Research On Human Action Recognition And Behavior Analysis Based On Kinect Human Skeleton

Human action recognition is an important field of artificial intelligence,the body feeling game,video surveillance,military training,health care and other fields has been widely used.For human action recognition based on computer vision is generally based on two dimensional color image processing to obtain information,so vulnerable to environmental and lighting effects,so in this paper,using Microsoft development device sensors to obtain depth image,has been treated by bone joint point of three-dimensional coordinates,according to the feature extraction of vector of human body structure,in view of the human body static posture and identification of behavior research,the main work is as follows: In understand the basic structure and performance of device sensors,on the basis of studying the depth information and the principle of human body skeleton information acquisition and perceive,determine the condition of static posture bone databases,but with the behavior selection of database research.According to posture database and behavior which human body each joint point of three-dimensional skeleton information,identify the basis of human motion is to extract the skeleton characteristic value of the information,the three-dimensional skeleton information obtained in this paper,based on the device body structure is constructed vector,Angle of structure between vector and vector modulus ratio combination described as human body posture and behavior characteristic vector,so the feature vector has the translation and scaling invariance,and can easily describe the complete information,is very suitable for human action recognition.This paper static posture and the dynamic behavior of human action recognition,the first to use posture training neural network database,will be to identify the action into the trained neural network to classify recognition,recognition is improved by using genetic algorithm to optimize neural network,and then based on dynamic time neat algorithm will be to identify action and behavior in the database to match,get and use the improved dynamic time neat algorithms reduce the recognition time.Through the system test shows that this paper uses genetic algorithm to optimize neural network with the improved dynamic time neat algorithm to identify the pose and behavior has a better effect,illustrates the feasibility of the method of feature vector selection and gesture recognition,research work has a certain application prospect and academic value.