Automatic Soccer Video Analysis

Current approaches for (semi-)automatic soccer analysis rely on special cameras or sensors to perform analysis such as player detection and player tracking. Other types of metadata – e.g. highlight annotation, player identification must be collected manually. We perform our automatic analysis on broadcast content without relying on dedicated cameras or other specific sensors. By analyzing broadcast content, we have to deal with additional challenges. Content from moving cameras usually exhibits motion blur that makes automatic analysis more challenging, as well as possible compression artifacts and interlaced content.

We are developing solutions for automatic collection of the following metadata:

  • Identification of the active camera view (overview, medium, close-up or out-of-field)
  • Replay detection
  • Visualization of the camera visible part of the play field
  • Player detection and tracking
  • Assignment of player to their teams
  • Localization of player position on actual play field
  • Detection of sport specific highlights

This technology can be useful for:

  • Broadcasters and other content holders for fast search and retrieval as well as navigation in sports video archives
  • Media asset management companies to extend their range of applications and include automatic sports video analysis