Model Construction Of The Environment Around Unmanned Ground Vehicles Based On 3D LiDAR And Image Fusion

With the development of society,driverless technology is developing at an unprecedented rate and new attention paid to unmanned vehicles.while the unmanned vehicle is moving,it must sense and identify conditions of the surrounding environmen.The information on vehicles and their surroundings is a major requisite for autonomous positioning,path planning and navigation.However,the environment around unmanned vehicles is very complicated and there are a lot of uncertainty.The unmanned vehicle should have the ability to identify and model the dynamic environment.The environment modelling technology has become an important research topic in the field of unmanned driving.In order to model the surrounding environment,the algorithm of data processing is applied based on 3D Li DAR and image fusion,and the algorithms of feature point extraction and automatic image registration are proposed.Furthermore,a new modeling scheme has been put forward,which consists of separating feature points,matching characteristic point,integrated data and modelling the surroundings.The main content of this paper and the results are as follows.The fusion processing method of LIDAR and color image data is researched.The experiments were conducted.First,the point cloud data collected by the Li DAR is fused with the color images collected by the vision camera through pre-processing experiments.Then,the fused data were preprocessed with coordinate transformation,gross error removing,morphological filtering,etc.Experimental results show that this method can accurately separate the ground points from the non-ground points.The extraction algorithm of feature points targets is studied.Based on the extraction principles of normal vector feature points and curvature feature points,experiments of feature points extraction were conducted.Two kinds of non-ground feature points in two adjacent frames are extracted by calculating the normal vector and curvature of the point cloud data.Through experimental comparison,it shows that the feature target extraction based on normal vector feature points is fast,but the extraction accuracy is low,which is suitable for modeling base environment in complex scenes.The extraction algorithm based on curvature feature points is complex,but the extraction accuracy is high and the results can better represents the object’s feature information,which is suitable for modeling environment by small amount of point cloud data.The automatic alignment algorithm based on iterative feature points is presented.With comparative analysis of the Normal Distribution Transformation algorithm and the Iterative Closest Neighbor algorithm,this paper designs an automatic alignment algorithm for iterative feature points by combining two algorithms.First,perform coarse matching by the normal distribution algorithm,and then perform fine matching on the coarse matching result by iterating the corresponding feature points.Experiments were conducted using this algorithm on the extracted feature points,and the experimental results showed that the algorithm improves accuracy and speed of alignment.The experiments of modeling complex surrounding were performed.An automatic alignment algorithm of iterative feature points is applied to align non-ground points,and the triangulation model is constructed for the ground point data.By integrating the ground model with the non-ground model,a complete 3D environment map model around the Unmanned Ground Vehicle is constructed.The result verifies the feasibility and effectiveness of the complex scene modeling scheme based on 3D Li DAR and image fusion.The innovations of this paper are as follows.First,it originally proposes the ground point and non-ground point separation modeling scheme.The model construction is more effictive,and the scene rendering is more refined.Second,the automatic alignment algorithm with iterative multiple feature points is proposed,which can be used to select feature point types according to the mobility of the scene.It reduces the time of the fine matching process of 3D scene modeling and improves the accuracy of model matching.