Three-Dimensional Modeling And Road Feature Detection Of Outdoor Low-Speed Unmanned Vehicle

Autopilot can change the current transportation system,improve efficiency and reduce accident rate.Autonomous driving requires robots to have reliable perception systems in changing environments,and to navigate in complex environments with unknown and GPS constraints.Real-time three-dimensional environmental information acquisition provides the possibility for outdoor robots and autonomous vehicles to navigate safely in unknown environments.Multi-layer laser is the most suitable three-dimensional sensor for outdoor sensing.In view of the above problems,this paper focuses on the establishment method of laser odometer,the construction method of three-dimensional environmental map and the detection method of real-time road information.Firstly,this paper improves the method of establishing laser odometer based on feature points.According to the characteristics of outdoor environment and point cloud data,after pre-processing point cloud,this paper uses the ground separation method based on range image to separate the original point cloud into ground point sets and non-ground point sets.On the basis of using clustering algorithm to eliminate discrete interference points,the feature extraction method based on random sampling and line and surface feature extraction based on point roughness value are used to extract feature points from ground point and non-ground point respectively.The characteristics are matched,and the laser odometer is divided into two steps to obtain,and the optimization calculation of the laser odometer is carried out by L-M optimization algorithm.Secondly,this paper improves the method of constructing point cloud map of outdoor three-dimensional environment.Based on the acquisition of the laser odometer,the system adds a loop detection step and a graph-optimized optimization thread to the process of constructing the point cloud map.The key frame and the local map are established according to the matching degree of the point cloud information,and the pose sequence is obtained corresponding to the key frame,and the loop detection is performed by using the overlap degree between the current key frames,and the edge constraint is established for the first and last frames of the loop.The error function is obtained by the calculated value of the pose transformation and the observation valueof the sensor,and the error function is minimized to correct the key frame pose,thereby updating the odometer and map information.In addition,this paper presents real-time road information perception using multi-layer laser.The candidate roadside information is obtained by grid processing of point clouds and elevation information.The point clouds are converted into binary images,And the image processing steps such as expansion are continuously processed,and the complete roadside information is obtained by curve fitting using the sliding window to establish the iterative.Finally,based on an outdoor unmanned vehicle platform designed independently,the methods of establishing laser odometer,three-dimensional environment modeling and real-time road detection proposed in this paper are experimented,and the experimental results of the above methods in different scenarios of the actual environment are analyzed.