Abnormal Behavior Detection In Intelligent Video Recognition System

An intelligent video recognition system is a system depends on an algorithm of intelligent video analysis, which detects and recognizes specific object in selected area based on detection algorithm automatically. Whenever some abnormal behavior occurs, system will send warned message to notify surveillance man to deal with.The article takes the intelligent video system which used to recognize abnormal behavior as research content. The article takes advantage of opencv combine with algorithm in image processing and computer vision to achieve each part of system, and finally integrates them as an intact system which can fulfill following functions:people detection, cross detection for line and area, congestion detection, missing detection, and amount detection for car. The specific content show as follows:In pre-processing part, analysis and compare have been made for RGB and HSV color space. Though result of figure testing, a conclusion that HSV color space has a better effect in feature abstraction. In addition, the article takes a brief introduction about median filtering, and then makes analysis and conclusion on the basis of test result.In motion detection part, firstly, analysis and compare have been made for mainstream motion detection algorithm. In consideration of drawbacks in traditional GMM, for example hole in object area, and poor adaptability of background updating, the article put forward an algorithm of fast and slow background which takes advantage of different rate of background updating to amend foreground figure, and then test the combination of traditional GMM and algorithm of fast and slow background practically which have proved that the algorithm is effective.In post-processing part, firstly, an introduction of some edge detection operator in common use and corrosion and inflation in morphology operation has been made. After that, the article discusses open operation emphatically, and then testifies its effect in area connection. Furthermore, the article introduces curve fitting in convex hull and polynomial, and a comparison and analysis has been made with these two curve fitting algorithm in effect of sketch amendment. Finally, an overview about shadow removal has been made, and an algorithm of shadow removal based on HSV color space has been put forward which is proved by testing for its better ability in shadow removal compare to traditional algorithm.In judgment part, the article puts forward some standards toward judgment of abnormal behavior, and verify its validity by experiment.Finally, a demonstration about availability of the intelligent video recognition system designed by author with all detection functions has been made.