performed in real time using statistical feature extraction and deep neural network (DNN) classifier. Once a detection is complete, the drug is injected into the target area using piezoelectric actuated valveless double reservoir micropump. iDDS presents an unique piezoelectric actuated double-reservoir based drug delivery system for fault-tolerance as well as better drug control.

In this work, we have presented the validation of a stress detection model using cortisol as the stress biomarker. The proposed model uses two physiological signals: Galvanic Skin Response (GSR) and Photoplethysmograph (PPG) to classify stress in two levels. GSR and PPG signals were collected from a total of 13 participants along with their saliva sample during the duration of the the experiment. We have used 10 out of the 13 participants to train our model. Data of remaining 3 participants was used to test the robustness of the model in distinguishing stressed states from non-stressed states. We have achieved an overall accuracy of 92% with the model achieving precision, recall and f1-score of 93%, 99% and 96% respectively in predicting the occurrences of stressful events. Results indicates the promise of the proposed methodology in accurately detecting the presence of stressful events by generalizing the test data coming from different set of population than that of training data. Index Terms-Galvanic Skin Response

For decades, Calligraphy has been a popular artistic activity in Japan, China, and some countries. To assist its self-learning, we have explored the Calligraphy Learning Assistant System using projection mapping. A learner can practice it by following the letter writing video projected on the paper. In this paper, we present a letter portion practice function. Learners may practice weak portions with the video showing the writing by a teacher. Through applications to 12 novice international students, we confirm the effectiveness, where each student has significantly improved the skill.

This paper proposes an edge camera system using a compressed deep learning model for enhanced Video surveillance Management System of Smart City that analyzes existing recorded video. The proposed edge camera at the end terminal of the Video surveillance Management System send the analyzed image, information and warning to the central system according to the situation analysis based on the information obtained by compressed deep learning for low memory and real-time operation in embedded system. We tested with edge camera installed in street lamp and confirmed stable operating performance and high recognition rate compared to existing system.

The users' behavior in a house impacts the amount of electrical energy consumption in the electrical products of the household; therefore, energy consumption can be optimized by using sense, smart and sustainable products S3 as Social Products (SP) for saving energy in the housing through gamification techniques that can yield behavioral changes. To reduce energy consumption in households, a structure of three steps is proposed for SP that uses a gamification Human Machine Interface in each device to communicate between products and consumer in an SH. Once every social product has its local gamification, a Gamified Smart Home is achieved.

For agricultural cyber physical systems (CPS), we propose an intelligent robotic vehicle that uses reconfigurable hardware to implement neural network inference and cryptographic functions for enhancing system performance and adaptivity. We also present a crop growth model and a detection model of pests and diseases to support its decision-making mechanism.

As part of Intelligent Transportation Systems (ITS) public transportation plays a critical and essential role for the mobility in every modern city. In this paper, we introduce a novel method for the real-time prediction of bus arrival times in the various bus stops over a given itinerary. The proposed approach exploits machine and deep learning algorithms, including optimal least square (OLS) linear regression, support vector regression (SVR) and fully-connected neural networks (FNN). The experimental results obtained show that the FNN approach outperforms, in terms of mean absolute prediction error, both SVR (by 7,62 %) and OLS (for 15,74 %).

Missing people, animals or objects is a common problem. Many tracking systems are based on technologies that are expensive, demands big devices to be carried, or demands often battery charging. Approaches using cameras or RFID systems usually have the problem of requiring to cover the search area with sensors. This work takes advantage of the Internet of Things proposing the Surveillance Swarm, a system for finding targets using the Swarm, which enables opportunistically gather heterogeneous resources of third parties. A use case was deployed and evaluated.

An uneven ground formed by a low-level obstacle whose height is too low from the ground often stalls a robot's navigation in an indoor environment. Few-centimeter differences between an obstacle and a non-obstacle are difficult to be precisely measured in a constant distance at a mobile robot. In this paper, we let a wheeled mobile robot make physical contact onto an object, and use the cliff sensor(s) to characterize an object at firsthand. While adopting a simplified deep-learning architecture, we suggest rapid and accurate obstacle detection in real-time. We implemented our technique on an embedded robot platform.

With the advent of autonomous driving systems, the system verification has great importance for the safety. Since the engineers usually make highly accurate simulations, they require massive computing resources. In addition, they should develop the trustable test scenarios and the evaluation metrics, which may be difficult and time-consuming issues. In this paper, we propose an autonomous driving system verification framework, which can efficiently support cost-effective simulation by means of FMI/DDS and cloud computing. Some formal assessment methods including quantitative evaluation metrics will be given in order to make easy verification of the features of the autonomous driving systems.

This paper presents a commercial implementation of CNN-based classification of parking slot type using around view images. We constructed an extensive dataset composed of labeled 480,000 images acquired on various environments using around view monitoring camera mounted in commercial vehicles. We subdivided parking slot types into ten categories and finally derive three parking slot types for actual applications. To operate the classifier in real vehicle, we designed CNN model suitable for embedded system and implemented it using GPU. In experiments, the classifier achieves accuracy of 94.15% and processing time of 3.67ms per frame on the NVIDIA Tegra CX embedded system.

