Baseball hand tracking from monocular video

There has been a consistent interest in tracking players and their actions in a sports video among the computer vision community. Further, tracking specific body part of an athlete can give insight for expert analysis, graphics animators and game developers. This can be a laborious task as athletes tend to move at high and varying speeds causing motion blur. Moreover, there is no definite equation depicting their motion making it hard to develop a predictor. Hence in this paper, a technique to track the pitching hand of a baseball player is presented. Taking cues from the little information available from the predictor, the possible SURF feature points that might have the pitcher's hand in the vicinity are found. A simple skin detection can confirm the presence of the hand near the SURF keypoint. A curve is fitted on the detection over the length of a pitching video. The proposed technique overcomes speed variations and blurry features to detect the pitching hand of the baseball player.