P‐2.31: An Emergency Rescue Action Recognition Method Based on Improved Spatiotemporal Decomposition Network
Zhang Yong-mei, Zhao Tianxiang · SID Symposium Digest of Technical Papers · 2023
In order to solve the problem of too large computer overhead for convolutional neural network in the rescue action recognition, this paper proposes a rescue action recognition method combining spatiotemporal decomposition network and channel attention mechanism. After adding the channel attention mechanism, the model can generate a weight value for each feature channel, and then weight the normalized weight to each feature channel to improve the recognition accuracy of the model. The combined network model has better performance in both computer overhead and recognition accuracy, and the recognition accuracy has improved compared with the original model (S3D) in the case of only RGB video as input on rescue action data. In addition, the recognition accuracy on the public UCF-101 and KTH datasets has also improved. The experiment results show the proposed method can effectively improve the accuracy of action recognition in both the rescue action and public datasets.