| 吉田 誠 | D, 中間発表 | ユビキタスコンピューティングシステム | 安本 慶一, 岡田 実, 諏訪 博彦, 藤本 まなと |
|
title: Detecting the state of people and objects using vibration sensors
abstract: We propose sensor hardware and deep learning/machine learning algorithms for accurately estimating the state of people and objects from vibrations generated by the movement of objects and people.Recognizing the state of people and objects using vibration has the advantage that there is little privacy-related information from the data acquisition stage, and even if the data is leaked, it is expected to reduce the damage.In addition, the data volume is smaller than that of camera images, making it superior in terms of computational cost. However, vibration signals have a large dynamic range, from minute signals to shock signals, making it difficult to increase the amplification in the signal amplification circuit of a sensor, making it difficult to achieve both high sensitivity and low noise.In addition, vibration reaches the sensor through a complex system with different transmission speeds and directions, making it difficult to detect the direction of movement, which is necessary for state recognition.In this paper, we present a unique vibration sensor using a piezoelectric element, an amplifier circuit that can handle a wide dynamic range, and a software algorithm that uses machine learning and deep learning to learn the vibration signal transmitted through complex transmission paths. language of the presentation: Japanese 発表題目: 振動センサを用いた人・物の状態検知 発表概要: 深層学習・機械学習を用いて物や人の移動に伴って発生する振動から、人や物の状態を高精度で推定するためのセンサハードウェアとアルゴリズムを提案する．振動を用いた人や物の状態認識はデータの取得段階からプライバシーに関わる情報が少ないという利点があり、データが万一流出したとしても被害の低減が期待できる．また、カメラ画像と比較してデータ容量も少なく、計算コストの面で優れている．しかし、その一方で振動信号は、微小な信号から衝撃的な信号まで、得られる信号のダイナミックレンジが大きいため、センサの信号増幅回路において増幅度が上げにくく、高感度化と低ノイズ化の両立が難しかった． また、振動は伝達速度も方向も異なる複雑な系を伝達してセンサに到達するため、状態認識に必要な移動方向などの検出が困難という課題があった．本発表ではピエゾ素子を用いた独自の振動センサと広いダイナミックレンジに対応できるアンプ回路および複雑な伝達経路を伝わってくる振動信号を学習させる深層学習及び機械学習を用いたソフトウェアアルゴリズムを実装することで、高精度で物や人の状態を検出を実現した． | |||
| 平野 陽大 | D, 中間発表 | ユビキタスコンピューティングシステム | 安本 慶一, 中村 哲, 諏訪 博彦 |
|
title: Extension of the problem and construction of algorithms in automatic tour planning systems
abstract: With the expected expansion of tourism demand, an automatic tour planning system (ATPS) has been proposed and put into service, which automatically selects sightseeing spots in a specific area and recommends a tour plan. However, in general, ATPS only suggests the shortest route or the cheapest travel price, and does not take into account the important indicators for users, such as satisfaction with sightseeing, money, time, and physical strength. In this study, we defined MOATPS, a multi-purpose extension of the existing ATPS, and constructed algorithms for each of A Posteriori approach and A Priori approach as a way to solve MOATPS. We also evaluated the effectiveness of the algorithms by comparing the algorithms of the two approaches. language of the presentation: Japanes | |||
| 内田 朋希 | D, 中間発表 | 知能コミュニケーション | 中村 哲, 安本 慶一, 田中 宏季 |
|
title: Simple analysis tools for health information using questionnaires: Assessing glucose metabolism status and water intake
abstract: 463 million people worldwide had diabetes as of 2019. Fortunately, Lifestyle modifications can reduce the risk of developing diabetes. Therefore, it is important for individuals without diabetes to understand their glucose metabolism status and to take appropriate measures for preventing diabetes. Screening tools for diabetes and pre-diabetes have been developed. However, they require the linkage of laboratory data and the input of many factors. This may increase the difficulty of implementation and limit the widespread use of the tools. Moreover, although these tools use known diabetes risk factors, there may be other unknown factors associated with glucose metabolism status. In addition, there are no tools to determine glucose metabolism status in non-diabetic individuals. In this study, we developed a machine learning model to identify glucose metabolism status in non-diabetic Japanese adults. The explanation variables of the model were only questions about lifestyle and physical information that can be answered on the spot. They included both known and unknown diabetes risk factors. language of the presentation: Japanese | |||
| WEI XIN | M, 1回目発表 | サイバネティクス・リアリティ工学 | 清川 清, 佐藤 嘉伸, 内山 英昭, 磯山 直也 |
|
title: Unobtrusive Self-Monitoring of Eye Conditions Using Eyewear
abstract: We propose an unobtrusive, standalone, and wearable solution for monitoring eye conditions. For some eye diseases, the patients can avoid the consequences upon timely diagnosis and treatment. While professional eye examinations lead to a reliable diagnosis, our goal is to detect the eye disease symptoms that are often overlooked in the early stages. Our solution is based on a smart eyewear and combines photo reflective sensors, electrooculography (EOG) electrodes to collect the information related to eye movements. The key idea is to analyze the skin deformations around eyes and the changes in electric potentials via build-in sensors and eyewear with computing power. The whole system with no interaction and no additional components enables long-term eye monitoring. language of the presentation: English | |||