Research On Video Action Recognition Based On Deep Learning

With the development of Internet technology and multimedia technology,the application of video is becoming more and more popular.Meanwhile,videos without supervision will also have a bad effect on the public,especially teenagers.Therefore,the screening of video content is of great social significance and research value.In recent years,deep learning has become the main means to deal with video semantic analysis because of the breakthroughs in deep neural network.On the other hand,due to the development of the new generation of high efficiency video coding technology HEVC,video compression can maintain more video information while achieving higher compression.In this paper,we propose a video action recognition method using HEVC compressed domain information.This paper first reviews the recent research on video semantic analysis,and then introduces the key technologies of deep neural network.The main work of this paper is as follows.In the non-compressed domain of video,several common methods of recognition are studied.It includes two-dimensional convolution network based on single frame image,three-dimensional convolution network based on multi frame fusion,recurrent neural network and two-stream neural network based on optical flow information.The theoretical research and simulation results of the models show that the two-stream neural network obtained the best accuracy by fusing spatio-temporal information.In the compressed domain of video,we extract the motion vectors inter frames to represent the motion information of the video.After that,we firstly filter the noise in the motion vectors,and then use RANSAC algorithm to compensate the global motion.We apply the two-stream neural network to the processed motion vectors,which achieves the expected recognition accuracy.At the end of the paper,we summarize the work of this paper and then prospect the future works.