Project: Camera-driven Live Animation and Statistics for Sports

The purpose of this project is to develop a solution capable of automatically extracting the skeleton poses and motions of athletes during sport events. The solution shall be completely non-intrusive, that is it shall not rely on special equipment carried by the athletes. Instead it will run solely off camera footage of the athletes. Parts of the solution shall be real-time and capable of analyzing the motion of up to 30 simultaneous athletes._x000D__x000D_The above is a very ambitious target for a project this size and virtually impossible without a solid platform_x000D_to build from. Part of the underlying technology here is the Tracab Image Tracking System™ and all the_x000D_experience gained when developing and deploying it. This system uses image processing technology to identify the position of all moving objects, that is, players, referee, and ball, in a sports arena in real-time at 25Hz. The precision of the measurements is below one decimeter. The first commercial version of the system was released in 2005 and, hardware-wise, the 4th generation is to be released in the summer of 2013. Customers include broadcasters, websites, and leagues or their data Ps._x000D__x000D_The goal of this project is thus to significantly enhance Tracab’s data content by capturing as automatically as possible not only the players’ 2D ground trajectories but also the positions of body parts such as arms,legs, torso and head. This next generation of tracking data will enable a range of new products based on animations and statistics. It will make it possible to generate computer-game style 3D animations that can be used both for second screen broadcasting and coaching purposes, while avoiding the costly and tedious manual editing that commercially available systems require. For broadcasters, editorial value can be added by allowing a virtual camera to be placed anywhere on the field during highlights or expert analysis. Coaches can gain insight into team performance._x000D__x000D_Achieving these goals requires a solid scientific approach and well-engineered solutions. A key part of the_x000D_project will be to exploit the ground plane trajectories provided by the Tracab Image Tracking System™ since_x000D_this information can be used to constrain the problem. Furthermore, in accordance with the current state-_x000D_of-the-art, we will draw both from methods that detect specific postures in individual frames and those that_x000D_track the motion from frame to frame given an initial pose. Neither technique is clearly superior to the other,_x000D_and both are actively investigated. In previous work by consortium members, the two were combined. Also_x000D_in our previous work we exploited the fact that human motions often contain characteristic postures that are relatively easy to detect. Given two consecutive such postures, modeling intermediate poses becomes an_x000D_interpolation problem, which is much easier to solve reliably than open-ended tracking. Encouragingly our_x000D_recent research shows that, given multiple views of a footballer in a game at any given time and sufficient_x000D_training data, it is possible to automatically estimate a wide range of 3D characteristic poses._x000D__x000D_While the approaches discussed above increase robustness they are not yet at the point where they could_x000D_be deployed in a large-scale real-world setting. To bridge this gap and bring such solutions to market, the_x000D_proposed project will rely on the expertise and capabilities of all the Ps while exploiting Tracab’s market position to provide access to suitable data, customers, and production workflows.

Acronym CLASS (Reference Number: 8661)
Duration 01/03/2014 - 28/02/2017
Project Topic The goal of the CLASS project is to extend the state-of-the-art in camera-based player tracking in arena sports by estimating the full 3D pose of the subjects’ body parts. The project will generate new products and services based on statistics and free-viewpoint computer-game style animations.
Network Eurostars
Call Eurostars Cut-Off 10

Project partner

Number Name Role Country
2 ChyronHego AB Coordinator Sweden
2 ECOLE POLYTECHNIQUE FEDERALE DE LAUSANNE Partner Switzerland