Distortion Correction And Image Rendering Of Ultra-large Field Of View Near-eye Display Device

Virtual reality(VR)display technology is currently a widely used three-dimensional display technology,whose typical application form is the head-mounted device(HMD).Immersion is an essential measure for HMDs,and a large field of view(FO V)is one dominant factor to improve it.Nowadays,it is a mainstream solution to achieve a large FOV through optical splicing,and the correction of this type of device has not been systematically discussed.Therefore,the correction research based on the dual optical path splicing HMD is undoubted of great significance.For a display device with a large FOV,the camera may not be able to obtain the full-frame distorted image in a single shot.This paper proposed a computational correction method to achieve distortion correction for those devices.The method uses homography transformation to transform sub-sampled images to the same coordinate system for stitching and fusion,then performs the sub-regional Bezier patch fitting,which adequately describes the complex non-linear distortion,on the stitched distorted image.Finally,the mapping table can be utilized for subsystem correction.Experiment results show that the method reduces the maximum line fitting error of the projection image from about fifty pixels to less than two pixels,effectively eliminating the distortion of the optical subsystem.For the splicing HMDs,this paper first proposed an image registration scheme to solve the problem of imaging misalignment at the joint of monocular subsystems.Based on the structure of the HMD,a corresponding rendering model is designed.The distortion mapping table and the FOV of perspective projection are modified through feedback adjustment to achieve the correct stitching of monocular images.Based on the unique application scenario of this paper,we also designed a fast and straightforward line detection algorithm and applied it to the feedback adjustment program to realize the quantitative evaluation of the misalignment.Experiment results show that the image registration method can effectively improve the image misalignment at the splicing position.The stitching accuracy around one pixel can be achieved in the central field of view,and a good stitching effect can be ensured when the user's line of sight deflects.