Moving Detection And Tracking In Video Surveillance

Traditional video surveillance systems primarily based on watching the monitorscreens by human and auto recording. Because the persons'attention are limited, abnormalor hazards situation can not be found in time. Usually, the monitor staff calls the originalsurveillance record after the accident. While such video surveillance can provide great helpto find crime, it dose not avoid the occurrence of crime in time. To address theshortcomings of traditional monitoring, computer vision technology has been applied invideo surveillance and made intelligent surveillance developed. Intelligent surveillancesystem can analyze the video data and drop the data which people do not care about. It onlyprovides useful information for further analysis. When abnormal situation or dangeroussituation is detected, it gives alarm to the person worked on surveillance. In this way, themonitor staffs can deal with the situation in time.This paper is based on the project–Content-Base Retrieval of Network VideoMonitoring Machine. The main work is motion detection and tracking. The purpose isproviding information for classification and identification in the later. By studying andresearching on computer vision technology and reading a lot of excellent essays at homeand abroad, a new motion detection and tracking algorithm for video surveillance is put andimplemented in ARM mini2440 experimental platform.First of all, by studying the existing motion detection algorithm deeply and doing alarge number of experiments, comparisons were drawn between several commonly usedmoving object detection algorithms on algorithm complexity, noise sensitivity, and theability to solve the problem that moving objects keep in background. In order to eliminatenoise like flying birds or swing trees, I combine frame difference with backgroundsubtraction and it works well. Threshold is very important in moving detection. By studyingon Oust Threshold segmentation, I choose Oust Threshold segmentation with a differentpercentage in different experimental .The result of the experimental shows that dynamicthreshold works better than fixed threshold in dynamic environment.Because of the complex and changing environment and diversity in the color ofmoving object, motion information contains a lot of noise and some holes in foreground.These make the motion detection information is different from real object and also make apernicious influence on later classification and recognition. In order to restore motioninformation, by studying on image processing technology, the author put a new method oneliminating noise in the experiment data.Secondly, the author learns and researches on the theory of Mean Shift and put it intoobjects tracking. By establishing model of colors for object, mean shift vectors can find where the object is. To reduce the complexity of the algorithm to meet the requirement ofreal-time monitoring system, the author convert RGB color space to HSV color space andestablishing model of H -component color only. Experiments show that the algorithm haslower time and space complexity and has the same tracking effects as before. Mean shiftdose not works well on tracking objects that move fast. Kalman filter can predict theposition of the moving objects in next frame through the information of position, speedacceleration in the current frame. Mean shift tracks fast-moving objects very well throughpredicting position by kalman filter.In order to meet the real-time video surveillance, there is a new method has been put inthis paper to track moving objects. This method combines physical characteristics of themoving objects and mean shift. Because the position of the moving object's centroidchanges steadily between frames at time t and t-1, centroid is chose for tracking movingobjects. It tracks moving objects in a good result when there is no collision in movingobjects at current frame. Mean shift and kalman filter will start to tracking instead whenthere is some collision in the image.Last, tracking moving objects algorithm is implemented in experiment environment ofARM mini2440. It contains two main parts. Preliminary moving detection will be done inARM mini2440 and the farther moving detection and tracking will be done in PC. Framedifference is chose for detecting motion in ARM mini2440. All frames are marked andtransported to PC when there is some motion has detected by frame difference, or a frame istransported in ten frames in order to keep the video fluency. In this way, we can knowpreliminary information in monitoring environment and reduces the amount of networktraffic. Through receiving frames by network, the marked frames are chosen to fartherdetect and track in PC. The author combines image difference and background subtractionto detect the moving objects, then, removes image noise. The result of moving detectionwill be used for tracking. The author combines physical characteristics and mean shiftwhich has been introduced above to track moving objects.It detects moving objects and tracking with excellent results in the environment thatthere are not very many persons or cars and meets the requirement of real time. But it doesnot work well in complexity environment such like parks, squares, traffic intersection, etc.The purpose of moving objects detection and tracking is providing information for targetclassification and behavior understanding. A good result of motion detection and trackingwill improve the accuracy of target classification and behavior understanding very much. Inthis way, the intelligence surveillance system will analyze video data better and be moreintelligent.In sum, by researching on motion detection and tracking deeply, a new method is putto tracking moving objects in intelligence surveillance system in this paper. It got excellentresult in some kinds environments. Today, intelligence surveillance system has been used insome environment to improve the functionality and efficiency of surveillance system. It may instead of persons who work in security fields to process security work. With thedevelopment of the computer vision technology, intelligence surveillance system will beincreasing towards perfect.