Combining Photometric Features and Relative Position to Detect and Track Target Person

Tracking a target person is a vital job in human-robot interaction. The robot must always notice a particular person as the interaction target partner. However, it is sometimes very hard to distinguish the target person because there are many other persons around the target. In this paper, we propose a target person detection and tracking system by combining person’s frontal photometric features such as face and clothing color, and coordinate of the person’s location in the real world. We apply an illumination invariant face recognition method named OptiFuzz. Hue-Saturation histogram (HS-histogram) is used to obtain the clothing color feature, and a location of the person is acquired from a calibrated single camera view. All these features are then fed into an algorithm of Naive Bayes to discriminate between the target person and others. Our experimental results indicate a successful outcome as it is always possible to detect and track the target person.