Application of motion based object detection to real-time 3D soccer ball tracking

S. Larndorfer. Application of motion based object detection to real-time 3D soccer ball tracking. 2005.

  • Stefan Larndorfer
Abstract The impartial analysis of soccer games has a wide variety of potential target groups: Spectators, referees and soccer coaches can benefit from accurate three-dimensional position data of the soccer players and the soccer ball. So far, the realization of a tracking system that is suitable for such a high-demanding environment has turned out to be very tricky. Current electronic tracking devices that have to be integrated into the sports equipment are not yet small enough, are not accepted by professional soccer players and are disapproved by the FIFA. On the other hand, current optical tracking systems that do not have these shortcomings suffer from a high unreliability of the computed position and are mostly not designed to operate in real-time. Based on a detailed evaluation of current tracking approaches, this thesis proposes an optical soccer ball tracking system that can compute three-dimensional coordinates of the soccer ball in real-time. Multiple video images from several different views are processed simultaneously in several distributed image processing units that use efficient image processing algorithms and fuzzy rules to locate the soccer ball in the image. The proposed real-time image analysis that utilizes methods like background subtraction, blob analysis and fuzzy logic is subject of a detailed discussion. Furthermore, the hardware and software architecture of the implemented prototype system is presented. The proposed implementation contains solutions for coping with the enormous amount of image data and shows the imperative necessity of a test-driven development process. Finally, test results of the implemented system from a variety of different test cases are evaluated. iii