Stereo analysis allows to reconstruct a dynamic 3D scene from two or more input videos that were taken from slightly displaced viewpoints. The computed 3D information (or depth map) forms the basis for a variety of applications such as novel view synthesis, virtual camera motion, or the combination of real with synthetic video content for augmented reality scenarios, which can greatly enhance the sports viewing experience. Our goal is to review the stereo processing pipeline and stereo matching algorithms from the particular perspective of sports applications. While the available literature focuses mostly on coarse depth reconstructions of sports scenes (for example, for extracting soccer players from the background), we aim at generating high-quality depth maps that reveal the subtleties of the athlete´s pose and movement. We give an overview of state-of-the-art computer vision algorithms for stereo analysis and discuss the special challenges that are posed to stereo matching by different types of sports scenarios. Examples of such challenges are fast and/or complex body movements, occlusions between team players, and unfavorable illumination or surface reflectance conditions. We show results that were obtained by applying state-of-the-art stereo matching algorithms and semi-automatic 2D-to-3D conversion techniques to sports scenes. Furthermore, we give suggestions for future stereo capture and processing systems that are tailored to the peculiarities of the sports domain and discuss potential applications.
M. Gelautz, F. Seitner, C. Kapeller, N. Brosch: "Stereo Reconstruction of Sports Scenes: Algorithms, Applications, and Challenges"; Talk: International Conference on Technology and Innovation in Sports, Health and Wellbeing, Vila Real, Portugal; 12-01-2016 - 12-03-2016.
Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.