Fully automatic pitching location analysis system using high-speed cameras

Abstract

In this paper, we propose a ball image saving framework for an automatic baseball pitching location analysis system using high-speed cameras. In the existing system, it is difficult to manually select the frames in which the ball is included from the experimental pitching videos. The proposed automatic framework extracts ball candidate regions by differencing three consecutive frames from the acquired sequence in the buffer. The ball candidate regions are identified as a ball by comparing circularity with a pre-defined threshold value. If the ball exists among the ball candidate regions, the current sequence is saved to storage. Since the system utilizes high-speed cameras, we apply a ball detection method that has low computational cost. Experiments on many images captured by the proposed system demonstrate that the proposed framework and ball detection method can automate the pitching location analysis system.