This work presented an end-to-end training model for parking slot detection in automatic parking systems (APSs), which combined both a line and a point semantic segmentation models based on multi-task learning. The proposed models generate images of entrance line and center points of corners, and are used to determine the coordinates of actual parking slots in the post processing step. The recall, precision and F-measure rate of the proposed method are 92.94%, 99.40% and 96.06%, respectively, which are better than existing state-of-the-art methods with end-to-end training.

The demand for vehicles which conserve energy is increasing. Reducing the weight and power consumption of headlights is also required. Many of the concept cars developed by car manufacturers are designed with thin headlights. In this situation, we have developed the RIR optical system as a new headlight module that can realize ultra-thin body with a lens height of 20 mm and uses less power consumption, 19 W.

This paper presents a simulink model-based approach of using fuzzy logic for accurate detection of PVC beats in ECG signals. Clinical diagnostic criteria were used to develop the ECG fuzzy rule set that classified each ECG beat into the two beat-groups: Normal and PVC. A severity index of the PVC was also formulated. The model was tested using ECG signals from the MIT-BIH database (Physionet). The accuracy of the proposed model is 96.6%. The sensitivity and specificity of the model are 95.6% and 96.8%, respectively.

In the present study we determined the coherence among different locations of EEG, heart rate and respiration with envelopes of songs during listening to music. We observed an overall increase in coherence among different locations of EEG. Our results showed that slow tempo song had the largest impact in making all variables synchronized to its envelope profile. Songs with expected similar cognitive effects produced higher synchronizations of the all measured physiological variables with their envelope. These results show the important role of tempo and cognition of songs in synchronizing different physiological variables.

To best of the authors' knowledge, this is the first work that uses Kriging for early detection of seizure. There is a need for a more time-sensitive approach to seizure detection. Here, we propose a real time seizure detection model in an edge computing paradigm using signals collected through the electroencephalogram (EEG). The EEG signals were de-noising with the Discrete Wavelet Transform (DWT), and then classified using the ordinary-Kriging method which gives a training accuracy of 99.4% and a perfect sensitivity. The presults show a comparable classification accuracy and a lower mean detection latency of 0.85 sec.

Accurate Face Expression Recognition (FER) on a smart phone is useful in many applications which respond to a user's emotional state. We introduce a new lightweight Deep Learning model, MobiExpressNet, for FER. The model relies on depthwise separable convolutions and a fast downsampling approach to keep the model size very small. Our best network model gave an accuracy of 67.96% on the challenging FER2013 dataset which exceeds human accuracy by 2.5%. The MobiExpressNet model size and FLOPs are shown to be over 5 times smaller than the smallest MobileNetV2 model which makes the developed model very attractive for real-time applications.

In this paper, we present a comprehensive literature review of existing big data clustering algorithms and their advantages and disadvantages for various use cases in hospitality industry. We implement the existing big data clustering algorithms and provide a quantitative comparison of the performance of different clustering algorithms for different scenarios. We also present our insights and recommendations regarding the suitability of different big data clustering algorithms for different use cases. These recommendations will be helpful for hoteliers in selecting the appropriate market segmentation clustering algorithm for different clustering datasets to improve the customer experience and maximize the hotel revenue.

In this paper, we propose a vision-based robot learning from demonstration (LfD) system, which uses action and object recognition. In this system, a robot is developed to learn from human-demonstrated actions. The vision-based robot LfD system employs RGB cameras as sensing devices, and integrates object detection achieved by YOLO deep learning architecture and multi-action recognition carried out by an I3D deep learning network incorporating a proposed statistically fragmented approach to separate the overall demonstration into a few sub-actions. Finally, the corresponding robot moving trajectories are subsequently planned such that the actions can be performed successfully.

In this paper, we propose a fall detection method using IR array sensors to inform others of fall and enable rapid treatment after the fall.The IR array sensor enables fall detection that is inexpensive, non-violate privacy, and non-requirement for wearing.The fall detection is performed by classifying actions using machine learning.The learning data are a series of data for two seconds of the temperature distribution acquired every 0.1 seconds by the IR array sensors.We compare a plurality of learning algorithms by their accuracies to select an algorithm of machine learning.

This paper proposes an artificial intelligence (AI) edge computing-based wearable assistive device for assisting zebra-crossing walking of the visually impaired people. The proposed assistive device is composed of a pair of smart glasses, a walking cane, and a waist-mounted box. When the visually impaired pedestrian prepares to cross the zebra-crossing, thus the visually impaired pedestrian will receive the current traffic light signal message. The visually impaired guidance voice service will be provided via Bluetooth (BT). Moreover, when the walking offset occurs on the zebra-crossing, then the visually impaired pedestrian will be reminded through the voice prompt of the earphone.

