基于CB-ViT的青少年视线估计算法研究OA北大核心CSTPCD

Gaze estimation technology is widely applied in the fields such as human-computer interaction(HCI),virtual reality,and medical diagnostic assistance.However,the existing public datasets are primarily adult-oriented,so the gaze estimation algorithms trained on these datasets show suboptimal performance when applied to adolescents.To address this issue,a youth-specific gaze dataset named ″Young-Gaze″,which encompasses gaze data from 107 adolescents,is collected.In addition,a novel 2D gaze estimation algorithm is proposed.This algorithm is on the basis of ViT(vision transformer)and incorporates a context broadcasting(CB)module,which significantly enhances the feature representation capability of the network model by integrating both eyes' features at different levels.Experimentally,this algorithm demonstrates superior performance on the dataset Young-Gaze.Its error is kept within 5.42 cm,so it surpasses the other existing 2D gaze estimation methods.Besides its notable performance on Young-Gaze,it also shows good results when trained and tested on the public 2D gaze datasets GazeCapture and MPIIFaceGaze.The above facts indicate that the proposed algorithm is not only suitable for the adolescent,but also applicable for the adults effectively.