A Research On Surveillance Video Based Human Abnormal Behavior Analysis Algorithm In Supermarket

With the development of science and technology and popularization of intelligent devices,video surveillance systems are spread all over the country's supermarkets.At present,the supermarket monitoring system still requires security personnel to take turns on duty,but it still cannot avoid the huge economic losses caused by theft and other criminal acts to the supermarket every year.At the same time,there is little research on abnormal behavior detection in supermarkets,and intelligent video surveillance is still not popular in native supermarkets.Therefore,the identification and analysis of the abnormal behavior in the supermarket monitoring video can reduce the work pressure of the monitoring staff in the supermarket and reduce the cost budget of the security personnel.Therefore,this paper has important theoretical and application value for the related research on abnormal behavior analysis in supermarket monitoring.This paper focuses on the abnormal behaviors that occur in supermarkets.The main work and results are as follows.(1)This paper gives the analysis rules of abnormal behavior in supermarket surveillance videos.By analyzing multiple pieces of supermarket surveillance videos,observing people's shopping behavior in the supermarket,and consulting relevant literature,the abnormal behavior in the supermarket is defined as shoplifting.The size of the product is limited in the rules,which must be larger than the hand of the identification target,and only for single-person targets and the "hiding" action is visible for abnormal behavior recognition analysis.(2)Aiming at foreground object extraction,combining the RLBSP texture features of image and HSV color space,a background modeling algorithm based on statistics is proposed.This algorithm uses RLBSP texture features' insensitivity to illumination changes and the HSV color space is more in line with the human eyes' color system.The background model initializes the color and texture features at the same time.In order to solve the problems of foreground resident and environmental lighting changes affecting subsequent segmentation,the post-processing method is used to modify the detection results,and the background template is updated with the corrected foreground detection results.(3)Using the skin color feature and the positional relationship of the hands in the upper limbs,a hand detection algorithm is proposed.Determine the approximate position of the skin and narrow the detection area through the skin color detector.At the same time,use the arm detector to obtain the position of the arm,according to the prior knowledge of the hand and arm and the relative position relationship,and combine the skin color area obtained by the skin color detection to obtain the human hand position.The experiment proves that the algorithm can detect the position of the hand more accurately.Kalman filter-based Camshift target tracking algorithm is used to track the hand target,and the analysis rules are based on the behavior of "hiding goods",and the current tracked hand results are compared with the standard "no goods in hand" state.Perform a similarity measurement of the Bhattacharyya distance based on the color histogram and LBP texture feature of the state of "goods in hand" to determine the current state,and then determine whether the current hand is within the body area,and integrate the hand state and position information to detect the "hiding product" behavior.