Urban 3D Reconstruction Based On Smartphone

Urban 3D reconstruction is an important part in the construction of digital cities and intelligent cities.Traditional 3D reconstruction techniques rely on expensive 3D scanning devices,such as 3D laser scanners,which are difficult to popularize.At present,research on street scene reconstruction technology at home and abroad mainly acquires video data through on-vehicle imaging devices.And then,an image-based 3D reconstruction algorithm usually runs to build a 3D street block on high-performance computers.The cost of research traditionally is high.With the rapid development of more than ten years,smartphones are generally equipped with high-performance cameras and also have certain computing power.Therefore,this thesis proposes the research in 3D urban reconstruction based on smartphone,and to explore low-cost,easy-to-popularized,and high-efficiency 3D urban construction technology.The main work in the thesis is the construction of a smartphone-based processing framework of 3D reconstruction for street scenes.The framework includes smartphone imaging,image/video preprocessing and cloud transmission,image-based 3D reconstruction and rendering algorithms.Firstly,the mobile camera is utilized to get the block video,and the key frames are extracted in the smartphone.Then the key frames are transmitted to the cloud server for feature detection and matching in-between consecutive frames.Secondly,based on the feature-point pair with high correct rate,the sparse point cloud of the scene is reconstructed based on the SfM algorithm,and then the dense point cloud is reconstructed based on the stereo correction.Finally,an existing method is employed to mesh the point cloud,and the mesh is sent back to the mobile phone to render and display.In this processing framework,feature point matching is the first key step.The accuracy of the matching results will seriously affect the performance of the subsequent point cloud reconstruction.In this paper,the SIFT method is used to detect the feature points in key frames and match the feature point pairs between adjacent frames.Then,based on the basic idea of RANSAC,feature point pairs are processed to eliminate mismatched point pairs.The second important part of the framework is stereo correction.In this paper,the stereo correction method is used to correct the actual imaging system of two cameras to the ideal binocular system to obtain disparity.Then the depth map is calculated by using the disparity of the image.After that,the depth map is transformed into a point cloud to create a dense 3D point cloud.Finally,the point cloud is registered and fused to obtain a dense point cloud of the entire scene.In summary,the main contribution of this thesis includes a smartphone-based processing pipeline is established for the 3D reconstruction of street scenes,and based on the idea of RANSAC,an algorithm for eliminating mismatched point pairs is proposed.