Research On Algorithm Of Extracting Moving Object Contour For Visual Tracking

Visual tracking has already been one of the research focuses in robotics, and computer vision. Among traditional visual tracking strategies, region based ones especially such as Mean Shift Algorithm, are characteristic of their outstanding real-time feature, easy integration property, and excellent robustness to occlusion. However, the algorithm is only sensitive to finding the region in which the moving target object is, while the informations with which the object itself is recognized are lost. It leads to a heavy restrict to intelligence degree in visual tracking. Contour implies most of the important informations of an object. And through one of active contour models, especially GVF Snake model, one can effectively extract the contour of the object. But it is sensitive to its initial position. That's to say, the initial position of the contour must, in general, be close to the true boundary or else it will likely converge to a wrong result.In this dissertation, contour of a moving object will be extracted base on a new algorithm which combines Mean Shift Algorithm and the GVF Snake model. Firstly, a region of the target is robustly extracted in real time based on Mean Shift Algorithm, which greatly reduce the searching region for an exact contour. And then, the extracted target region oriented image preprocessing is implemented. Finally, the contour of the moving object will be extracted based on GVF Snake model. The experimental results show that the proposed contour algorithm is effective.