Since a consumer home generally consists of various IoT-devices which are with different capabilities and from different vendors each other, it is difficult to apply centralized high-performance security solutions to their smart home systems. A blockchain system can be alternative but it is needed to improve the high resource problem of the mining process to apply a blockchain to a smart home system. This paper presents a lightweight blockchain which uses a reverse sequence hash chain instead of POW and then applies it to a smart home system.

As submicron technology scales, SRAM bitcell density increases on the chip. This results in an increase of soft errors due to radiation induced MBU. The probability of MBUs in the SRAM bitcells in 16nm/7nm technology has increased considerably. There are powerful ECC codes which can correct these soft errors but it comes at the cost of extra parity bits, translating into the SRAM size increase. We are proposing an area efficient ECC(72,64), ECC(39,32) code with no extra parity bit cost which can detect and correct adjacent 2-bit error and detect adjacent 3-bit error.

Vehicular networks use a vehicle to infrastructure to improve traffic safety by gathering information from vehicles such as traffic conditions and accidents. This approach is based on the vehicle's location information and the reliability of the data collected from vehicles. At the same time, the voluntary participation of drivers requires privacy protection and compensation for sharing trusted information. In this paper, we propose a blockchain-based decentralized reward solution considering drivers' privacy. The proposed solutions also enable service providers can verify the reliability and integrity of the messages transmitted by legitimate vehicles

Proper security protocols should be regarded as a critical requirement in today's society to prevent adversaries from being able to gain access to valuables. Although numerous preventative protocols have been implemented, the attackers adapt and develop new approaches to infiltrate new technologies. The results of these experiments reviled vulnerabilities in the radio frequency communication of cars and garages in remote keyless entry (RKE) systems as well. In this work, we present a timestamp-based solution to enhance the security of existing RKE systems and demonstrate it through a basic prototype implementation.

In this paper, a new low-power technology for binocular displays is proposed. This technology includes darkening one eye image for the binocular display. Through a psychophysical experiment, we verify that even if binocular images are darkened to some extent, people cannot distinguish them from the original ones. The proposed technology is more efficient than darkening both eye screens to save power without losing visual quality. And we find out a threshold condition where 17.3% more power can be saved without a negative effect on visual quality.

RGB-D cameras have been widely used in various applications but the resolution of depth maps is lower than that of an RGB image, which significantly limits the potential applications of depth maps. The challenge with the depth map super resolution for RGB-D cameras is that we do not have high-resolution depth maps for training models. We propose a deep self-learning approach for color-guided depth map super resolution. Experiments demonstrated that the proposed method achieved a reasonable result even with a low-resolution image (internal data) only for training.

Facial expression generation,especially dynamic facial expression generation from a static natural (expressionless) face image, plays an important role in the fields of entertainment,game,and social communication. Several approaches based on machine learning including deep learning techniques have been developed or proposed for facial expression generation. However, most of them are focused on static facial expression generation. In this paper,we propose a static-to-dynamic model (static2dynamic) based on a 3D conditional generative adversarial network. Herein,the dynamic facial expression image is treated as a 3D image. The effectiveness of the proposed method is demonstrated by generating dynamic happy and angry facial expression images.

A content-based image retrieval (CBIR) system can support radiologists in making clinical diagnosis through image analysis. Multiphase contrast-enhanced computer tomography (CT) images are effective in detecting and characterizing focal liver lesions (FLLs). This study proposes a deep learning method for the CBIR of FLLs using multiphase contrast-enhanced CT images. We use deep convolutional neural networks (DCNNs) to extract the temporal-spatial features from multiphase CT images. The effectiveness of the proposed method was demonstrated through experiments with our multiphase FLL CT dataset, which is called as MPCT-FFL dataset. The mean average precision (mAP) was improved from 0.76 to 0.84.

We propose an RF distance sensor that utilizes the transmission wave itself for the distance measurement between the transmitter and an individual for power level management. The harmonic signal generated by a rectenna is used for the return signal. Since the frequency is different from the transmitted wave, the detector is not saturated by the transmitting wave and can thus be set to high sensitivity. We measured the propagation delay from the transmission timing to received wave for the distance sensing. The time detector by utilizing an equivalent sampling technique suppressed the accuracy of a distance within ± 3 cm.

Audience Meter is an application that estimates the number of attendees in an event throughout the time by applying a state-of-the-art face detection algorithm. In order to detect faces at large distances, high resolution images are required. This makes unfeasible the computation of the algorithm on the device and requires the use of MEC modules from the emerging 5G networks to boost performance. An average boost factor of $10x$ is achieved by using MEC in comparison with CPU devices for high resolution images.

In a typical smartphone, the camera layout and display are different, rendering the gaze often not front-facing. Although various gaze-correction methods have been proposed for this problem, many of them cannot generate sufficiently natural images. In this paper, we propose a gaze-correction method using a deep-generative model. This model can determine the naturalness of the resulting images and learn to provide natural results. And we use super-resolution techniques to generate images that have a larger size and higher resolution than those generated by conventional methods.

