Moving Object Segmentation Method Research In Complex Scenes

Intelligent video surveillance in recent years has increasingly become hot in the field ofcomputer vision, and it has a very wide range of applications. Intelligent video surveillanceincludes several parts which are object detection, segmentation, tracking, behavior analysisand so on. Object segmentation is to segment moving objects from video, which is the mainprecursor of object recognition and behavior analysis.The research on object segmentation has made great progress in recent years. In allusionto videos with background change, background motion caused by moving camera,background and foreground have similar color, results of classic object segmentation methodsare not very well. Consider of time complexity and effectiveness, a new segmentation methodin complex scenes based on Stauffer and Grimsons’ pixel-level Mixture Gaussians Model isproposed.Classic Stauffer and Grimsons’ method is to establish pixel-level mixture Gaussiansmodel for background. According to the mean and variance changes, determine if the currentpixel is in accordance with the background and segment it. According to space correlation,this paper proposed a block-level mixture Gaussians model, and integrated color channelsintersecting into Stauffer and Grimsons’ method and improve its efficiency.Through quantitative analysis and qualitative analysis on several sequences exhibitstationary or dynamic backgrounds, illumination changes, and can have been shot by a littletremble camera, summary the choice method of parameters in different scenes which includecolor channels, block size, learning rate, number of mixture Gaussion models and so on.Experiments and comparisons to other motion detection methods demonstrate the proposedmethod has better performance for video analysis in complex scenes, which effectiveness isproved through quantitative analysis.