Research Of Moving Object Detecting And Robust Tracking Methods In Unstable Background

The technology of moving object detecting and tracking in video is a core area of research in image understanding, computer vision and so on. Now it is widely used in video surveillance, robot vision navigation, traffic control and other fields. Therefore the research of moving objects detecting and tracking has important significance.In view of the "ghosting" and smear phenomenon appear in the moving object detection and the problem of partial occlusion in the tracking multiple objects. The paper has researched of moving object detecting and robust tracking methods in unstable background. The paper has yielded the following results:1. In view of the slow execution speed of mixture Gaussian models and easily lead to "ghosting" and other issues when detected foreground object. This paper proposed a Quick Gaussian Mixture Background Model. In this method, through giving the constraints on the weight of the Gaussian distribution and survival time, we establish a mechanism for exiting the Gaussian distribution. The model selected the number of Gaussian distribution according to the scene for each pixel number. So the model accelerates the pace of implementation of the algorithm. In the model update process, through integrating with the frame difference, we solve the "ghosting" and smear phenomenon.2. After the Analysis of the existing shadow detection algorithm, an adaptive shadow detection algorithm based on CIE LUV color space and single Mixture Gaussian Model is presented. The paper uses the shadow features in the CIE LUV color space, establishes a single Gaussian shadow model, and achieves adaptive shadows detection. Tests show that algorithm of this chapter can remove the shadows of moving objects well.3. In view of the problem of partial occlusion in the tracking multiple objects. In real time, the anti-covered object tracking method based the Quick-SIFT feature is proposed. The method solves the problem of partial occlusion by matching Quick-SIFT feature points of moving target before and after occlusion. In the feature points extraction process, a new description of the key points of operator reduces the dimension of the key points and improves the SIFT feature point extraction speed.4. The paper puts forward the integration of multi-target tracking method—Mean Shift multi-target tracking method in researching the feature point tracking algorithm and Mean Shift algorithm. This method combines the advantages of Quick-SIFT feature anti-covered moving object tracking algorithm and Mean Shift algorithm, making the robustness of the tracking a moving object in video sequences more strong.