Human behavior recognition method based on double-branch deep convolution neural network

Aiming at the poor robustness and the low accuracy of 2D image recognition, this paper presents a method of human behaviors recognition based on double-branch deep convolution neural network. Firstly, the features of the input image are extracted, and the feature maps are input into the double-branch deep convolution neural network to obtain the joints information of human body and the joints connection information of human body respectively. And we used the Hopcroft-Karp algorithm to optimally match skeletal joints to obtain the human skeletal sequence diagram. The human behaviors are identified by the multi-classification support vector machines. Finally, through the training and testing of standing, walking, running, waving, bending, squatting and sitting on seven kinds of general human behaviors, the experimental results show that the proposed approach has good accuracy and robustness.