We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
SMA-GCN: a fall detection method based on spatio-temporal relationship.
- Authors
Yang, Xuecun; Zhang, Shanghui; Ji, Wei; Song, Yijing; He, lintao; Xue, Hang
- Abstract
With the aging population in our society, falls have become a major cause of injury to the elderly in their daily lives. Based on this, fall detection for the elderly has received wide attention. However, under complex background and illumination changes, skeleton map feature extraction has stronger robustness than direct image feature extraction, but the existing methods for skeleton map data detection have the problem of insufficient mining of spatiotemporal features. Therefore, this paper proposes a fall detection method based on spatio-temporal features to accurately detect falling behavior using the spatio-temporal dependency of the skeleton. In terms of network design, first, we design a new spatial graph convolution method to extract spatial features of the skeleton, which focuses the network on potential features between joint points that are not physically connected in human skeleton information. Obtaining potential features between joint points that are spatially non-physically connected. Second, the fusion of temporal features of different granularities through multi-scale temporal graph convolution improves the network is ability to extract temporal information; finally, considering the spatio-temporal correlation of actions, we introduce a spatio-temporal attention module to extract deeper spatio-temporal features. The experimental results show that the algorithm in this paper achieves 98.6% precision and 98.86% recall on the fall dataset and can effectively detect the occurrence of fall behavior.
- Publication
Multimedia Systems, 2024, Vol 30, Issue 2, p1
- ISSN
0942-4962
- Publication type
Article
- DOI
10.1007/s00530-024-01293-0