Research On Localization And Mapping Technology Based On Fusion Of Multi-line Laser And Monocular Camera In Large Scenes

With the increasing attention and support of intelligent mobile robot technology,the localization and mapping technology as the core and premise of mobile robots has attracted more and more attention.In recent years,localization and mapping techniques based on vision or laser have gradually become the mainstream of SLAM.With the rise of autonomous driving,SLAM technology for large outdoor scenes has become a hotspot in the field of robotics research.This paper focuses on the location and mapping technology of monocular camera and multi-line lidar in large scenes.We study the SLAM algorithm based on vision and the SLAM algorithm based on laser in large scenes respectively.Combining the advantages of the two algorithms,we propose and implement a Localization and mapping algorithm based on 3D laser and monocular camera with high accuracy and robustness for large scenes.The main contents and results are as follows:1.A pose estimation algorithm combining 3D-3D,3D-2D and 2D-2D constraints is proposed.Starting from the structure and principle of camera and lidar,we study the visual Odometry algorithm based on sparse depth information for large scene.First,the sparse depth information is provided to the image by laser point cloud data,and then the ORB feature algorithm is used to extract and match the image features.Then,the proposed algorithm is used to solve pose estimation between images,which overcomes the shortcoming of sparse depth information in large scenes and realizes visual odometer in large scenes.Finally,the experimental comparison with the traditional PnP algorithm proves that the algorithm has better robustness and accuracy in large scenes.2.A registration algorithm of point cloud between frames is improved.We study localization and mapping algorithm based on single lidar in large scenes.First We extract the edge feature and plane feature.Then,We study the registration and distortion correction algorithm of point cloud between frames based on edge features and planar features and then in view of the shortcomings of the algorithm in large scenes an improved algorithm is proposed.We propose a hybrid algorithm used grid and elevation threshold to separate the ground and only non-ground feature points are used for iteration constraints of cloud registration to improve efficiency as well as ground planar feature are used as z-axis constraints to ensure the accuracy.By comparing the experimental results of the algorithms before and after improvement,it is proved that the improved algorithm not only improves the efficiency but also ensures the accuracy.3.Combining the above research on vision algorithm and laser algorithm,a location and mapping fusion algorithm based on laser and monocular camera for large scene is proposed.The motion estimation obtained by visual odometer is used as the initial value of the registration and distortion correction algorithm of point cloud between frames to solve defect that laser algorithm is prone to errors in the section where the feature of point cloud is not obvious.We construct the submap referring to cartographer algorithm and establish the Pose Graph by the pose constraint between the sub-map and the laser frame to realize Global optimization.Loop detection is achieved using the BoW algorithm based on image ORB feature to restrain the trajectory divergence and reduce the cumulative error.Finally,the experiments show that the robustness and accuracy of this algorithm are better than that of localization and mapping algorithm based on single sensor in various large scenes and it is a complete and effective localization and mapping algorithm.