Research And Application Of Multi-person Pose Estimation Algorithm

In recent years,with the development of deep learning,the field of human pose estimation has developed rapidly.Human pose estimation is an important basis for human action recognition.Which can help computers understand human behavior and promote the development of the intelligent robots,virtual reality technology and better human-computer interaction.In addition,human pose estimation has a wider application in the field of medical assistance and video surveillance,which can assist humans to monitor video content and avoid false detections caused by fatigue.Based on the research of OpenPose multi-person pose estimation algorithm,we find that the OpenPose algorithm has the problem of false detection of key points in where was no people.To solve this problem,we propose an algorithm which combines top-down and bottom-up structure.The multiperson pose estimation algorithm effectively solves this problem.The main work of this paper is as follows:(1)We analyze several classical object detection algorithms which could be used in multi-person pose estimation algorithm of the top-down structure.In addition,we also discuss the representative algorithm in the multi-person pose estimation algorithm and the fall detection algorithm.(2)We propose a multi-human body pose estimation algorithm combining top-down and bottomup structure.The algorithm adopts a top-down structure to detect the area of the human body in the image through a human body detection algorithm,and then combines these areas and then send the combined areas into the pose estimation algorithm of the bottom-up structure for pose estimation.(3)In this paper,we propose a fall detection algorithm based on pose estimation.Firstly,we extract the key point information of the human body through the pose estimation algorithm proposed in this paper,and then classify these key points by an SVM classifier to get the human body's status.The status of the human body is then analyzed to determine if there is a fall event in the video.And,the algorithm is robust to the video which exist more than one people.(4)We applied the pose estimation algorithm proposed in this paper to video surveillance and developed an intelligent monitoring system.In response to the needs of the Jiangsu Food and Drug Administration's application,the system mainly monitors the kitchen for canteen,including key points extraction module,clothing identification module,face detection and recognition module,and item detection module.