Recognition Of Human Abnormal Behavior In Intelligent Video Surveillance System

In recent years, with the development of computer technology and the increasesecurity needs, the intelligent video surveillance has become a hot research area ofcomputer vision. Compared with the traditional video surveillance, the intelligent videosurveillance using intelligent algorithm to gives computer autonomous decision-makingability related field. Computer-assisted and instead complete the monitoring task,which required by staffs and thus reducing the burden on them. Intelligent videosurveillance system is able to detect and identify abnormal behavior in time. This abilityis used to reduce the incidence of various types of abnormal events, thus can save a lotof human, financial and material resources. Because intelligent video surveillance hasbig advantages and huge demand along with tremendous economic value, domestic andforeign scholars and research institutions concerned about the intelligent videosurveillance system.Abnormal behavior recognition in intelligent video surveillance system is a relatedto the cross-application of computer vision, digital image processing, patternrecognition, machine learning and many other areas. This paper is focused on the keytechnical issues in the direction of several applications of intelligent video surveillancesystem such as target tracking, human detection and fall behavior recognition. Thispaper is organized as follows:(1) A combination of mixed target detection method which based on dynamicbackground modeling moving target detection and super-pixel feature target detectionmethod are created to achieve the goal. The variation of pixel values in studiesmonitoring the scene to determine the dynamic background updating models. Shadowcolor feature modeling is used to achieve the target of shadow suppression. This paperanalyzes the reason why using Gaussian mixture background modeling methods toextract a moving target empty. Then different background and foreground update rate isusing to solve the hollowing out of the problem.(2) In tracking of target, human is non-rigid target whose shape in videosurveillance is central standing committee deformation. Accurately track human targets for the long-term, we use the image geometric invariant moments Hu, the direction ofthe target gradient statistical description of the normalized histograms and10randompixels as the target template updated forecast target model with online learning methodto achieve target tracking.(3) In the area of the abnormal behavior detection, in this paper first use thecentroid position difference of the target, the high aspect ratio and the tilt angle as thecategories of normal and abnormal behavior characteristics, and then use a prioriknowledge of the above three features to quantify normal and abnormal behavior of aperson. the quadrant features based on the energy difference of the image of the sceneand human skeleton similarity characteristics are defined as the feature of fuzzyclassifier of fall behavior recognition. The results of sitting down, falls and walking testis given.