Detection Of Abnormal Behavior In Supermarket Based On Hand Image Feature Analysis

With the increasingly prominent problem of public security and the explosive growth of video data,intelligent monitoring has gradually become a hot research topic in computer vision.The analysis and detection for the abnormal behavior of supermarket theft phenomenon,can not only reduce the monitoring personnel labor pressure,also decrease economic losses of supermarket.At the same time,also have a certain role in promoting stability and unity of society.Therefore,it is of great theoretical and practical value to analyze and detect abnormal behavior in supermarket monitoring.To analyse and detect abnormal behavior in supermarket monitoring,the first step is to detect moving objects.Because the theft behavior "taking" and "hiding" the goods in the supermarket is mainly done by the hand,therefore,according to the skin color feature and size,position and geometric characteristics of the hand in the moving target region detected,the hand can be detected and segmented.Then tracking the hand,the Mean Shift tracking method combined with Kalman filter is first used in the hand tracking of supermarket monitoring,the iterative initial position of the Mean Shift algorithm was predicted by the Kalman filter;and according to the change of hand motion to judge the "Time-space" of "take goods" and "hidden goods".The experimental results show that the tracking algorithm reduces the error of the Mean Shift algorithm in the case of hand's fast movement and complex background,and realizes the accurate tracking of the hand's area.For abnormal behavior detection,this paper proposed a method for detecting abnormal behavior in supermarket based on hand image feature analysis.After the detected "take goods time-space",extracting color and texture histograms of the hand and moving pixel region around,and getting a similarity measure with no goods in hand using the Bhattacharyya distance,and compared with the mean similarity to judge "take goods".To distinguishing "hidden goods" behavior,the premise is that "take goods" behavior is done,after "hiding goods time-space",extracting color histogram and texture histogram features of the hand and movement pixel area around,the similarity measure is obtained with the goods in hand and no googs in hand using the Bhattacharyya distance,and detect the "hidden goods" behavior compared to the mean similarity.In the end,the feasibility and effectiveness of the proposed algorithm are verified by many experiments,and analysis the influence with different monitoring distance and commodity size,and by comparing with other experimental results,the validity of the proposed algorithm is further verified.