Road Scene Modeling Based On Lidar Datas

The road scene model is the basic data of the autonomous driving simulation platform,car driving simulator,3D digital map and other systems.The use of automated modeling methods to quickly establish accurate road scene models is of great significance for improving production efficiency in these fields.This paper presents a method to build a road scene model using the 3D point cloud data collected by vehicle LIDAR,including: point cloud data preprocessing,point cloud registration,3D reconstruction,reconstruction result evaluation and other steps.The method can be used in the automatic modeling of road scenes and the automatic construction of 3D digital maps on various driving simulation platforms,and can be used in the fields of digital cities and digital road management.In this method,the LIDAR is designed to be installed on the top of the car,and the car drives at a constant speed to collect continuous point cloud frame data.Through the combination of statistical filter and conditional filter,noise point filtering and ground point separation are realized.The preprocessed data were registered using SAC-IA(Sample Consensus Initial Aligment)and ICP(Iterative Closest Point)algorithms.In order to improve the time-consuming problem of FPFH(Fast Point Feature Histograms)feature extraction in the SAC-IA registration process,this paper adopts CUDA to accelerate the extraction of FPFH features,and designs a multi-threaded structure to calculate the SPFH(Single Point Feature Histograms)and FPFH of each query point in parallel.The scene segmentation is performed on the registered point cloud data by combining prior knowledge and filters.Then use Poisson reconstruction and fast triangulation reconstruction algorithm for 3D reconstruction.In order to evaluate the reconstruction accuracy of the road scene reconstruction method,a simulation evaluation method based on virtual LIDAR is presented in this paper.The point cloud data obtained by the virtual LIDAR scanning the preset virtual scene is used as the input of road scene reconstruction,and the reconstructed scene model and the reconstructed object with known parameters are compared in terms of volume,elevation and contour,and then obtain the quantitative evaluation result of the reconstruction algorithm.In order to verify the method in this paper,experiments are carried out based on actual road and virtual road scenes.The test results of the actual road experiments prove the feasibility of the method in this paper,in which the use of CUDA acceleration improves the efficiency of the feature extraction process by about 91.2%.The accuracy of the model established by this method is quantitatively evaluated by simulation experiments.The experimental results show that the method in this paper can quickly build a 3D model of the road scene with sufficient accuracy by using the vehicle LIDAR data,which is an effective method for rapid modeling in related fields.