Research Of Human Behavior Recognition Based On Kinect

Human behavior recognition based on Kinect is according to the principle of Kinect machine,in the video,three bones and depth information are obtained from it.The application of the bones and depth information can be identified effectively in human recognition.In traditional video behavior recognition,the acquired data solves the problem of low recognition accuracy or confusion between behaviors caused by the effects of wear or other coverings,and also solves the error between the human body and the background.This method usually gets 15 or 20 key joints’ key information of the human skeleton,then,according to the geometric relationship between the joints to describe the behavior of the human body.In some simple behaviors,Limb movement are only related to parts of movements and information,so how to improve this phenomenon is the key to solve the problem.In this paper,to improve human behavior recognition accuracy,we analyses shortcomings of human behavior recognition based skeleton carefully and improve it.The main work of this paper includes the following two aspects:1.In method of human behavior recognition based on Kinect,we use absolute joint position information or relative joint position information to represent skeletal information generally,but those information often have lots of data noise and can’t distinguish the similar actions on the same limb,In this paper,a new method of joint fusion is proposed to solve it.First,the whole skeleton is divided into five parts according to the structure of the human body,that is trunk,left and right arm,left and right leg,fusion joints information which in five parts.Then,the information of fusion is calculated according to the formula of the absolute joint position information and the relative joint position information respectively,and cascade the absolute joint position information and relative joint position information corresponding to get new descriptors.2.In order to eliminate irrelevant or redundant features information of body joints cut down recognition accuracy,an improved CSO(Chicken Swarm Optimization)is used to optimize features information of body joints by filtering unnecessary data.The Good Point Set and OS(Ordered Subsets)are added tothe improved CSO algorithm.The Good Point Set is used to uniform initial population.The OS method divides entire ordered population into three parts correspondingly.During the process of updating position information,regard the Monkey Algorithm’s gaze afar motion as the process of gazing afar when chicken looking for food,and establish revised formula.It improved the convergence speed of algorithm and avoid falling into local optimal.