| 研究課題/領域番号 |
18K11408
|
|---|
| 研究機関 | 法政大学 |
|---|
研究代表者 |
Jianhua Ma 法政大学, 情報科学部, 教授 (70295426)
|
|---|
| 研究分担者 |
Huang Runhe 法政大学, 情報科学部, 教授 (00254102)
|
|---|
| 研究期間 (年度) |
2018-04-01 – 2021-03-31
|
| キーワード | Wearable / Recognition / Modeling / Activity / Emotion / Personality |
|---|
| 研究実績の概要 |
Based on the previous study of the multi-wearable platform and related key technologies, our research in 2019 fiscal year was focused on various kinds of wearable applications using multiple wearable devices. These applications are fallen into two categories, human activity recognition (HAR) and mental characteristic recognition (MCR). The HAR covers simple activities such as standing and walking, and complex such as activities eating and deskwork, which are involved in a series of simple activities. The MCR covers human mental states such as stress or relax, and personal character such as mood and personality. One of core research issues is how to effectively use the data sensed from multiple wearables to improve the recognition accuracy in HAR and MCR. The data from wearables is divided basically into three types, i.e., inertial data, contextual data and physiological data. We have made a series of research output in various HCR and MCR using combined multi wearable devices, which have been published in three journals and presented in both international and domestic conferences.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
理由 The research was carried out basically as we planned. Because our laboratory had previously done much research in HAR and MCR using a single wearable device, our main research work was to integrate the datasets simultaneously from multiple wearables and find out the suitable machine leaning algorithms for recognition of human activity and characteristics. For HAR, we studied two new approaches; the one is to recognize complex activities in a hierarchical way, and the other is to fusion activity using both inertial data and contextual data. The experimental datasets were taken from seven devices that were worn in the different locations in a participant’s body. The experiment results have shown that the both processing time and recognition accuracy are improved using the two approaches. For MCR, we used the physiological data, i.e., heart rate, aspiration rate and eyeblink from devices Emotiv and Spire to recognize personal trait affective intensity and emotional stability. The result has shown that the recognition accuracy increases when data from multiple sources are used. This result has been further proved in personal character modeling. We also conducted the research to recognize human states of stress and concentration, respectively.
|
| 今後の研究の推進方策 |
The research in FY 2020 will be carried out from the following three aspects. First, we plan to extend from the human activity recognition to human attribute recognition. The attribute means human gender, height, weight, etc. An associated key issue is to fusion a personal attribute more precisely using data from multiple wearables. Second, we shall study recognition of other mental characteristics including preference, short-term stress, and focus degree. Besides the heart and aspiration rates, brainwave and EDA data will be used for possible improvement of recognition accuracy. Deep learning algorithms will be also adopted in the recognition. Third, we are going to study various fusion techniques for human character modeling. Scenario-based feature fusion, decision fusion and incremental fusion will be implemented to figure out the best fusion approach and algorithm in character recognition. In addition, we are also going to study recognizing activity of animals such as dog, and wearable-based interaction between human and computer.
|
| 次年度使用額が生じた理由 |
One main reason is from the unexpected coronavirus, due to which a planned oversea travel has been canceled and an invited visitor cannot come. In addition, the device with expected specification is not available to purchase.
From the late half of 2020, the situation caused by the coronavirus will be possibly recovered. And we also expect to have more research output. So, we shall have more chances to present work in both domestic and international conferences and be able to exchange research physically with external researchers.
|
