Team Possession Analysis for Broadcast Soccer Video Based on Ball Trajectory

摘要

This paper presents a framework to automatically analyze the team possession for broadcast soccer video. Our framework requires the position of the ball and the play/break structure information as input. This problem is important in the cases of coaches and viewers evaluating team performance, and in the border context of video content understanding. The local minima of ball velocity curve over frames are a good start for finding possession transition points. The local minima and maxima in X and Y curves of the ball position, the ball and player locations, and acceleration starting points of the ball are used to infer the touching points, which are potential ball possession transition points. A Support Vector Machine is employed to find out which team touches the ball for each touching point. Experimental results on two consecutive 1000-frame sequences illustrate that the proposed framework is promising.