A robust and fast model-based athlete contour tracking in diving videos

This paper presents a novel tracking scheme to detect the athlete contour in diving videos, which is an important step in on-line motion analysis for diving training. This scheme is mainly based on the active contour model in linear shape space. First, an initial template is built on the first frame of the diving video and is updated by a curve-fitting algorithm. Then in each following frame, the template transforms and adjusts itself to track the athlete contour, according to the information from both prior model knowledge and current frame analysis. Meanwhile, the changes of template between frames are detected and recorded to improve the efficiency of the tracking process in the following frames. Finally, the athlete contour is tracked out frame by frame and experiments demonstrate that the proposed scheme is both efficient and robust for the athlete contour tracking in diving videos.