Estimation of Ball Possession Statistics in Soccer Video
Saikat Sarkar, Amlan Chakrabarti, Dipti Prasad Mukherjee · 2018
In this work, we have estimated ball possession statistics from the video of a soccer match. The ball possession statistics is calculated based on the valid pass counts of two playing teams. We propose a player-ball interaction energy function to detect ball pass event. Based on position and velocity of the ball and players, a model for interaction energy is defined. The energy increases when the ball is closer and about to collide with a player. Lower energy denotes that the ball is freely moving and not near to any player. The interaction energy generates a binary state sequence which determines a valid pass or a miss-pass. We assess the performance of our model on publicly available soccer videos and have achieved close to 83% accuracy.