Detection method for driver's gaze direction based on improved VGG network
Haowen Wang, Qijie Zhao, Xianggang Wang, Weichi Zhao · 2023
Gaze detection is the basis for driver state monitoring and intent recognition as well as vision-based human-vehicle interaction. This paper proposed an appearance-based gaze direction vector detection model, taking the eye image, head posture and pupil center position as sample data and the gaze direction vector as sample labels, and using the improved VGG deep learning network for training to obtain the driver's gaze direction vector, and completing the gaze direction detection of the driver's gazing at different target regions. In the real car environment, we carried out experiments to recognize 6 different regions of the windshield, and the results show that the gaze direction detection method proposed in this paper is basically unaffected by changes in head posture and driving environment illumination, and the average detection accuracy for 10 drivers is 96.53%.