Study On Human Abnormal Activity Recognition Based On Video Image

Human abnormal activity recognition analyzes human behavior within the scope ofmonitoring range, human behavior is to match the preset abnormal behaviors, if they matcheach other, the human behavior is judged to be abnormal behavior, then alarm automatically.Traditional monitoring system is simple, mainly rely on the manual operation of themonitoring personnel to monitor short and random unexpected and unusual event, requiringmonitoring personnel to monitor the screen continuously, it requires not only a large numberof monitoring personnel, but also it can not do a complete real-time monitoring to result inomissive alarm and incalculable loss because of the lax of the monitoring personnel andexcessive monitoring points.This thesis research includes: firstly, based on the background subtraction method,Gaussian mixture model is improved as a background model and Gaussian mixture modelparameters for the mean and variance use different learning rate to improve the accuracy ofthe background extraction. Markov random field theory is introduced to the foregroundsegmentation and improves the segmentation accuracy of moving target, and this paperproposes a method which human movement can be detected effectively when moving target isat rest. In the superiority of shadow detection of HSV color space, a fast and accurate methodis proposed to eliminate the shadows; Secondly, in order to be able to track instantly andautomatically and avoid the effect of keeping out each other, a new method that Kalmanalgorithm is combining with Mean shift algorithm is proposed to obtain human movementinformation in advance and achieve quick search; Finally, a feature extraction method isproposed, Hidden Markov Model is used for training and classification to identify humanabnormal behavior. Experimental results of the proposed method show that the proposedalgorithm can identify faint, smashed cars and other human abnormal behavior recognitioneffectively.