People’s Behavior Analysis Under Dynamic Scenes

People’s behavior analysis under dynamic scenes, based on people walking posturechange, is a new technology, and it mainly relies on statistical signal processing theory andmathematical statistics probability theory. With the development of computer technology,multimedia technology, pattern recognition and machine vision, people’s behavior analysisunder dynamic scenes gradually become a hot research topic and key research directions.Compared with sensor technologies, the proposed method easily accessible, simple to use,and can be used as a non-contact multimedia video monitoring, pattern recognition, machinevision technology to carry out research. This research has important significance and a widerange of application prospects in the field of production and life safety, community safety,medical diagnostics, the security zone monitoring behavior. People’s behavior analysisincludes four main sections: background modeling, image segmentation, the moving objectextraction and analysis of people’s behavior; People’s behavior is divided into threecategories: normal walking class, fall classes and the squatting classes. Extracting andtracking moving object is the key action of the people’s behavior analysis.This paper focuses on image segmentation, and moving object extraction，afteroperating mathematical morphology, the foreground and background can be segmented withthe target can be extracted by improved Gaussian pyramid optical flow method. In the deepresearch of people’s behavior analysis under dynamic scenes, there were some works done.Such as: the research background, the research meanings and people’s behavior analysisunder dynamic scenes. According to related research at home and abroad, this paperpresented comprehensive overview in the related field, compare and analysis the classicbackground modeling methods and moving target detection algorithm. This paper adoptedimproved drift means method to create a background, and then segmented the images basedon graph theory, followed extracted a moving target under dynamic scenes, classed movingtarget into three categories by tracking them.There are there innovation in this paper, the first one is it proposed an improved driftmeans algorithm to create a background model, which ensures it meet the need to adapt tothe dynamic scenes, the second is this paper invites an image segmentation algorithm basedon graph theory, which can separated the background and from foreground well, and the thirdis based on the moving targets tracking by online learning mechanism, this paper proposedimproved Gaussian pyramid optical flow method to extract the target in the image. In thispaper it classes human behavior analysis into three types; other behavior can be classifiedaccording to the analogous instructors, and then launch research. The experimental materialin this paper includes the videos shot by myself, Weizman video database and KTH videodatabase.