Violence detection based on attention mechanism
Chenglong Pan, Shumin Fei · 2022 41st Chinese Control Conference (CCC) · 2022
Violence detection task is very useful in video surveillance and other scenes, such as prisons, schools and other public places. In order to fully extract the temporal and spatial features and temporal features in video surveillance, and accurately detect and classify violence, a violence detection model combining the advantages of two-stream network, 3D convolutional neural network and convolutional LSTM network network is proposed. In order to better extract human motion features, the model uses YOLO target detection method to segment human targets and extract local features, integrates CA attention mechanism into 3D convolution network, so as to improve the feature extraction of video channel and space, and replaces part of 3D convolution layer in neural network with convlstm layer, so as to obtain better video timing relationship. The experiment was carried out on RWF-2000 data set. The results show that the proposed model has high accuracy in violence detection.