| 研究実績の概要 |
Our research in 2020 fiscal year was focused mainly on human attribute recognition and emotion characteristic analysis using multiple wearable devices. The six wearable devices were placed in different body positions, namely chest, waist, left wrist, right wrist, thigh, and ankle, for each of twenty-nine subjects during their walking, and going up-stair and down-stare. We first perform classifications of gender, height, and weight based on each of the six devices using machine learning algorithms including SVM, LR, k-NN and RF, then perform decision fusion based on classified results from these devices in different activities. The data of electroencephalography (EEG), heart rate variation (HRV) and galvanic skin reaction (GSR) were used to analyze and estimate human emotion states. We found that the high accuracy can be achieved in recognizing three concentration states in PC office work using EEG and HRV. We applied machine learning algorithms of SVM, RF, and k-NN to determine stress states, and achieved at 95.7% in two-state stress estimation and 81.2% in four-state stress estimation when using RF. Multi sensors were also used for human gesture recognition and dog activity recognition. In addition, we have proposed a scenario-based fusion framework to for personality computing using various kinds of data.
|
| 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
The research in 2020 fiscal year was carried out basically as we planned. Using multi wearables is one of the main features in this project. The relevant basic research questions when using multi are (1) what positions in wearing devices will be better in recognition; (2) what devices can be combined to achieve high recognition accuracy; (3) how to effectively fuse the data from multi wearable devices. Therefore, multi devices were worn simultaneously by a subject in all studies in 2020FY. The six devices placed different positions were used to get inertial data in recognizing human attributes including gender, height and weight. Our research has shown that devices placed on the body’s upper part enable higher accuracy as compared with placing devices on the body’s lower part in attribute recognition. Furthermore, the combination of two devices placed in the upper part can further improve recognition accuracy. This result is also confirmed in our studies on stress estimation using both HRV and GSR data, on continuous gesture recognition using two wearable devices called LPMS-B2, and on dog activity recognition using two wearable devices on dog collar and back. The approaches of weighted-sum and majority-voting are used to fuse classification results from different devices and activities, which have obtained higher recognition accuracy.
|
| 今後の研究の推進方策 |
The research in 2021 will be carried out from the following aspects. First, we plan to focus on applying wearables for recognition of activities of daily life, such as food intaking, tooth brushing, and hand washing, which is very important in the current COVID-19 pandemic. In this study, the wrist-worn inertial sensor, pressure sensor and EMG sensor are to be used. We also plan to study wearable-based gesture recognition when a use is moving, i.e., walking. Second, we shall extend the study of emotion recognition to the special mental behavior such as fear as well as the physical and psychological combined fatigue using physiological sensors including EEG, ECG and GSR. We will also explore recognition of dog emotion using both inertial and HRV data. Third, the physiological data such as heartbeat and breath will be used for study on human biometrics and authentication. We shall conduct various experiments to see how accuracy can be achieved to recognize human with the physiological data.
|
