It is also possible to avoid some of the problems in tracking multiple skin blobs by using colored gloves. Some of these problems included confusion when blobs would severely occlude or pass by each other and confuse the correspondence. Thus, colored gloves allows the system to be more robust to such occlusions as well as other various environments. Therefore, the vision system can be switched into this optional operation mode whenever skin tracking is unreliable. Instead of using only a probabilistic density model of skin color, each user wears two differently colored gloves and thus a unique probabilistic color model is used for each object. For instance, the face is tracked using the default skin color class, the left hand employs a blue glove color class and the right hand uses a red glove color class. This multi-color object tracking is a more reliable mode of operation and is used to initially train the system. It provides cleaner data and should result in much faster convergence of the behavioural learning. Subsequently, when the system has started learning, and has some limited behavioural model, it is possible to remove the gloves on the assumption that the behavioural learning has acquired some understanding of the role of the three blobs as distinct entities (left hand, right hand and head). Otherwise, the correspondence between blobs might not be resolved properly.