Research On SLAM Technology Based On LiDAR And Its Application In Autonomous Driving Vehicles

The rapid development of autonomous driving vehicles technology is promoted by some major factors including the rapid development of science and technology,the increasing living standards of people and the rising use of 5G Internet of Everything.In the future,autonomous driving vehicles are expected to replace all traditional automobile and make traffic environment free from pollution,traffic jams and traffic accidents.The research contents of this thesis are issues concerning simultaneous localization and mapping(SLAM)in autonomous driving vehicles.The main challenge of this research field is to realize the positioning of the vehicle and the creation of the environmental map simultaneously by using the on-board environmental sensing sensor when the autonomous driving vehicles are in an unknown environment without the external positioning information.In this thesis,the LiDAR environmental sensing sensor is chosen to achieve precise positioning of autonomous driving vehicles and to create highprecision 3D environment maps.The main research contents of the thesis are as follows:First,the traditional ICP inter-frame matching algorithm is introduced in this thesis.Aimed at the deficiency of the ICP algorithm,an improved inter-frame matching algorithm based on feature points is designed.Differing from the point-to-point data association of the ICP algorithm,the improved algorithm designed in this thesis first extracts the feature points in the point cloud,and then performs point-to-line and point-to-plane data association between the extracted feature points and the reference point cloud composed of multi-frame feature points.The pose estimation obtained by this data association method is more accurate than the traditional ICP algorithm.In order to further accelerate the point cloud matching speed and elevate the matching precision,this thesis also designs a radial gradient ground extraction algorithm and a conditional grid filtering algorithm applying to ground removal and noise reduction of the point cloud respectively before inter-frame matching.Subsequently,an improved graph optimization SLAM algorithm is designed in this thesis.Pose graph is directly used to optimize all poses in the algorithm based on the fact that point cloud is dense and inappropriate to extract landmark.Meanwhile,the incremental solution method is used to speed up the optimization.The constraint between adjacent poses in the pose graph is directly constructed by the odometer estimate,and the loop-closure pose constraint is constructed by the loop-closure detection section.In order to achieve faster and more accurate loop-closure detection,a fast loop-closure detection algorithm is proposed based on point cloud in this thesis.Via the introduction of the pose estimation on the front-end odometer,the loopclosure detection issues are converted to candidate loop-closure pose search and fast loopclosure verification,thus realizing loop-closure detection fast and accurately.Finally,the SLAM software is designed based on the ROS environment,which integrates the inter-frame matching algorithm based on feature points and the improved graph optimization algorithm.The node and topic mechanism in ROS are used for the software design.By means of designing different functional modules as independent nodes and establishing connection between nodes through topic,consequently running speed of the software and the real-time performance are improved.The software is validated via the experimental platform of autonomous driving vehicles,and it is proved that the SLAM software designed in this thesis can effectively complete the task of localization and mapping simultaneously.