Identification Of Human Abnormal Behavior Based On Video Image

In recent years, human motion analysis is a very challenging research topic in thefield of computer vision at home and abroad, widely used in medical diagnosis, safetyprotection, human-computer interaction, and other fields, with broad application prospectsand high research value. Mainly focused on the abnormal behavior of human in the interiorcorridors, this paper proposes the identification method of human abnormal behaviorsbased on motion characteristics, that is, detecting in real time, tracking people in the videoand then identifying the abnormal behaviors. It is mainly divided into three parts: targetdetection, target tracking and identification of human abnormal behavior.In the target detection part, use the background subtraction method of blockingupdate to achieve target detection. Update the background in real time through theblocking of difference image. The advantages of blocking is that the background block willnot be judged into foreground one and has a certain anti-interference ability, reducing theamount of computation, so that can extract the full body contour.In the target tracking part, a tracking method based on the combination of associationarea and color histogram has been proposed, carrying on the tracking of human body bythe color histogram-matching rate assisted by the area-matching rate. This method suppliesa better solution for leaky tracking, false tracking and other phenomena for the problem ofinaccurate tracking of targets when they move nonlinearly, improving the stability.In the part of identification of human abnormal behaviors, for the identification of"walk","run","jump","squat","climb" and "wander", these acts except for "walk" aredefined as abnormal behaviors. According to the results of human detection and tracking,deeply analyze the moving trail, moving posture and moving time three parameters, putforward different criteria for the different behaviors, and finally identify the five abnormalbehaviors and alarm.The development tools used in this paper are Microsoft Visual Studio2010. OpenCVcomputer vision library software programming is used to establish video database by shooting video and test the accuracy of the algorithm for many times. Experimental resultsshow that the algorithm in this paper can not only detect and track human body in real time,but also quickly and accurately identify the abnormal behaviors, with a simple andpractical feature.