The Research Of Moving Target Detection And Shadow Elimination Based On Video Sequence

Moving target analysis technology based on video sequence has been widely applied tointelligent video surveillance, terrain matching, image-guided navigation, the analysis of hu-man motion details, artificial intelligence and other fields. The moving target detection is themost basic of Moving target analysis technology, which belongs to the basic work, completeand accurate moving targets extracted from the video image is the key of subsequent videoimage motion analysis can proceed smoothly.Domestic and foreign researchers for movingtarget detection technologies do a lot of detailed and in-depth research,and have achieved goodresults in many ways. A variety of moving target detection algorithm technology is one of thehot areas of computer vision research projects.Firstly, the research background and significance of the subject are described in detail,and the status quo and development trend of related areas of research are reviewed. Through-out the entire process of moving target analysis techniques were outlined, focus on targetdetection, shadow detection and elimination, to eliminate image noise and related algorithm.Secondly, this paper focuses on the target detection algorithm based on backgroundmodeling,try to find problems and improving methods.Background model for the directextraction method, statistical average, W4background modeling method, a single Gaussianbackground modeling method and Gaussian mixture background modeling method wereanalyzed.In the initialization process of the background,the traditional Gaussian mixture modelrequires no moving objects in the scene background,otherwise tend to put the object into thebackground model.For this shortcoming,in the initialization process of the back-ground,obtaining accurate initial frame using median filter even the presence of moving objectsin the background scene.When conventional Gaussian mixture background model update, themean and variance of the model have taken the same value,This will lead to convergence andstability of the background model can not be met.This paper also presents an improvedmethod.At the beginning of the mixed gaussian background model training, new ways to makeupdate rate of mean and variance are larger value, which makes the mean and variance of modelupdating speed becomes faster, makes the model learning speed becomes faster,and this will beto more quickly obtain a model that matches the background.In a larger update rate aftertraining for a period of time, let the mean update rate unchanged, variance update rate becomesmaller, there would be make background model will not have the too big fluctuations, stableperformance is better.Finally, because the difference between the shadow attributes and background properties,as shadows and moving targets with the same motion characteristics,in the target detection process, moving target was always accompanied by the shadow. In this paper, shadow elimi-nation algorithm based on statistical and color features invariant and color models wereanalyzed, shadow elimination algorithm of HSV color space can accurately eliminate sha-dows but with a great amount of computation, so an improved method is proposed, it can re-duce amount of calculation and improve the real-time performance.Through the experiment,the method can effectively eliminate the shadow.