3D Gaze Estimation via Binocular Disparity and Feature Refinement
Chengxin Wang, Qiuxia Chen, Ying Zhong Tian, Jinfeng Yang · 2024
The eyes are the windows to the soul and are crucial for studying human behavior. Therefore, gaze estimation has attracted much attention in the field of computer vision. In recent years, convolutional neural networks have driven significant progress in gaze estimation research. Despite the remarkable achievements of gaze technology based on full-face images, the inability to obtain full-face images in specific scenarios makes research based on eye images particularly important. This paper proposes a three-dimensional gaze estimation model that takes binocular images as input, using VGG16 as the backbone network for feature extraction. To enhance the feature extraction capability, we have embedded channel attention and spatial attention mechanisms. Meanwhile, to improve the stability of gaze estimation, we classify real gaze angles. First, we perform a preliminary screening of gaze direction through a classification task, and then perform linear regression with the true angle. Finally, we optimize gaze features using binocular vision differences to finetune the gaze estimator. Experiments have shown that our method performs well on three public datasets.