Single-Camera Gaze Estimation Based on ShuffleNetV2
Yixin Guo, Liankui Qiu, Jingpei Zhou, Xuyang Shi · 2023
The estimation of the user's gaze is useful in many aspects and future development trend is single-camera gaze estimation, which is conducive to the application of gaze estimation technology on mobile devices. At present, the accuracy and rapidity of head pose estimation are necessary for single-camera gaze estimation. Therefore, a ShuffleNetV2 based head pose estimation is proposed in this paper and verified in data sets of 300W-LP and BIWI. The result shows that the proposed algorithm is superior to the current advanced algorithms. Then the head pose and the simplified eye model are combined to ensure the accuracy of gaze estimation and the robustness of the user's head movement. The result shows that the average accuracy of gaze estimation can reach 5°under the extreme state of head motion.