A novel computer vision technique used on sport video

Abstract

A method based on computer vision technologies is presented to achieve the function that the simulated motion in sport simulation system and the motion in sport video are presented on the same screen and at the same view point. The proposed method first applies the camera self-calibration theory to obtaining camera intrinsic parameters in sport video according to the 2D video features correspondence. Next it makes use of the feature 3D reconstruction to get a feasible estimation of extrinsic parameters. The extracted camera parameters information is applied to sport simulation and training system to achieve the function that the simulated normal motion of 3D virtual athlete in sport training system and the athlete motion in sport video are presented on the same screen and at the same view point. So we can quickly and accurately find the difference between the athlete motion in sport video and the simulated motion. It is very helpful to coaches and the training of athletes.