Research On Human Abnormal Behavior Recognition Algorithm

The development of the economy increases the difficulty of social security.In recent years,many accidents occur in airports,railway stations,schools,banks and other public places.However,the technology of real-time video monitoring automatic alarm are not reliable enough.The traditional video monitoring system cannot meet the demands of people,so research of surveillance system becomes important.Abnormal human behavior recognition is one of the hot research topics,which extract the body characteristics to identify abnormal human behavior and alarm quickly.To improve the accuracy of human behavior recognition algorithm,after human body target detection,moving target tracking,two kinds of human behavior recognition algorithm are studied.One is based on improved FSVM employing dynamic and static features,another is template matching method based on Hu-moment and texture.After target detection,a variety of methods being compared,background difference method model is improved to propose double background model method,the static background using the average background modeling,dynamic background using the optical flow modeling.Moving target tracking,an improved particle filter tracking algorithm is adopted based on two features combined.In the aspect of human behavior recognition,two kinds of recognition methods are studied.One is template matching method,which combines Hu-moment and texture features,the similarity is computed by Mahalanobis distance.Another is based on improved FSVM employing dynamic and static features.Fuzzy membership degree of fuzzy support vector machine(FSVM)is improved.