3D Shot course estimation system for tactics analysis support of handball games

In this paper, we propose a 3D shoot course estimation method for tactics analysis of handball games using the video recordings captured by two interlaced video cameras. To achieve the accurate estimation of the 3D position of the ball, an algorithm for temporal and geometrical calibration is employed to calibrate two cameras because the speed of the ball is high and the sub-frame calibration is important for our system. To estimate the 3D position of the ball, it is important to detect the 2D positions of the ball in respective frames captured by each camera. However, many false detections occur because of occlusions caused by frequent hard contact play and highspeed moving ball thrown by a player to the goal. Therefore, we eliminate the false detections based on the consistency of the ball positions through multiple frames using the fact that the trajectory of the ball is straight in almost all of the case.