Research On Movingtarget Detection And Tracking System Based On Video

In recent years, the production of security cameras in China is growing at about20%per year. This wave has introduced a lot of video data processing tasks. Thetraditional video monitoring system cannot meet the growing need. Study ofintelligent video monitoring system has attracted more and more attention. This articleis a deeply study based on a static camera surveillance scenario. This paper consists offour parts, which are the moving target detection, shadow removal, moving targettracking and behavioral understanding.Moving target detection, which is the foundation of the whole process of videoprocessing, are highly related to the subsequent process. In this part, several commonused algorithms for moving target detection were introduced and analyzed. Moreover,the background difference method was improved. Due to the difficulties ofbackground difference method and the amount of calculation are focused on thebackground modeling and update process, this paper introduced the method ofbackground difference to solve these problems. On the basis of guaranteeing theaccuracy, the method improved the efficiency of the background modeling and updateprocess. Results can been found from the experiments presented in this part.Because of the influence of the light, moving target detection sometimes can beaffected by the interference of shadow. Therefore, shadow removal is a very importantprocess after detecting the moving targets. Through the analysis of the characteristicsof shadow and the algorithms for shadow removal, the author introduced an improvedalgorithm for shadow removal, based on RGB color space. The algorithm used thetheory of color constancy and characteristics of shadow, to detect the area duringmoving target detection process. The algorithm can efficiently remove the shadow andaccurately show moving target.Moving target detection technology can detect different moving target indifferent images. However, moving target tracking technology can track one movingtarget in continuous image sequence. To get the direction, the track and the speed of amoving target is to track a moving target. The author carried out a research on themean-shift algorithm. The principle and the process of the algorithm is simple.Moreover, it is easy to be improved, combining with other algorithms. Results can befound from the experiments in this part. Behavioral understanding technology has always been a hot issue in the field ofcomputer vision. Since there is no general method, which can completely solve theproblem of behavioral understanding till today. Most of the methods are limited to aparticular environment or a certain behavior. The author introduced and analyze thelayered behavior model, and proposed his own suggestion, wishing to help otherresearchers in this field.