Estimating the Number of Soccer Players Using Simulation-Based Occlusion Handling

Noor Ul Huda, Kasper H. Jensen, Rikke Gade, Thomas B. Moeslund; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 1824-1833

Estimating the number of soccer players is crucial information for occupancy analysis and other monitoring activities in sports analysis. It depends on player detection in the field that should be independent of the environment and light conditions. Thermal cameras are therefore a better option over normal RGB cameras. Detection of non-occluded players is doable but precise estimation of number of the players in groups is hard to achieve. Here we propose a novel method for estimating number of the players in groups using computer graphics and virtual simulations. Occlusion conditions are first classified by using distinctive set of features trained by a bagged tree classifier. Estimation of the number of players is then performed by maximum likelihood of probability density based approach to further classify the occluded players. The results show that the implemented strategy is capable of providing precise results even during occlusion conditions.