Title | ||
---|---|---|
Skeleton-Based Explainable Human Activity Recognition For Child Gross-Motor Assessment |
Abstract | ||
---|---|---|
Human activity recognition (HAR) is a basic technology for understanding human behavior and motion, and is used in various fields and products related in the industrial informatics. This paper focuses on child gross motor (GM) skills as the target of explainable HAR and proposes a visualization method of recognition reason. Subsequently, a motional time series image conversion method and its data augmentation are proposed, and the recognition accuracy to thirteen type of GMs was extremely improved to 99.5% using presented deep network. Finally, a skeleton-based method for visualizing the reason of discrimination using Grad-CAM was proposed, and its usefulness was investigated by an agreement analysis. As a result, the similarities / differences between humans and machines were found, and a network structure of explainable HAR suitable for obtaining a high consent by human was confirmed. |
Year | DOI | Venue |
---|---|---|
2020 | 10.1109/IECON43393.2020.9254361 | IECON 2020: THE 46TH ANNUAL CONFERENCE OF THE IEEE INDUSTRIAL ELECTRONICS SOCIETY |
Keywords | DocType | ISSN |
activity recognition, CNN, gross motor, explainable AI | Conference | 1553-572X |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Satoshi Suzuki | 1 | 99 | 15.23 |
Yukie Amemiya | 2 | 0 | 0.34 |
Maiko Sato | 3 | 0 | 0.34 |