Offline Swimmer Cap Tracking Using Trajectory Interpolation

In this paper, we present a preliminary attempt to solve the difficult problem of tracking swimmer cap in swimming videos to facilitate swimmer performance assessment. Due to the great challenges posed by moving camera and severe figure-background occlusions, an offline approach based on trajectory interpolation is adopted. Firstly, each frame is searched for hypothesized positions of the target cap using mean shift mode seeking. Secondly, most outliers due to ambiguities and noise are eliminated using lane constraints, and the hypothesis in the space-time volume are clustered into trajectory segments based on a spatial and temporal closeness criteria. Finally, cubic spline trajectory interpolation is used to infer the target cap position in occluded frames. Experiments show that satisfying tracking results are achieved by our approach.