Computerized object counting shows potential for conservation population estimates as an alternative to manual counting, but remains unsatisfactory for minuscule objects, such as those in UAV-produced images. We improve UAV data collection by using a novel path planner which shifts altitude to maximize deep learning-based object detection confidences from a Faster Region-Convolutional Neural Network, considering energy consumption trade-offs. Using an empirical altitude - confidence relationship, our adaptive path planner ("DCRPP") adjusts UAV altitude based on confidence, yielding better quality data given energy constraints. DCRPP achieves 11.92% greater accuracy compared to fixed-height methods in our conservation-aimed simulation.

Measuring the heart rate is for arbitrary applications in medicine, sports and psychology. To avoid crucial disadvantages of conservative techniques contact-free measurement techniques were developed recently. The most promising technique is the analysis of humans face video data. While these techniques were developed further they stay overall limited to only one participant simultaneously. As a reason of this a concurrent approach for up to 9 persons using CNNs is proposed and optimized regarded to run time and accuracy.

Due to the significant burden of deep learning computation, it is well known that the AI technologies are hard to be embedded to the edge devices. In order to alleviate the computational burden imposed to a single device, we propose NCDL (Network based Cooperative Deep Learning) methodology in wireless IoT Video Transmission Environments. In particular, the optimal distribution algorithm for the deep learning task is designed by the factors of the number of assisting nodes and the traffic activity based on numerical analysis.

Most consumer products are equipped with methods of network communications, and nondeterministic tests, which are originated from random message exchanges, should be carried out. Therefore, the tests consume much time to design and conduct. For reducing labor of designing test cases, algorithms and tools, which help engineers to convert specifications written in a natural language into semi-formal descriptions, and generate test cases as decision tables, are proposed in the paper. We applied the algorithms and the tools to generate test cases and confirmed that the algorithms and tools were succeeded in generating test cases from a document.

For autonomous navigation or performing missions, robots need a variety of information. Especially, in order to operate in outdoor environments such as weather conditions, night and daytime, multi-type of sensor data is required. In this paper, we propose a multi-modal sensor module for various sensors. In addition, the sensor data can be usefully calibrated. For example, distance values that are difficult to obtain in RGBD images can be measured using LiDAR data. Therefore, the proposed module is useful in most outdoor environments and can be easily installed in various robots.

Fun'iki Navi is proposed for distracted walking accident prevention, where LED light color and sound of FUN'IKI Ambient Glasses express direction, route deviation, and arrival at the destination in navigation system. To verify effects of casual alert of the proposed Fun'iki Navi, the Fun'iki Navi is compared with the Google Maps navigation in terms of a viewing time of the smartphone or downward direction and subjective assessments. As results, the proposed Fun'iki Navi decrease 97.1% of the viewing time of smartphone or downward direction on average. The proposed method realizes society with few distracted walking accidents.

in this study, a new sensor network system that remotely monitors various environmental information of water (e.g., water temperature, pH, dissolved oxygen) affecting growing conditions of aquaculture and cultivation, and that quantitatively estimates degree of plant/fish growth is proposed.

For video coding, since the video signal is ultimately received by the human eye, the visual perception of the human eye cannot be ignored. Thus, we proposed a perceptual RDO scheme based on Lagrange multiplier by considering human visual system characteristics. The temporal perceptual features are extracted by considering the temporal perceptual correlations for each coding tree unit. These Lagrange multipliers are adaptively adjusted using these perceptual features. Experiments demonstrate that with SSIM as metric, it can improve the subjective video coding performance. Compared to the HM16.9, it can achieve 5.6% in average bitrate reductions with negligible subjective quality loss.

An effectively perceptual preprocessing filter method will reduce the sequence's perceptually insignificant details, which can help to alleviate the video coding's pressure and benefit the coding results. In this paper, a perceptual rate SSIM optimization based preprocessing filter algorithm is presented to achieve better video compression performance. The evaluation results confirm that the proposed video filter algorithm has achieved reliable rate-visual quality improvements with -4.35% and -2.96% BD-Rate performance gain on JM18.3 and HM16.9, respectively. Additionally, the subjective quality comparison results show that our algorithm has reduced significant bits with negligible visual difference.

The rise in people's consciousness towards their daily eating habits has attracted significant attention from the field of automatic food analysis. Specifying in Japanese daily food items, which may become helpful for almost all people at every age in society. In this paper, we propose a mobile application that can take the picture of the meal of the user as an input, and output the corresponding categories. The system is based on an improved Faster R-CNN including center loss to increase the classification accuracy.

We proposed a power demand forecasting system with a personalization function. A power demand forecasting system is a system that performs power demand prediction by regression analysis. The screen operation behavior varies depending on the target consumer. The personalization for information systems was proposed, we shown the target value by personalization without effect reflected in the actual system confirmation. In this paper we applied a personalization method from business systems to a power demand forecasting system and conducted experiments to confirm the effects. Through experiments, we confirmed the personalized power demand forecasting system provides screen operations to user's usage classification.

