Semi-automatic Multiple Player Tracking of Soccer Games using Laser Range Finders
Yuma Kabeya, Fumiharu Tomiyasu, Kenji Mase · 2016
The diversity of sports broadcasting formats increases fans' interest in watching games. Soccer video is one such example ---it may include data analyses of player performances, highlighting of target players, and other useful features. Such video also supports player coaching and training. Numerous studies have attempted to estimate exact trajectories using camera information. However, camera information is not robust in situations with varying brightness levels, complicated backgrounds, and so on. To address these issues, we propose a soccer player tracking method using laser range finders. Our proposed method can precisely acquire the positional information of players. Over 95 % of the temporal connections can be easily established with a graph construction method. In order to further reduce manual connecting operations, we adopt automatic corrections using a particle filter. Experimental results show significant improvements in tracking results, which can reduce manual operational costs.