Automatic video sequence segmentation using object tracking

Emerging video standards address content-based functionalities and interactive multimedia. MPEG-4 segments the frames into video object planes (VOP) with each VOP corresponding to one semantically meaningful object. In this paper a new automatic video sequence segmentation algorithm is presented that can extract such VOPs. A binary model for moving objects is automatically derived and tracked in subsequent frames using the generalised Hausdorff distance. The model is updated every frame to accommodate for rotation and changes in shape. For that an update mechanism consisting of two components is proposed. To improve the segmentation and reduce the computational complexity the stationary background is filtered by a novel technique. Experimental results demonstrate the performance of our proposed algorithm.