Joint Motion Information Extraction and Human Behavior Recognition in Video Based on Deep Learning

Video surveillance has played a huge role in the safety and security tasks of various places for many years, and its application range is also expanding. The core problem that needs to be addressed in intelligent video surveillance is the identification and analysis of human behavior in video. In this paper, the structure and joint motion feature extraction of the two-channel deep convolutional neural network model are studied and designed. In the design of the network structure, the ventral and dorsal channels used in the processing of visual signals by the brain visual cortex were simulated. The spatial channel network and the time channel network are used to process static information and dynamic information respectively, and the two types of features are separately extracted. The superiority of the dual-channel structure is verified by comparing the recognition effects of the two-channel model with the single-channel model. Finally, experiments were performed on the KTH behavioral dataset. The results show that the human behavior recognition algorithm based on deep learning can achieve high recognition accuracy based on the good extraction of joint motion information.