A trajectory-based ball detection and tracking system with applications to shot-type identification in volleyball videos

Abstract

This paper presents a trajectory-based ball detection and tracking algorithm for volleyball videos to classify the different types of shots in the game. In case of volleyball games, the task of ball detection and tracking become more complex due to the presence of too many number of players in a rather small space. The direct detection methods often fail due to the high rate of occlusion of the ball image with players. The distortion of the ball image due to the effect of ball and camera motion leads to a number of wrong detection. In this paper, the trajectory information of a volleyball for different shot sequences is studied and used to estimate the ball locations. The ball candidates are generated using the shape and size features. A Kalman filter-based approach is used to generate a set of candidate trajectories. The actual ball trajectory is extracted by analysing the physical characteristics of the ball motion. The trajectory informations are then used to classify the different shot sequences in the volleyball game for better representation and analysis.