Research On 3D Registration Algorithm Based On Fusion Of Marker And SLAM Localization In AR System

In recent years,with the rise of the concept of Metaverse,augmented reality has become a research hotspot again.As the core technology of augmented reality,3D registration is responsible for the tracking and positioning of real scenes,and plays a decisive role in image synthesis and dynamic interaction of virtual and real scenes.With the rapid development of society,it is urgent to introduce more mature and accurate augmented reality application systems in the fields of industry and medicine.In the process of building 3D registration algorithm,there are several core problems to be solved.The first is to improve the accuracy of the algorithm so as to ensure the authenticity of the composite images and the operability of interaction,the second is to improve the speed of the algorithm and reduce the delay so as to ensure the real-time performance of the system,and the third is to improve the robustness of the algorithm so as to ensure the stable operation of the system.Based on the above three aspects,this paper focuses on the current popular algorithm including localization based on markers and based on SLAM.By analyzing their advantages and disadvantages,this paper proposes a 3D registration algorithm based on the fusion of the above two localization algorithm.The primary works of the paper are summarized as follows.(1)In order to solve the problem of uneven and unstable illumination in the real scene,the homomorphic filtering method is used to adjust the brightness of the real scene image,and the image is enhanced with the restricted contrast adaptive histogram equalization.By using the pre-processed image to match the ORB feature points,more matching points can be obtained stably.Through the comparative experiments on KITTI and Eu Ro C datasets,it is found that the optimized image can significantly improve the accuracy of ORB＿SLAM3 tracking.(2)This paper proposes an algorithm to optimize the ORB＿SLAM3 initialization module by using the markers,which solves the problem of scale uncertainty in the initialization process of single SLAM.On the one hand,using the position and pose information of an marker can speed up the initialization of ORB＿SLAM3 system.On the other hand,based on the physical information of the known markers,the positionattitude transition between the markers and the camera is obtained,and the point-cloud map of SLAM with real scale is constructed by using the triangulation principle.(3)The long-time tracking of ORB＿SLAM3 will produce cumulative error,which leads to the tracking precision decrease and tracking instability.In this paper,we propose an algorithm to correct the offset of SLAM pose by using markers.When a closed loop occurs,the detection and recognition of marker is performed to screen the loop candidate frames more quickly,accelerate loop correction,and shorten the drift time of SLAM tracking.The comparative experiments show that the tracking accuracy of SLAM trajectory is obviously improved.(4)In order to solve the problem of tracking failure when the object is occluded or the angle is not ideal,we fuse the marker information into the ORB＿SLAM3 point cloud map and enable the SLAM tracking when the marker tracking is lost.On the other hand,when SLAM tracking fails due to image smoothness or lack of feature points in realworld scene,marker tracking is enabled.The experimental results show that the proposed fusion algorithm has better Robustness for the tracking environment and the system is more stable.In order to verify the validity and feasibility of the proposed algorithm,an augmented reality system based on the above-mentioned fusion localization algorithm is developed.The system running results show that the performance of the proposed algorithm is obviously improved in the aspects of tracking initialization speed,tracking accuracy and robustness to environment.