Comprehensive Analysis Of Dangerous Movements In Human Population

With the rapid development of science and technology,the rapidly emerging deep learning technology makes target detection technology develop rapidly and begin to be widely used in various fields.A comprehensive analysis of dangerous movements in people will focus on human postures including "standing","sitting","walking","falling",and holding dangerous objects.The technology involved is target detection technology and key point detection technology,but the traditional detection technology has some problems,for small targets and missing targets detection effect is not good.At the same time,the complexity of human movements is too high and small key points will appear,leading to blurred or covered key points that are not easy to be detected.In view of the above problems,this paper conducted a comprehensive analysis of dangerous movements in people based on deep learning.The comprehensive analysis research involves the study of the holding of dangerous objects in the population and the dangerous actions in the population.Then,YOLO target detection algorithm is improved based on Shuffle Net structure to detect handheld small objects.Then,YOLO network is combined with Openpose network to optimize dangerous behavior detection research.Through key points,different parts are connected to form effective limb bones,and then human behavior is determined by the change of bones.Through these two studies,we can better understand the risk factors in the population.From the actual needs,this article focuses on solving the problems of dangerous goods in the crowd and people suddenly falling down in the crowd.The target of hand-held dangerous goods is usually relatively small,including individual hand-held guns and sharp controlled knives,etc.These hand-held dangerous goods are easy to be covered or blurred and difficult to be identified.This paper proposes a target detection algorithm based on Shufflenet improved YOLO,and the improved network improves the accuracy by about 9.6% compared with other networks.When someone suddenly falls down in the crowd,it can be identified through the human body key points.When the key points are difficult to see or blurred or covered when the distance is long,based on this,the target detection method of YOLO network combined with Openpose network is proposed to solve the above problems.The experimental results show that,The accuracy of remote attitude recognition has also achieved a good effect,the accuracy of 95%,while maintaining a good real-time.