Bilkent University
Department of Computer Engineering
S E M I N A R
Multiple Player Tracking for Soccer Video Analysis
Sermetcan Baysal
Accurate tracking of multiple players in real-time is the key issue for soccer performance evaluation systems and it requires detecting players on the video, finding their positions at regular intervals and linking the spatiotemporal data in order to extract the trajectories. It is common to encapsulate the descriptive information of a soccer match (such as position, velocity, and appearance of the players) into states at each time frame and to model the game as a collection of temporal states. Then, multiple player tracking problem can be perceived as a stochastic process where the objective is to estimate the state of the game based on the previous observations. However, using a joint representation of the target space and unified observation model for all players results in a huge state-space which is intractable. In contrast, decoupling states of the players and employing a separate tracker for each target fails to grasp the global state of the game and relation among the players resulting in well-known problem of identity hijackings. As a solution, we propose a new methodology for accurately tracking multiple soccer players in which relative efficiency of employing separate probabilistic trackers is combined with effectiveness of joint-state models. Furthermore, we describe the complete steps of a real-time soccer player tracking system that has been successfully tracked players covering a total distance of 45000 km in near 200 professional soccer league matches.
DATE: 10 November, 2014, Monday @ 15:40
PLACE: EA-409