Study And Realization Of The 3D Reconstruction Techonology With A Single RGB Camera

3D reconstruction is one of the main techniques of Virtual Reality,Augmented Reality and other computer vision applications.It captures the shape and appearance of real objects to build high-quality 3D models,which are suitable for digital processing.In this thesis,a passive image-based 3D reconstruction method is discussed.In the system pipeline.Firstly,the tracker estimates the camera pose based on the input RGB frames.Then,the depth map computation module computes the depth map of current frame based on stereo matching.At last,the output depth map is fused into the volume model according to the captured camera pose.This thesis realizes the architecture and algorithm design of the 3D reconstruction system with a single RGB camera.What's more,focused on the purpose of enhancing system performance,an intensive study is constructed on camera tracking and stereo matching algorithms.In this thesis,ORB-SLAM is incorporated into the system to realize camera tracking.The camera tracking algorithm incorporates three threads that run in parallel:tracking,local mapping and loop closing.The loop closure thread is designed to reduce the accumulated tracking drift and the bag of words place recognition module ensures the good invariance to changes in viewpoint and illumination.Based on the framework of local stereo matching which uses minimum spanning tree for aggregation,in this thesis,a high accuracy method is implemented.Census feature is introduced in the process of matching cost computation.It improves the accuracy of disparity in textureless areas.Also,a two-step disparity repairment strategy and disparity refinement method based on plane fitting are adopted to improve accuracy near depth edges and slanted surface.Using the methods mentioned above,the output disparity map is accurate,even in complex regions.The evaluation of the camera tracking algorithm shows that the proposed method achieves 35.9%relative pose error reduction compared with PTAM algorithm based on the TUM standard dataset.For local stereo matching algorithms,the proposed method attains 22.1%bad pixel percentage reduction in Middleburry standard dataset.Based on the improvement of tracking and stereo matching algorithms,our system achieves high-quality 3D reconstruction.