Head orientation estimation for covert-tracking robot

Covert-tracking robot refers to the mobile robot that not only can follow a human objective, but at the same time can control itself to keep away from the human´s visual field. This paper proposes a head orientation estimation method based on probability model, which can help the robot to implement its covert behaviors. First, the elliptical head contour is tracked out by using a method based on quadrant arcs; then it is normalized into predefined size and is partitioned into 24 sub-areas. According to skin color model, an orientation probability model is built for each discrete angle. The final estimation is obtained by weighting each discrete angle, where the weight is calculated out by matching the current input with the model corresponding to the discrete angle. Experiment results confirm the method´s good performance, strong robustness to different distance and high real time property.