People Recognition using Color and Texture Features: Application to Camera based Video Games

L. Hamoudi, J. Boonaert, and S. Lecoeuche (France)


Image sequence processing, color and texture, pattern recognition.


This paper proposes an appearance based approach for automatic player recognition and face and hands tracking in the context of camera based video games. The process uses background subtraction in two color spaces (RGB and YCbCr) and edge segment based approach to detect players. Then, the whole blob of a detected player is divided to three sub-blobs (top, middle and bottom). A feature vector is calculated for the whole blob and for each of the sub-blobs. The features are a collection of measures and histograms of different color and texture spaces. A Gaussian Mixture Model is used to model each player appearance in the training phase, and for classification of players at runtime. Then, a voting scheme is carried out for a final decision. The face and hands tracking algorithm is based on skin detection in "THS" space (Texture, Hue & Saturation).

Important Links:

Go Back