Knowing an opponent’s strengths and weaknesses is crucial for success in combat sports. Yet, the manual performance analysis required to analyse opponents is very time consuming and expensive. The aim of this project was to develop prototype performance analysis software that is capable of automatically detecting and categorising basic actions performed in a combat sport.
Indeed, we have successfully created this prototype by developing software that analyses footage from commercially available Microsoft Kinect 3D motion cameras. Previously these sensors were not able to distinguish two combatants when entangled within a fight. We have developed prototype software that can disentangle the actions of two people who are in contact with each other (e.g., two judokas in combat) on the basis of depth and colour information from the Kinect camera. Using this disentangled information, we can now automatically detect basic prioritised actions within a fight, providing the capacity to detect a range of more subtle actions in the future.
Krabben, K., Mann, D.L., Coolen, B., Ravensbergen, H.J.C., Smit, R. & van der Kamp, J. (2016). Disentangling duelling judoka: Developing a system to automate performance analysis in combat sports (final report). Vrije Universiteit: Amsterdam. [on request]