Research Of Human Abnormal Behavior Recognition Based On Video Sequence

Human abnormal behavior recognition technology based on video sequences can discover and deal with human abnormal behavior in time,it has great practical value in ensuring social security.This paper is mainly about how to identify human abnormal behavior from the video sequence accurately and quickly,based on target detection and tracking,and according to the corresponding human body features to achieve the identification of human behavior.First of all,we should choose proper test video sequences,and carries on pretreatment to filter the noise in the process of video acquisition or transmission,offering reliable guarantee for evaluating algorithm performance objectively.Analysing and comparing the advantages and disadvantages of three common motion target detection method,and in this paper,we adopt three-frame difference method to detect the moving target,it has good resistance to noise and low complexity.Owing to the traditional three-frame difference method can't extract a complete moving target,this paper proposes a moving target detection algorithm through the combination of three-frame difference algorithm and SUSAN edge algorithm.It get the outline of the moving target through three-frame difference,and then obtain the static edge of the current frame image by SUSAN edge operator,conduct ‘and' operation on the two results can extract precise contour of moving object,and for further edge connection,area filling and morphological operation will extract the complete moving targets.This paper extract rectangle length-width ratio and centroid feature to realize human falling behavior recognition.In human fighting behavior recognition,aiming at large calculation and bad real-time characteristic of traditional optical flow method,this paper puts forward a fighting behavior recognition method based on target distance relationship and optical flow feature.Taking use of the distance characteristic of fight behavior,join the target distance constraints in traditional method,only conduct optical flow computation on the image which meet the distance constraint,and overcome the poor real-time shortcoming of optical flow and satisfy the the real-time requirement of behavior recognition.