A comparison with covariance features on player uniform number recognition

Broadcast sports video report is expected be the generated player data in real time. Therefore, uniform number recognition need to fast computing. Covariance feature has possibilities to do it in real time. We present a method for uniform number recognition using the covariance feature as a region descriptor. The covariance feature is represented by the covariance matrix of image features such as spatial position, intensity, higher order derivatives, etc. Thus, the covariance feature vary by used image features. Feature matching is a simple nearest neighbor search under a distance measure. In the experiments, the six covariances compared. The recognition rate is 79% in a multiple player game.