| In recent years,with the development of artificial intelligence technology,its applications have become ubiquitous in daily life.As a prevailing direction in artificial intelligence,unmanned driving is also developing rapidly.High precision map,as an important part of driverless technology,plays a significant role in perception,positioning and decision-making of driverless technology.Simultaneous Localization and Mapping(SLAM)is a kind of high precision point cloud map generation algorithm.According to the different sensors used,SLAM is also divided into visual SLAM and Li DAR-SLAM.Visual equipment is very susceptible to interference from external light.In contrast,the point cloud data captured by lidar is more stable.At the same time,the point cloud information collected by lidar can better reflect the structural information of the surrounding environment.Therefore,this paper researches for the Li DAR-SLAM.Li DAR-SLAM includes front-end odometer,back-end optimization,and mapping steps.The role of the front-end odometer is to use point cloud registration to obtain the rigid transformation between two adjacent frames of lidar.The result of IMU pre-integration can provide the initial pose transformation matrix for the front-end odometer.Because the front-end odometer only considers the position and pose transformation relation between two adjacent frames,there will be accumulated errors.Back-end optimization reduces the impact of accumulated errors by adding constraints such as loopback detection and GNSS absolute value.Mapping is to generate a point cloud map by splicing each adjacent point cloud.At present,the point cloud registration algorithm used by many Li DAR-SLAM algorithms is a registration algorithm based on feature points.For example,LIO-SAM algorithm uses the point cloud registration algorithm based on curvature features.The algorithm does not consider the moving objects in point cloud registration,which will affect the accuracy of point cloud registration.The existing loop detection algorithms are not robust.For example,the loop detection algorithm based on Euclidean distance cannot detect loop frames at a large distance,and the loop detection based on Scan Context has higher requirements for the stability of the environment at the same location.In view of the above problems,the main research contents of this paper are as follows:(1)In order to solve the problem that the point cloud registration algorithm based on curvature feature does not consider moving objects,this paper proposes an improved point cloud registration algorithm based on curvature features.Compared with the original algorithm,this algorithm uses point cloud reflection intensity information to determine whether it is a point cloud on a moving object during the registration process.The accuracy of point cloud registration is improved.(2)In order to solve the problem that the existing loop detection algorithms are not robust,this paper proposes a loop detection algorithm that combines Scan Context and Euclidean distance.This algorithm combines the advantages of the two loop detection algorithms and improves the robustness of the loopback detection algorithm.This paper uses this method to detect looped frames and establish a new constraint relationship in the factor graph.It can reduce the impact of accumulated drift errors.(3)This paper uses the sensors and vehicles provided by the project team to complete the collection of real campus data.By aligning the time of the data,this paper establishes a small data set containing GNSS,IMU and lidar information,and verifies the effectiveness of the algorithm in this data set and the public KITTI dataset. |
