A Real-Time Court Detection and Tracking System to Tennis Videos

To analyze a tennis video sequence at a higher semantic level, the first step is to determine the position of the court. Thus, it is required to track the court in the scene. In this article, we proposed a new automatic system for tennis court detection and tracking in real time. Our system comprises a structural analysis system that extracted the court view images to be processed by classifying the tennis video to non-court view or court view frame. Then, the initialization step is dedicated to detecting the court in the first frame. It consists of extracting the tennis court line using the Probabilistic Hough Transform and estimating the homography between the tennis court model and the detected court lines. To ensure the accuracy of the obtained homography, it should pass a rejection test. We used the camera motion as a supplementary information, and we tracked the court only in case where the camera is moving, which improves the stability of our system. In addition, for each court view frame, we updated the homography, and we fitted the projected model based on the algorithm of Levenberg–Marquardt.