For checking out the absentees strictly, efficiently, and safely, we propose a Classroom Roll Caller system on face recognition which includes MTCNN based face detection and FaceNet based face matching. In working process, user should take a group photo by mobile phone and send it to PC. The system supports two manual detection modes, normal mode and hard mode(working in weak light). Then system will return the call roll result in seconds, and there is a manual correction to make the result right. Face library can upgrade automatically in every roll call. Test proves that the system meet our expectation.

On-board diagnostics (OBD) is important especially for commercial truck providers. Deep neural networks (DNNs) could create powerful data-driven models for OBD. Two factors could contribute to the improved performance of DNNs for this task: large amount of data by truck makers and enabling more variables as input. In this paper, we introduce the implementation of a data recording system able to record indirect measurements with deep learning (DL) besides CAN message. To let our customers easily access the data, the data are uploaded to cloud in real time and could be easily accessed on a user friendly web page.

Digital instruments have a checking tool that can automatically read data, but analog instruments have limitations in collecting data because they are managed directly by the administrator. A lot of research is in progress to obtain data after converting analog data to digital data from analog instruments, but there is a limitation of the environment and cost problem. This paper proposes a system that automatically converts analog data to digital data with portable equipment using image processing algorithms.

This paper proposes a smart ``thing'' which can convert the regular steering to smart steering with the help of the Internet of things. This device works with the touch of the human driver, collects and analyses the physiological data of the person and performs the analysis in a microcontroller. With the help of the analyzed data, the decision of sobriety of the human is made and sent to the car's infotainment as a notification. The blood alcohol prediction is made with an exact level of the concentration present in the human body with an accuracy of approximately 93%.

In the IC design process, wafer map defect recognition is an import part. Existing tests rely on additional analysis of testing result by engineer to determine the failure. Thus it could take an additional amount of time and cannot make adjustments of the process immediately. In this paper, we proposed a reduced-weight architecture to classify failure type based on depthwise separable convolutions. The entire work is verified by using the real-world wafer map dataset (WM-811K). The accuracy is 96.63% in test set.

To investigate cyber attacks accurately, we must collect, manage, and maintain log data on devices related to the system. We herein propose a distributed management method for logs using a blockchain scheme. The blockchain scheme has the following two features. One makes managed data tamper resistant. The other makes access to the same data possible with multiple terminals, thereby increasing access availability. A log management method that adapts these features can provide log data that are guaranteed to be complete when needed. This paper presents our evaluation of the feasibility of our method in a prototype.

Modern applications Internet of Things demand efficient routing protocol to exchange huge amount of information among deployed nodes and control station. Many routing protocols have been proposed but it is challenging to achieve both reliability and energy-efficiency. In this paper we proposed Multi-hope Heterogeneity-aware Energy Efficient Routing Protocol (MHERP) which deals with two level heterogeneous cluster formation for Internet of Things (IoT). MHERP adopts cluster formation based on location and residual energy resources indexing. We further extended MHERP to Multi-level MMHERP to accommodate advancement in network heterogeneity level. Our Proposed routing protocols outperform the existing state of the art protocols.

We have designed and prototyped a lightweight and low-power active personal cooling system which uses the non-toxic gallium-based liquid-metal, Galinstan, as the coolant. The Galinstan absorbs body heat as it is pumped through tubing attached to a cooling vest and after absorbing the body heat, the warm Galinstan is cooled by passing it through the tubing embedded in cold phase-change material in a cold pack. To the best of authors knowledge, the liquid-metal-active personal cooling system is 1/3rd of the weight, 1/8th of the price, and cools for 4 times longer compared to the current state-of-art active personal cooling systems.

This paper presents the design and evaluation of a side-channel detection and exploitation framework that follows a machine learning based plugin oriented architecture thus allowing side-channel research to be conducted on a wide-variety of side-channel sources.

Digital halftone deals with transforming a gray scale image into its printable binary version. In this paper, a generative adversarial network-based model is proposed to perform both halftoning and its effective reconstruction. The GAN model is based on the concept of unpaired image to image translation and the model learns both transformations simultaneously. For optimal training, the model is feed with maximal information patch (termed effective patches) and the model architechture is also modified in accordance to this transformation problem. From results, it has been validated that the model performs with consistent accuracy in both cycles.

In this paper, a new compensation method to reduce luminance variation due to IR-drop in active matrix OLED displays is proposed. This paper decomposes the IR-drop into global IR-drop and local IR-drop, and then estimates the amount of them. The proposed method cancels the IR-drop using the estimated IR-drop. In experimental results, the proposed method decreases the maximum difference of luminance due to the local and global IR-drop by 95.2% and 93.96% respectively. In addition, when the global and local IR-drops are occurred simultaneously, the proposed method suppresses the luminance error between the expected and the measured by 99.32%.

