A Feedback Control, Gaze Following Approach for Human-Robot Interaction

Abstract

Gaze following is one of the key requirements for successful Human-Robot Interaction (HRI). In this paper, a feedback controlled approach for facial features detection and gaze following is proposed. The goal of the method is to cope with uncertainties present in the context of HRI, namely different poses, occlusion and variable illumination conditions. The key elements of the method are the local and spatial estimators of the facial features, the Gaussian Mixture Model (GMM) used for segmenting the face, as well as the feedback control way in which the parameters of the whole processing chain are adapted. The system has been evaluated against temporal sequences of moving human agents, acquired via a stereo imaging system mounted on a mobile robotic platform. As performance metrics, the mean and the maximal normalized deviations between the manually determined ground truth and the estimated positions of the facial features have been chosen.

Keywords

• gaze estimation
• feedback control in image processing
• facial features detection
• Human-Robot Interaction