Online Multiplayer Tracking by Extracting Temporal Contexts with Transformer

Sports competition is one of the most popular programs for many audiences. Tracking players in sports game videos from broadcasts a nontrivial challenge computer vision researchers. In videos, direction an athlete’s movement changes quickly and unpredictably. Mutual occlusion between athletes also more frequent team competitions. However, rich temporal contexts among adjacent frames have been excluded consideration. To address this dilemma, we propose online transformer-based learnable framework end-to-end fashion. We use transformer architecture to extract successive add them network training, which robust complex multiplayer tracking. demonstrate effectiveness our method on three video datasets by comparing with recently advanced trackers.