Abstract

Full-body human performance capture has been extensively explored in constrained environments, but less attention has been applied to the task of performance capture in sports scenes. Sports datasets provide a wealth of challenges: player contact and occlusion; fast motion; and low resolution and wide-baseline cameras. We present a method that uses multi-view pose to disambiguate players and overcome some of these issues. The applications are broad, including player performance analysis, sports broadcast and immersive experiences. This paper leverages previous work on multi-person 3D pose estimation and tracking in sports, and model-based human shape and pose estimation. These techniques are combined to produce an estimate of the body shape and pose for all players in a sports scene. We demonstrate results on a soccer dataset comprising over 20 subjects.

Paper

Full-body Performance Capture of Sports from Multi-view Video (Short Paper)
Lewis Bridgeman, Jean-Yves Guillemaut and Adrian Hilton
The 16th ACM SIGGRAPH European Conference on Visual Media Production (CVMP 2019)



Acknowledgments

This research was supported by EPSRC Grant (EP/N50977/1).