This paper presents a technique for integrating heterogeneous information for tracking soccer players. First, our proposed method observes the texture, color, and shape of the target to detect its position, size and/or velocity. Then, Kalman filters are used to adaptively fuse the detection results depending on the occlusion status and on the reliability of the detection. Experiments with soccer sequences validate the technique's robustness to occlusion and deformation. Its applications to broadcasting services are also briefly discussed.
展开▼