Soccer Ball Tracking using Dynamic Kalman Filter with Velocity Control

In this paper, we propose the ball tracking method that is tracking the ball adaptively and robustly in the soccer video. In the latest works, people have used the Typical Kalman Filter to track the ball. But when the ball is disappearing due to the occlusion with players, Typical Kalman Filter has no choice but to make a pool prediction and especially if the player take the ball for a long time, the error is produced much more. To overcome these problems, we propose the Dynamic Kalman Filter algorithm. Dynamic Kalman Filter robustly tracks a ball in the dynamic condition by using player information and reduces the error in the situation of occlusion by controlling the velocity of the state vector. The experimental results show that proposed Dynamic Kalman Filter shows better results than the Typical Kalman Filter and the Adaptive Kalman Filter that is proposed to overcome occlusion problem in the video sequence.