A semantic video object tracking algorithm using three-step boundary refinement

This paper describes a semi-automatic algorithm for semantic video object tracking. The proposed algorithm consists of three steps: object boundary projection, uncertain area extraction, and boundary refinement. The object boundary of the previous frame is projected on the current frame by using the object-based motion compensation. Then, uncertain area is determined via two modules: MC error-test and color similarity test. Finally, the exact object boundary of the current frame is obtained from extracted uncertain area by boundary refinement using color information of the current frame. The proposed video object extraction method provides efficient tracking results for various video sequences by using only two consecutive frames.