Research And Implementation Of 3D Reconstruction And Tracking Registration In Augmented Reality

In recent years,augmented reality has been one of the popular use cases in computer vision technology for the general public,while the application of 3D reconstruction has an important significance in augmented reality.The effect of 3D reconstruction determines the actual effect of augmented reality technology in the application.Tracking and registration technology is the core technology in augmented reality.Although tracking and registration technology has been developed for a long time and gradually matured,there is still a lot of room for improvement from the perspective of pursuing perfection.In this paper,we mainly focus on feature extraction and matching,3D reconstruction of point cloud data and tracking and registration technology in computer vision.The specific work is as follows.(1)Improvement of ORB algorithm.The traditional feature extraction algorithm with scale invariance cannot well meet the real-time requirements of augmented reality systems.Although the ORB algorithm with better real-time performance can obtain scale invariance by constructing Gaussian pyramids,it causes partial loss of details and blurred edges when performing Gaussian filtering,so a nonlinear multiscale space is introduced to obtain more feature points on edges and textures.(2)The improved ORB algorithm is applied to the 3D reconstruction.Although the traditional multi-view-based 3D reconstruction has a better performance,the real-time and accuracy of 3D reconstruction are more demanding in recent years in application scenarios such as augmented reality,mixed reality and autonomous driving.Therefore,by introducing the improved ORB algorithm for feature extraction and matching work in 3D reconstruction,the speed and quality of 3D reconstruction are significantly improved.(3)The tracking registration algorithm is improved.In the augmented reality system feature matching of each frame with its previous frame for the whole image will incur great time cost,in order to improve the tracking registration speed choose to track the fixed target in the image and only feature matching of the fixed target region in the upper and lower frames to reduce the computing time.Therefore,this paper adopts the improved TLD-based tracking algorithm to achieve the tracking of the target in the field of view,and then achieves feature extraction and matching of the tracking region by the improved ORB algorithm to complete the acquisition of the camera pose,after which the virtual object is registered to reality.