Robust depth camera based multi-user eye tracking for autostereoscopic displays

Abstract

Autostereoscopic display systems require effective and continuous eye tracking techniques to provide fluent stereo vision experience to the users. This paper introduces an emerging framework for 3-D range image sensor based eye tracking in human-machine interactions. A novel approach to multi-user eye tracking based on generic depth cameras for autostereoscopic displays is presented. Depth and color image sequences generated by a single depth camera are used as system input. Depth image sequence is utilized for multi-user tracking. An eye localization algorithm of our previous work, which combines appearance and feature based methods using SVM classification to gain robustness, is adopted in this approach with optimizations. The experimental results demonstrate that the proposed method satisfy the requirements regarding multi-user eye tracking for autostereoscopic displays. In addition, through the comparison with state-of-the-art object tracking techniques our method provides superior results.