Name
Abstract | ||
|---|---|---|
Sensor-based activity recognition can recognize simple activities such as walking and running with high accuracy, but it is difficult to recognize complex activities such as nursing care activities and cooking activities. One solution is to use multiple sensors, which is unrealistic in real life. Recently, learning using privileged information (LUPI) has been proposed, which enables training using additional information only in the training phase. In this paper, we used LUPI for improving the accuracy of complex activity recognition. In short, training is performed with multiple sensors during the training phase, and a single sensor is used during testing. We used four published datasets for evaluating our proposed method. As a result, our proposed method improves by up to 16% in F1-Score to 67% compared to the baseline method when we used random-split cross-validation of each subject. |
Year | DOI | Venue |
|---|---|---|
2021 | 10.1007/978-981-19-0361-8_3 | Sensor- and Video-Based Activity and Behavior Computing |
DocType | Volume | ISSN |
Conference | 291 | 2190-3018 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
4 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Adachi Kohei | 1 | 0 | 0.34 |
| Paula Lago | 2 | 3 | 4.76 |
| Yuichi Hattori | 3 | 36 | 7.57 |
| Sozo Inoue | 4 | 176 | 58.17 |