| Virtual reality and augmented reality are hot cutting-edge technologies all over the world.They are widely used in medical research,culture and entertainment,life services and other fields.Object detection in VR/AR field can effectively detect the position,category and other information of surrounding objects without manual identification,which greatly reduces the workload of manual analysis and sorting.However,the undifferentiated complete detection through the existing object detection algorithms will produce a lot of edge information and irrelevant objects,which makes the detection results chaotic and redundant,and can’t clearly get the object information concerned by users.It will be very difficult to analyze the psychological change process of users,the direction of interest and personalized recommendation.Based on this,this paper creatively combines eye tracking technology with object detection technology,and proposes two object detection methods based on eye tracking for VR/AR field.This paper studies the following aspects:(1)In this paper,eye tracking object detection method based on foveal visual region mapping is created to detect the target in the fixation area of human eyes.Through the establishment of two channels to process the human eye image and the foreground image.The creation of the mapping model is to determine the size of the visual area of the fovea by using the visual characteristics of the human eye,and combines the gaze tracking equations in the horizontal and vertical directions.Then,the data information of the two channels are matched with each other,so that it can be determined that the line of vision area of the human eye in the VR/AR scene corresponds to the position relationship and object information in the foreground image.The optical flow method and perceptual hash algorithm are used as the basis for object detection and pupil tracking,which greatly improves the speed of the model.(2)When using the multi-step and multi module method to complete the object detection in the fixation area,it is easy to cause the accumulation of errors.Therefore,this paper proposes an end-to-end model,which realizes the object detection function based on eye tracking.The model improves the feature fusion module on the basis of yolov4 network,establishes a new CSPANet network.Through a loss function,the model can detect the objects in the line of vision area.The experimental results show that the average angle error of the model is as low as 0.73 °,the rate increases by 30.77%,and the superiority of user experience is greatly enhanced.(3)This paper designs an object detection system based on eye tracking,which can be deployed in VR/AR equipment. |
