Semi-automatic video segmentation for object tracking

This paper describes an algorithm of semi-automatic video segmentation for object tracking. In this semi-automatic method, a user specifies an object of interest in the first frame; then the specified object is to be tracked in the remaining frames. The proposed tracking algorithm consists of two steps, i.e., object prediction and boundary refinement. For object prediction, we propose two novel object prediction methods based on motion estimation and color, respectively. By using these two object prediction schemes, pixels are assigned either to the object or background. Then, if some pixels are judged differently by these two schemes, they are considered as uncertain areas. By refining these uncertain areas, we can obtain object boundaries closer to the real ones compared to conventional algorithms. Simulation results show that the proposed method provides efficient tracking results for various video sequences.