Knowing an opponent’s strengths (and weaknesses) can make the difference between winning and losing in sport, but obtaining this knowledge is typically very expensive and time-consuming. In combat sports, performance analysts are still required to watch and code video footage of opponents manually in order to identify the actions performed by each player. Given the time and expenses necessary to do so, a system is desirable that automatically detects and categorises the athletes’ actions.
The aim of this project is to develop automated performance analysis software for use in combat sports (e.g., judo, karate, boxing, taekwondo) using Microsoft Kinect sensors. These sensors are not presently able to distinguish two opponents who are in contact with each other (as is the case in some combat sport), and so in this project we will develop algorithms to (1) distinguish two opponents on the basis of footage from Kinect sensors, and (2) categorise particular actions performed by the two opponents.
Method and design
The project requires Microsoft Kinect sensors to record three-dimensional depth images of judo bouts. Computers with high-end graphic cards will be used to process the images by identifying and distinguishing the opponents, and then will use algorithms designed to identify basic actions.