Research On Moving Object Detection And Tracking In Video Sequences

The technology of object detection and tracking in video sequences is one of the hotspots in the field of computer vision. It concerns many fields such as image process, video process, pattern recognition and artificial intelligence and has wide application in intelligent transport, television surveillance, human-machine interface, video compression and coding, and so on.Some common approaches in the field of object detecting and tracking are introduced briefly, including frame subtraction, background subtraction and optical flow in object detection and model-based tracking, region-based tracking, feature-based tracking, deformable-template-based tracking in object tracking. All those algorithms are analyzed and their advantages and disadvantages are concluded. Then the related knowledge on morphologic operation and connectivity analysis is discussed.In terms of object detection, the background subtraction is emphasized in this paper and two improved algorithms are proposed. One is moving object detection based on RGB color space. It improved the median modeling in the stage of background modeling; in the stage of foreground object extraction, the information of color is used to avoid mistaken detection caused by the similarity of gray between some background pixels and foreground pixels. The other one is moving objects detection based on pixel classification. In this algorithm, the pixels are classified by its characteristics into two categories: the single-model and the multi-model. The former uses median method for background modeling while the later use Gaussian Mixed Modeling (GMM). The experiments shows that this algorithm, comparing with GMM, is computational efficient and can be used for real-time systems easily.About object tracking, this paper pays great attention to Mean Shift algorithm. It is a good algorithm, and many scholars in foreign countries have developed this algorithm in recent year. The OpenCV tool is also introduced. Some improved algorithms are analyzed including fast moving objects and object occlusion, which are experimented on VC6.0 platform.