Human Behavior Recognition Based On Video Key Frame Optimization

Video-based human behavior recognition is one of the hot-spots in the field of computer vision research in recent years,and is widely used in human-computer intelligent interaction,video surveillance,virtual reality and other fields.With the rapid development of multimedia technology and network information,a large number of video data are flooded around us.How to retrieve effective and critical information from a large amount of video data within a specified period of time is a key issue that needs to be resolved urgently.The key frame is a frame or several frames of images that reflect the main content of the camera.It can not only describe the main visual content of the video in a simple and general way,but also provides a good data preprocessing function for later applications.In the research field of human behavior recognition,extracting video key frames can effectively reduce the amount of video indexing data,thus improving the accuracy and real-time of motion recognition.In order to improve the representativeness of video key-frame,this paper proposes methods of key-frame sequence optimization for human motion video,daily surveillance video and other data,and then using the optimized key-frame sequence for behavior recognition.The method of classifying and identifying the extracted initial key frames in the traditional behavior recognition method is improved,and the recognition time is shortened on the basis of ensuring the recognition rate.The main work includes the following two aspects:(1)The Kinect depth camera is used to obtain the human skeleton information and describe the 3D human skeleton features including the location and angle information of the key nodes,using K-means clustering algorithm to extract the initial key frames in the human motion video sequence,and then perform secondary optimization based on the position of the key frame in the sequence to extract the optimal key frames and obtain the optimized key frame sequence.Finally,using the CNN classifier to identify behavioral videos based on the optimal key frames.(2)In the phase of video key-frame optimization,a color feature-based video key-frame extraction and optimization method is proposed for daily video surveillance data.Firstly,the RGB color feature is constructed for the video frame image,and then the key frame sequence is initially selected according to the similarity degree between adjacent frames.Finally,the Canny edge detection operator is used to eliminate the redundant frames by calculating the edge matching rate,which also achieves the optimization goal.Experimental results show that the method has good self-adaptability and can effectively avoid repeated extraction of key frames due to differences in grayscale distribution.The final extracted key frames can well summarize and express the original video content.