Human Behavior Recognition Research Based On Hierarchical Classification

The human behavior recognition is an important subject in the field of computer visionresearch, which is a very attractive and challenging problem in recent years. The human actionof visual analysis is a new frontier field of research, involved in the multi-discipline,such aspattern recognition, image processing, computer vision, etc. It can be widely used in the field ofmotion analysis, motion capture, intelligent robot, human-computer interaction, video intelligentmonitoring system;therefore,the human action behavior recognition has greatly research valueand practical significance. However, the human body movement is a rigid body motion,existence difference in appearance, shape and motion mode, lead to human body behavioridentification difficulty is greater, there is no a general behavior recognition model, most of theresearch are in a particular setting.The purpose of this research is to realize a single body go, waved, jump and bend down towait for a few kinds behavior recognition, and the recognition process of the algorithm usedsome research. Due to the moving target detection is the first step of human behavioridentification, test results will have a direct impact on behavior identification results, so in thispaper, at first the human movement detection algorithm is studied. At present the moon are usedfor the comparison of the moving target detection method has been cond ucted, and for eachalgorithm and analyzes the advantages and disadvantages of selected at the end of the paper,gaussian mixture model in this paper as a method for moving object detection algorithm. Thehuman body for detecting target, this paper, according to the location of the object, get the targetof minimum rectangular area.The detection of human movement, this paper again feature extraction, and featureextraction algorithm was studied. This paper mainly adopts based on time and space of interestpoints feature extraction algorithm, in the point of interest around extraction based on3dgradient and description to describe the son, the son do clustering operation, then use supportvector machine for behavioral model of training and the final behavior prediction.The method to use MATLAB simulation, and then use C, C++, MFC, OPENCV makeapplication software. Mainly in the Weizmann video database and KTH video database trainingand testing, and in recent years, some of the most advanced methods are compared. Theexperimental results show that the method has high accuracy. For further practical applicationprovides a good theoretical basis and experimental basis.