Flash storage is increasingly used in various fields such as mobile devices, internet of things (IoT) devices, etc. Generally, failure rate of a storage device follows the bathtub curve. Though error rate of flash storage such as SSDs follows the bathtub curve, most of SSDs use the ECC scheme with fixed strength against failures. Hence, we propose Flexible Strength RAID (FS-RAID) scheme that employs RAID configuration and dynamically adjusts its strength against failures according to the typical failure rate of the device lifetime. In addition, we propose a method to greatly improve reliability by combining RAID parity and existing ECC.

In this paper, we present an efficient curve rendering method useful in smartphone and mobile device that have dynamic workload changes between CPU and GPU. To do so, we propose a Hybrid scheme that adaptively tessellates a curve into a set of small triangles and sub-curves to render them effectively on GPU. Experimental comparisons show that our scheme not only outperforms other schemes but also reduces the power consumptions. As the result, we can effectively enjoy our scheme in any smartphones and mobile devices regardless of any workload environment.

Nail painting machine is getting prevailing of consumer device, but a smart technique to automatically define nail printing area becomes necessary. In this study, it proposes a computer vision method to mark the area of nails, which generate the nail image and merges the user selected patterns. The segmented-and merged result is then sent to the nail painting machine for printing.

The Brazilian digital television system is migrating its middleware approach from profile Ginga-C to an integrated broadcast-broadband strategy, known as profile Ginga-D. The present work proposes a methodology able to extending current implementations, in a systematic way, with the goal of harmonization maintenance and conformance regarding current standards.

Each year, different kinds of trauma generate about five million deaths worldwide and an increase in people with disabilities in their most productive years. These problems represent a social and economic impact on families. These accidents occur due to lack of information on prevention. This document details a robotic assistant to support the teaching about trauma accidents to children (3 - 6 years old). The results show that the information provided by the robotic assistant during the teaching process is retained by the children. It provides the child with an alternative way of learning about prevention.

The main proposal of this paper is to design hardware sharing architecture for DCT and IDCT between in MPEG-2 and HEVC.Using the common factor of MPEG-2 and HEVC 8×8 DCT coefficients,a mapping technique suitable for hardware design is proposed.Based on hardware design aware,transform coefficient of MPEG-2 is developed by a basis of 2 and added by the compensation coefficient matrix.The transform coefficient of MPEG-2 and HEVC are verified and can be sharing.We propose a hardware architecture design in FPGA for resource sharing.Its architecture 2D-DCT takes 2424 slice LUTs and the maximum achievable clock frequency for the proposal is 209.732 MHz

This paper proposes a system which implements the Cincinnati Prehospital Stroke Scale (CPSS), the widely used screening method for the initial symptoms of a stroke, in a communication robot. AI on cloud analyses an acquired video through a conversation with the robot in real time and automatically determines the abnormalities. The judgement result is informed to his/her families by SNS. This study implemented two of the three CPSS scales such as "Arms" and "Speech", we confirmed that the system enables to acquire, analyze and notify the information in real time.

Voice is very important in our daily life. However, some people cannot speak because of throat surgery. Electrolarynx (EL) is a way for them to speak again but this voice is flat and noisy like a buzzer. Therefore, we propose a new Electrolarynx system based on WaveRNN. When we use it, the EL voice is converted into a voice close to the natural voice. We assume that this system will be provided as a smartphone application. In the experiment, we evaluated the converted voice by the system and found that naturalness improved.

With FDA-approved portable x-ray handheld units available, there is a need of on-device classifier. In this work, a small-scale and cost-effective model is implemented on embedded system to demonstrate the successive subspace learning. The embedded system is tested on well-known chest x-ray datasets from Open-I and NIH ChestX-ray14. This paper has two main contributions: (1) A subspace weakly supervised learning x-ray image classification model is developed with better than 90% accuracy. (2) The network is implemented on a small and an affordable single-board computer without any accelerators to form a low-power on-device AI x-ray image classifier.

Discomfort detection for young infants is essential, since they lack the ability to verbalize their pain and discomfort. In this paper, we propose a novel infant monitoring system, enabling continuous monitoring for infant discomfort detection. The proposed algorithm is robust to arbitrary head rotations, occlusions and face profiles. For this purpose, a Faster R-CNN architecture is first pre-trained with the ImageNet dataset, and then fine-tuned with a training dataset of different infant expressions. The presented system enables reflux disease analysis and remote home monitoring in a more relaxed environment, which is largely preferred by pediatricians and parents.

With FDA-approved ECG sensor and cloud-based interpretation, researchers are focused on end-to-end training on device. In this work, we present a semi-supervised learning classifier that jointly learns the parameters of a convolutional neural network (CNN) and using support vector machine (SVM) on the resulting features. By applying Saak(Subspace approximation with augmented kernels) Transform, we can update the weights of the network in a single multilayer feed forward step without backpropagation. We trained this CNN with the MIT-BIH arrhythmia database, on Raspberry Pi ARM platform. The resulting ECG beat classifier has 98.94% accuracy and 1.06% misclassification rate.

