A Robust Eye Localization System for Autostereoscopic Display Using a Multiple Camera

In this paper, a multiple camera eye localization for stereoscopic display is proposed. The multi-camera system contains an infrared camera and a binocular camera. Candidate face area for low-resolution infrared image is obtained by using YOLO-V3 network. Then face distance is calculated by an embedded ARM platform Raspberry Pi-3B+, which provides a priori knowledge for the face detection of the high-resolution visible light camera. AdaBoost and LBF algorithm are used for complete face detection and alignment, and finally they achieve the goal of eye location. Sliding window stereo matching method can improve the efficiency of stereo matching. The experiment results verify the feasibility of the proposed scheme. The speed, accuracy, and robustness of eye tracking are excellent.