Human Pose Recognition In Video Surveillance

In the rapid development of science and technology today,machine vision technology and human social life increasingly close connection.Human pose recognition is an important part in the field of machine vision,with great application value,widely used in medical research,sports,virtual reality,intelligent monitoring and other fields.The purpose of human pose recognition is to automatically detect,track and identify human body posture by using machines,so as to realize the interaction between people or the surrounding environment.This paper has done research in the following aspects:(1)Based on the traditional frame differential method in detecting foreground object detection,there will be the emergence of "noise group",this is because from the traditional frame differential method to get the template ghosting in the background of reasons,this paper used in target detection method is improved on the traditional frame differential method,can effectively avoid the appearance of ghost,hollow,ghosting,and makes the background pixels preserved,Compared with the traditional method,the algorithm has better real-time performance and higher accuracy.(2)In the process of moving target tracking,the traditional MS tracking algorithm tracks the moving target through the color or gray characteristics of the region,which is difficult to solve the tracking problem in the complex and changeable actual scene.The MS tracking algorithm proposed in this paper not only uses the color features of the target,but also integrates the LBP texture and color image gradient features,and combines the moment information of the target region to calculate the size and direction changes of the foreground target in the tracking process.The new algorithm provides a better solution for solving problems such as light changes and background interference,improves the fault tolerance rate of MS tracking algorithm for target tracking in the case of single color feature,and enhances the robustness of tracking algorithm due to the change of target scale and direction.(3)The algorithm of human key points detection is studied.The Open Pose algorithm constructed from bottom-up human skeleton is used to accurately predict the confidence graph and partial affinity fields of human key points through VGG19 and two branch networks.Human key points are detected and their respective nodes are assigned to each person through clustering.The framework construction and key points extraction are realized,and the algorithm is fast and time-consuming.(4)In the process of human body gesture recognition,using three human feature subset contour to the human body was described,the traditional support vector machine(SVM)based on the design of multi-core vector machine,can effectively solve the problem of increased because the data and the multivariate,can express vast amounts of data samples,the method is compared with the traditional gesture recognition algorithm,both accuracy and recall rate will be higher.