Discrete Particle Swarm Optimization for Player Trajectory Extraction in Soccer Broadcast Videos

Tremendous broadcast soccer videos demand automatic semantic representation for event and tactic analysis. Player tracking is an important step which can be further processed and analyzed by sport experts to evaluate player or team performance. A novel scheme for player tracking in soccer broadcast videos is proposed in this research. Following player detection using Adaboost, player labeling, occlusion handling and mosaic construction, player tracking is formulated as an optimization problem allowing us to extract player trajectories using Particle Swarm Optimization (PSO). PSO is an optimization method inspired by the flocking behavior of birds which was originally customized for continuous function value optimization. In this paper, a new application of discrete particle swarm optimization for player tracking in soccer videos is proposed. Updating equations for particle swarm optimization algorithm are modified based on problem characteristics and discrete variables to extract player trajectory. Experimental results show that the modified PSO is promising in solving soccer player tracking problem.