| 大西 晃正 | M, 2回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 清川 清, 荒川 豊, 諏訪 博彦 |
|
title: Stress measurement and analysis while performing daily living activitie
abstract: In recent years, there has been a great deal of research to improve the quality of life (QoL: Quality of Life). Although social efforts have been made, it is necessary to make efforts to improve QoL by one's self in the private space of a residence due to privacy and other factors. In this study, we aim to support household chores, which are indispensable in living in homes.For this purpose, it is necessary to understand the stress conditions of residents in the performance of household chores.Therefore, we conducted an experiment to grasp the stress condition of residents by sensing the subject doing household chores. The results showed that the accumulation of stress in household chores varied among individuals, and the degree of stress in each household chores varied. | |||
| 鳥越 庸平 | M, 2回目発表 | ユビキタスコンピューティングシステム | 安本 慶一, 清川 清, 荒川 豊, 藤本 まなと |
|
title: Strike Activity Detection and Recognition Using IMU Towards Kendo Skill Improvement System
abstract: In the field of sports, there are increasing opportunities to utilize inertial sensor units (IMU) to analyze player movements in games and practice, and improve their performance. In this study, I focus on Kendo, which is a traditional sport in Japan, and propose a method for detecting and recognizing strike activity using IMU towards realize a Kendo skill improvement support system. First, I collected sensor data of strike activities of 14 subjects who attached 4 IMUs to right wrist and waist, and to the Shinai (bamboo sword) Tsuba and Saki-Gawa. Next, I conducted an evaluation experiment to verify the effectiveness of the proposed method using the obtained sensor data. As the strike activity detection, I used the Dynamic Time Warping (DTW) distance between the training data and the sensor data in order to detect the strike activity. As the strike activity recognition, I classified 6 types of strike activities (Center-Men, Right-Men, Left-Men, Dō, Kote, Other) using a machine learning method. As the result, I confirmed that proposed method can detect and recognize strike activity with high accuracy. language of the presentation: Japanese 発表題目: 剣道上達支援のためのIMUを用いた打突動作検出および認識手法の提案 発表概要: スポーツの分野において、慣性センサユニット（IMU）を用いて試合や練習における競技者の動きを分析し、パフォーマンスを向上させる機会が増加している。本研究では、日本の伝統的競技である剣道に焦点を当て、剣道上達支援システムの実現に向けて、IMUを用いた打突動作検出および認識手法を提案する。最初に、4つのIMUを右手首と腰、竹刀鍔、竹刀先革に取り付けた14人の被験者の打突動作のセンサデータを収集した。次に、得られたセンサーデータを用いて、提案手法の有効性を検証するために評価実験を行った。打突動作検出では、打突動作を検出するために、教師データとセンサーデータ間のDynamic Time Warping（DTW）距離を用いた。打突動作認識では、機械学習手法を用いて、6種類の打突動作（正面、右面、左面、胴、小手、その他）を分類した。その結果、提案手法を用いることで打突動作を高精度に検出および認識できることを確認した。 | |||
| YANG FAN | D, 中間発表 | 知能コミュニケーション | 中村 哲, 清川 清, Sakriani Sakti, 伍 洋 |
|
Title: Domain Adaptation with Imperfect Source Data
Abstract: Deep-learning-based models are hungry for well-annotated data, however, it is costly to annotate new data for a specific target task. Since related data with annotations (i.e., source data) might be available, it is highly desirable to utilize them to reduce the target data annotation. Transfer Learning (TL) generally services for such purpose and reaches great success in a wide range of applications. Among TL, Domain Adaptation (DA) further reduces the domain divergence between source and target data to address more challenging TL tasks. Nonetheless, most of the existing DA studies focus on ideal scenarios (e.g., homogeneous to target data), while we suppose that the source data could be imperfect (e.g., heterogeneous to target data) since the source data are not designed for target tasks in real applications. Under such background, we raise an exploration to DA with imperfect source data. language of the presentation: English | |||