This paper presents a low-light image enhancement method using a convolutional neural network (CNN). Given a low-light input image, the proposed method converts RGB color space to CIELAB color space. The luminance and chrominance components are separately enhanced. The luminance channel is enhanced using a CNN to enhance the brightness. On the other hand, the chrominance channels are enhanced using a dilated CNN to reduce the color distortion. Experimental results demonstrate that the proposed method can successfully enhance low-light images of a vehicular imaging system without color distortion.

In order to enable convolution neural network (CNN) to be deployed on a Field Programmable Gate Array (FPGA), this study builds a lightweight convolutional neural network that can be separated by a depth to reduce the amount of parameters and computations stored. We replaced the standard convolution operation with a separate convolution operation, and proposed a hardware accelerator architecture that can handle differently sized depth-separable convolution operations, using parallelization to efficiently utilize hardware resources for depth separable convolution.

In this paper, we propose novel networks for stereo disparity estimation. First, deep features are extracted using efficient depthwise-separable convolutions. Next, the stereo matching costs are calculated from the deep features with a novel extended cost volume. Then, rich multiscale contextual information are aggregated with the proposed atrous multiscale network (AMNet). The proposed foreground-background aware network (FBA-AMNET) is trained with an iterative multi-task learning strategy to discriminate between foreground and background objects at multiple scales. The proposed networks advance the state of the art on challenging disparity estimation benchmarks, such as the KITTI 2012, KITTI 2015, and Sceneflow stereo benchmarks.

Using bio-signals, one can calculate motion, can authenticate individuals using fingerprints. These bio-signals are convenient because they can be measured easily, anytime, anywhere. However, there are cases of misuse of bio-signal based technology. To solve this problem, we develop an electromyogram (EMG)-based algorithm that can be used for biometric authentication. The proposed algorithm uses two channels to acquire EMG data, and hand motion recognition and authentication are carried out through signal processing. As a result of dividing 50 data sets into artificial neural networks and applying the ensemble technique, the authentication success rate achieved by the proposed algorithm was 83.8%.

Location-based Notification System (LNS) is a system that informs the user of events occurring in the street. This research proposes a LNS that predicts a user's path and informs the user only of events around the predicted path, and also develops mobile applications and simulates it with street events.

In this paper, we present a stochastic computing based AI system for mobile devices. As technology in AI advances, more complex computations are required. In case of mobile devices, it is hard to accommodate the entire computations due to power and area limitation of an embedded system. As a stochastic computing replaces the complex computations into simple computations, the mobile devices are available to include the AI system. In order to verify our design, the embedded AI system including stochastic computing is implemented on an field-programmable gate array (FPGA), and we successfully demonstrated the feasibility of the proposal.

Recently, flash based SSDs employ dozens of flash memory chips to achieve high performance and large capacity. However, employing plenty of flash chips leads to increased failure rate in SSDs. A RAID-like protection is one of the typical techniques to recover from such failures. However, with the advent of massive storage devices, various studies on the design of RAID-like protection techniques are needed.

To this end, in this work we propose a massive SSD-Aware Parity Logging (mSAPL) scheme which protects against n-failures at the same time in a stripe where n is protection strength from user selection.

Foreseeing the positions of the surrounding objects is a critical technique to autonomous vehicles. Referring to the positions predicted by the trajectories, moving vehicles could avoid possible collisions. Many trajectory prediction methods about have been proposed, from traditional handcrafted approaches to Recurrent Neural Network (RNN) models. But there are still exist some problems, such as predicting positions in obstacles, sharp turns that are not socially acceptable. We adopted the concepts of social GAN to propose a new model that take the social forces, scene information and motion direction into account when predicting trajectories.

In this paper, we present the concept of multilayer cross-platform graphics sharing in the automotive digital cockpit. Considering that automobiles today have around 150 ECUs (engine control units), managing all these ECUs is becoming a challenging task. For example, there is a controller (System on Chip - SoC) for every display in an automobile. This SoC is used for content rendering and data processing. The number of ECUs can be lowered by using SoCs with a hypervisor. A hypervisor is a concept that enables us to run two operating systems on one SoC in real-time. The content from both operating systems can be rendered and presented in the same display output. The proposed system consists of one SoC with two operating systems running on a hypervisor. With this proposed solution, we were able to simultaneously render content from both operating systems on one display output. The proposed solution also covers the rendering of media content on display that is hosted on a different operating system and therefore enables mixed criticality where safety-critical information, such as those presented in the cluster, are presented with no interference with the non-critical operations, such as media rendering. We also evaluate safety concerns and system performance when content is rendered simultaneously on both operating systems.

The constant expansion of automotive industry has led to a lack of technology and standards in the field that are needed to keep up the pace with automotive requirements. In order to support these new requirements, it is necessary to find a way to integrate new functionalities into existing systems without disruption of the system. The leading automotive industry standard, AUTOSAR, has been expanded with an Adaptive Platform (AA) that offers support for realizing new automotive features. In this paper scalable extension of the software within existing architecture with modules that provide communication with security-critical parts of the system is implemented within AA environment. The scalability in solution provides the flexibility in the development of the future systems. The AUTOSAR foundation ensures preservation of the automotive grade quality within the components.

