Research On Localization Algorithm Of Driverless Car Based On Fusion Of GNSS,Visual And Laser SLAM

In the field of driverless cars,vehicle localization is one of its core technologies.The Simultaneous Localization and Mapping(SLAM)technology is a widely used localization technology in robotics,driverless cars and other fields.Laser SLAM with LIDAR as the main sensor can achieve autonomous motion estimation based only on the point cloud data provided by Li DAR.However,the laser point cloud data only contains the spatial location information of the surrounding environment and cannot utilize the rich texture information in the environment,which can easily produce incorrect feature matching and thus affect the localization accuracy.To address this problem,this paper uses Li DAR as the main sensor,uses the image data captured by the camera as the auxiliary,uses RGB information as the constraint for feature matching,introduces GNSS positioning information to suppress trajectory drift,and proposes a COLSLAM algorithm based on Fusion of GNSS,visual and laser SLAM to improve the localization accuracy of driverless cars and construct sparse color point cloud maps.The main work of this paper is as follows:(1)A point cloud coloring and ground point cloud extraction algorithm is proposed.Firstly,the collected laser point cloud is projected under the pixel coordinate system of the camera,and the RGB information of the corresponding pixels of the point cloud is obtained to color the point cloud.Then,a ground point cloud extraction algorithm is designed for the problem of insufficient accuracy of the ground extraction module in laser SLAM algorithm.The point cloud is filtered using a method based on the RGB information and relative pitch angle,and then the ground point cloud is fitted and extracted using a planar model.A method based on a combination of density and Euclidean distance is used to cluster the non-ground point clouds and filter out the noise points that are not clustered to prepare for the feature extraction stage.(2)A COL-SLAM algorithm based on the fusion of the RGB information and laser SLAM is proposed.For the problem that laser SLAM can only rely on point cloud spatial location information with mis-matching,the RGB information is introduced to reduce the mis-matching between feature points.Firstly,smoothness is calculated and features are extracted from the pre-processed point cloud,and then the RGB information of the point cloud is used as a constraint in the front-end laser odometry and the back-end local map matching part to remove the wrong matching point pairs and improve the positioning accuracy and robustness of laser SLAM.Finally,the colored point clouds are registered into the global map in the mapping stage to construct a colored sparse point cloud map.(3)To address the problem of cumulative error during the operation of laser SLAM,GNSS positioning information is introduced and a loop closure detection algorithm based on Scan Context descriptors is used to eliminate the cumulative error.Firstly,the point cloud of each key frame is described by Scan Context method,and the loop closure detection is performed for all loop closure candidate frames within a certain range.Then the laser odometry factor,GNSS factor and loop closure detection factor are constructed,and the cumulative error of the system is eliminated by the factor map optimization,and the corrected global point cloud map of bit position and color is obtained.(4)The effectiveness of the algorithm in this paper is verified through comparative experiments.Firstly,the ground point cloud extraction algorithm proposed in this paper is verified based on the KITTI dataset.And the experimental results show that the algorithm in this paper can extract ground point clouds more accurately.The comparison experiments based on KITTI dataset and real campus environment show that the COL-SLAM algorithm proposed in this paper has higher localization accuracy than Le GO-LOAM algorithm.