Research On Human Behavior Recognition Technology Based On Deep Learning

Human behavior recognition technology is an important research direction in the field of computer vision,and it is the key technology in the fields of video monitoring,human-computer interaction,video retrieval,virtual reality and so on.The application of deep learning technology in the field of computer vision promotes the development of human behavior recognition technology.However,due to the high complexity and variability of human behavior,as well as the interference of complex background on human behavior recognition,the application of human behavior recognition technology in reality faces many challenges.This paper focuses on the recognition of human behavior in video,focusing on the key technologies of human behavior recognition.The specific work is summarized as follows:(1)In the aspect of human behavior feature extraction,based on DenseNet,a DenseNet3D convolutional neural network is proposed for human behavior recognition.In this network,3D convolution layer is used to extract features,and spatiotemporal information is introduced into the neural network to improve the network performance.When building the network structure,when the front and back layers of the network are directly connected,cross layer connection is realized between different network layers,so that the features extracted by the bottom layer network can be directly mapped to the top layer,so as to enhance the feature propagation and reduce the network parameters.Aiming at the high complexity and variability of human behavior,the network can make full use of space-time information and improve the utilization rate of features and recognition accuracy.In this paper,the method is validated in UCF101 behavior database.The experimental results show that the network can effectively improve the accuracy of human behavior recognition.(2)On the basis of DenseNet3D convolutional neural network,this paper proposes a human behavior recognition method based on two-stream DenseNet3D network.This method uses DenseNet3D convolutional neural network to extract the features of optical flow samples and RGB video image samples respectively.TV-L1 optical flow algorithm is used to generate the optical flow samples,and the algorithm uses two-way solution mechanism to reduce the amount of operation of extracting optical flow.In the classification and recognition stage,this method uses the scoring feature fusion mechanism of two-stream network to integrate the video image and optical flow through dens The RGB score feature obtained from the DenseNet3D network is fused with the optical flow score feature to generate the fusion feature and identify it.The fusion feature contains both video image information and optical flow information.The introduction of optical flow into convolutional neural network can reduce the interference of complex background on human behavior recognition in video image,and further improve the accuracy of human behavior recognition.