Video action recognition method based on attention residual network and LSTM
Yu Zhang, Pengyue Dong · 2021
A video action recognition method based on attention residual network and long-term memory network(LSTM) is proposed, which is to solve the problems that the existing human action recognition methods are prone to overfitting, susceptible to interference information, and lack of feature expression ability. In the beginning, the traditional data preprocessing method and sampling method are improved to enhance the generalization ability of the model. Then, a residual network with attention is proposed to improve the feature extraction ability of the network. At length, LSTM is used to recognize video actions. Experimental results on UCF YouTube dataset show that the proposed method can recognize the actions in video more effectively than other similar methods in this field, and the recognition rate reaches 95.45%.