Research On 3d Reconstruction Technology Of Traffic Scene Based On Laser Point Cloud

At present,data retention in the transportation industry is generally based on pictures,audio,video and other data.The content that can be stored is relatively simple and the information provided is relatively limited.The feedback of sudden traffic accident information such as car accidents or obstacles is not perfect and not timely.Pictures and videos alone cannot describe the positional relationship in depth.At the same time,three-dimensional reconstruction technology has developed rapidly,and it has shined in various industries such as construction engineering,geography and landforms,and mechanical tools.Its application in the field of transportation remains to be studied.To this end,this article proposes the application of 3D reconstruction technology in the field of transportation research,using 3D reconstruction technology to reconstruct the 3D model of the traffic scene.The main research contents of the article are as follows:The data obtained by lidar is a three-dimensional point cloud,which is easy to obtain and quick to measure.This thesis uses lidar to collect point cloud data of objects and scenes,and studies how to process the acquired data,so as to obtain higher-quality point cloud data for subsequent reconstruction algorithms.The research steps are: first use voxel grid filtering to downsample the experimental data,remove outliers through radius filtering,and then use filtering to reduce noise and smooth the data.After preliminary preprocessing,this article uses experiments to implement several existing point cloud data feature point extraction algorithms based on open source libraries such as PCL,visualizes the experimental results,and analyzes the different algorithms on the basis of the experimental results.The number of key points and the length of time.Finally,the calculation of normal vector,eigenvalue and curvature of point cloud data is studied.After point cloud preprocessing,feature extraction and description,etc.,high-quality point cloud data is obtained,this thesis takes Poisson algorithm as the main research object,uses the data to implement the existing algorithm and analyzes the problems of the existing algorithm.Aiming at the time-consuming problem,an adaptive octree partition algorithm is proposed with the point cloud density to determine the threshold;for the most common Laplacian basic solution of the Poisson equation and the known function convolution,this article tries The Poisson equation is solved by trilinear interpolation and multi-grid method.The original algorithm and the optimized algorithm are analyzed and compared from the three aspects of the reconstructed surface effect,error and time-consuming,and the feasibility of the optimized algorithm is verified.Finally,the article analyzes and discusses the feasibility of applying the research content in the field of intelligent driving,and puts forward suggestions for the current research deficiencies.The research in this thesis can realize the 3D reconstruction of a small-scale scene.For the problem of the time-consuming 3D reconstruction of the scene,this thesis proposes an adaptive octree partition threshold and the use of trilinear interpolation to directly solve the Poisson equation,reducing part of the reconstruction calculation process and time.Experiments prove that the optimized 3D reconstruction algorithm in this thesis can reduce the time-consuming experiment to a certain extent,and the triangular mesh model of the reconstructed scene is evenly distributed,the contour of the 3D gray scale model is clear,and the background and objects have a clear sense of hierarchy.The size identification is relatively accurate,and the reconstruction effect is good.