Abnormal Human Behavior Analysis Based On Video Streams

With the rapid development of the information age, video surveillance equipmentshave been adopted massively, the data of it has also increased dramatically, so it hasbrought great challenges for real-time monitoring alarms and video data retrieval effectively.When confronted with crimes, it often needs us to look up captured video clips manually tofind effective key clues to solve the case. However, this manually observation method isusually inefficient, poor real-time. There is an urgent need to take full advantage ofcomputer technology to tap the key clips from massive videos, and quickly complete theautomatic search for the key object based on retrieval conditions. Therefore, the study ofhuman behavior analysis based on video clips has a very important theoretical significanceand application values.The research background is based on the video environment, This research mainlyinclude human behavior detection and identification under complex backgroundenvironment. The research is divided into four respects as following: First, the rapid humanbehavior detection under videos, we have proposed a background subtraction algorithmbased on Gaussian background model to complete human motion detection aftercomparatively analysis of common motion detection algorithms. Second, the quick andefficient moving target tracking, after comprehensive compare with the relevant motiontracking algorithms, this paper selects the particle filter algorithm to complete movinghuman tracking, and proposed a tracking model based on target color information. Third,we established human skeleton and parametric model, after extracting video frames,extracting moving targets, refining the object to get its skeleton, extracting skeleton keypoints with improved algorithms, converting behavior analysis problems directly toparameters solution problems. Fourth, the identification of abnormal human behavior, weintroduced the fuzzy theory in this paper, try to take advantage of the method of themaximum membership principle to complete human abnormal posture identification. Inorder to take advantage of trajectory information obtained during the target tracking process,we attempt to calculate the speed and direction data of the trace points, so that we can fuzzyrecognize the abnormal behavior during the action processes.The deep study of behavior analysis will promote the rapid development of futureintelligent video analysis, completing intelligent alarms of abnormal behavior under videos,thus promoting intelligent behavior analysis to be widely applied to everyday life.