ScholarWorks@성균관대학교

초록

This paper introduces a real-time head-pose detection and eye-gaze estimation system for Automatic Driver Assistance Technology (ADAT) aimed at enhancing driver safety by accurately collecting and transmitting data on the driver's head position and eye gaze to mitigate potential risks. Existing methods are constrained by significant limitations, including reduced accuracy under challenging conditions such as varying head orientations and lighting, higher latency in real-time applications (e.g., Faster-RCNN and TPH-YOLOv5), and computational inefficiency, which hinders their deployment in resource-constrained environments. To address these challenges, we propose a novel framework using the Transformer Detection of Gaze Head - YOLOv7 (TDGH-YOLOv7) object detector. The key contributions of this work include the development of a reference image dataset encompassing diverse vertical and horizontal gaze positions alongside the implementation of an optimized detection system that achieves state-of-the-art performance in terms of accuracy and latency. The proposed system achieves superior precision, with a weighted accuracy of 95.02% and Root Mean Square Errors of 2.23 and 1.68 for vertical and horizontal gaze estimation, respectively, validated on the MPII-Gaze and DG-Unicamp datasets. A comprehensive comparative analysis with existing models, such as CNN, SSD, Faster-RCNN, and YOLOv8, underscores the robustness and efficiency of the proposed approach. Finally, the implications of these findings are discussed, and potential avenues for future research are outlined.

키워드