An efficient moving object detection (MOD) method for monocular camera with ego-motion is proposed in this paper. In general, There are keypoint-based foreground detection methods that find out outliers using epipolar geometry of two camera coordinate systems in consecutive frames. Determining of regions of moving objects from sparse keypoints is especially challenging. In this paper, we propose an optical flow noise removal and an efficient sparse-to-dense motion probability map. Subtraction method using perspective transform is applied to detect candidate regions of objects. Moving objects are detected by combining subtraction method with the proposed motion probability map.

New trends in the automotive industry, such as e-mobility and autonomous driving, give birth to challenges of engaging passengers in a vehicle. Once relieved from driving tasks, passengers may take up productive and entertainment-related activities, such as media consumption. It is assumed that by 2030 in-car cabin would mostly resemble a living room, whereas internet connectivity would become one of the most important enablers for new content delivery. Streaming media in such an environment poses a challenge, given that reliability and adequate coverage in 4G and coming 5G networks are mandatory. In this paper, we give a brief analysis of challenges which streaming media would bring to the user experience. We give a proposal of a software stack for a trajectory-aware playback, capable of predicting streaming quality drops and proposing appropriate fallbacks.

In this study, we propose a stress classification algorithm to determine whether the user is exposed to stress, using Empirical Mode Decomposition and Fast Fourier Transform on multiple bio-signals. Electrocardiogram (ECG) and Photoplethysmogram (PPG) signals were detected from the user's arm for accurate classification of stress. By using the EMD and FFT techniques, the bio-signals in the frequency band were analyzed in various stress levels. As a result of analyzing the stress, it was confirmed that R interval value and mean frequency value when stressed situation were lower values.

Gait monitoring is considered as a significant marker of disability, injury, and gait symmetry. The goal of this study is to develop a real-time consumer health monitoring system based on IoT sensors and Machine learning technique in order to detect health abnormalities such as, stroke onset. The proposed consumer stroke prediction system consists of IoT based gait monitoring sensors, real-time vital sign monitoring and machine learning based disease prediction model to predict the disordered gait and the healthy gait. This study will be useful for post-stroke gait coordination for rehabilitation and consumer health monitoring service.

This paper overviews a new consumer product that allows a consumer to detect gastrointestinal disorders, such as colon cancer and others, using new technology and artificial intelligence techniques. We review the concept, compare to the existing market technologies, and show the significant opportunities for consumer products in this "advise the doctor" space of products.

This paper describes our developed biofeedback device based on muscle fatigue scale (MFS). At present, Japan faces the issues of population decrease and a super-aging society. Extension of healthy life expectancy and reduction of health disparities are especially the ultimate objectives in Japan. Seniors who want to remain healthy and active must prioritize doing regular physical activities. Under the circumstances, we develop a biofeedback device based on MFS. By using the device, seniors, as well as any other persons who want to do muscle training, can do it suitable for the condition of a muscle.

High Efficiency Video Coding (HEVC) introduced many new coding tools to gain improved coding efficiency. However, it greatly introduces computational cost, and this will give rise to the demand for the fast algorithm that can satisfy the real-time encoding of HEVC encoder for low bandwidth environment. In this paper we introduce and evaluate a novel machine learning based approach that uses multi-view clustering to reduce the complexity of intra-coding algorithm based on the analysis of original texture and its neighborhood.

The lightweight U-Net based monaural speech source separation method is proposed. The proposed method utilizes U-shaped neural networks to segregate speech and interfering noises from noisy audio recording. To reduce size of the networks, the proposed method employs the multi-lane dimensionality reduction module to each convolutional layer. Compared with the conventional U-Net based method, the proposed method achieved almost on-par performance to those of the conventional one while using model footprint of 1.39 MByte, which is only 3.72% of size of the conventional U-Net. Moreover, the proposed method is implemented in off-the-shelf edge computing.

This paper presents the application of the backward feature elimination technique on an electronic nose (E-nose) to aid the rapid detection of pathogens using Volatile Organic Compounds (VOCs). The timely identification of pathogens is vital to facilitate control of diseases. E-noses are widely used for the identification of VOCs as a non-invasive tool. However, the identification of VOC signatures associated with microbial pathogens using E-nose is currently inefficient for the timely identification of pathogens. Therefore,we proposed an E-nose system integrating the backward feature elimination. Comprehensive experiments of backward feature elimination showed that they improve the classification accuracy.

Fault diagnosis is a well-known practice in the industrial, automotive fields, etc. but less used for consumer electronics. This paper analyzes a Cloud service based on a Machine Learning (ML) approach used to provide fault detection capabilities to household appliances equipped with electric motors and compare the results with on premise ML algorithms provided research tools. The purpose of this paper is to perform a preliminary comparison of ML algorithm performances provided by two software, namely Microsoft Azure (cloud solution) and MATLAB (on premise solution), on a study case.