Appearance-based multiple hypothesis tracking

Soccer is a popular sport in the world with the growth of demand for automatically analyzing matches and tactics. Since players are the focus of attention in soccer videos and they manage the entire game, player tracking is fundamental to most soccer video analysis. An efficient implementation of the multiple hypothesis tracking algorithm by evaluating its usefulness in the context of soccer player tracking is introduced in this paper. In contrast to the inherent linear assumption of multiple hypothesis tracking (MHT), which ignores appearance cues and occlusions, our approach relies on an appearance-based MHT (AMHT) framework by incorporating particle swarm optimization (PSO) to account for appearances, nonlinear movements and occlusions. Experimental results demonstrate the efficiency and robustness of the algorithm. An effective AMHT framework is introduced to account for players appearance, nonlinear movements and occlusions.A blob-guided PSO player detection is introduced into the MHT, which is capable of detecting and labeling multiple players and resolving partial occlusions simultaneously. Only teammates in occluding blobs are detected by erasing the previously detected ones.Player labeling is solved within the player tracking step.