Research On Multi-sensor Fusion Localization And Mapping Based On LiDAR Point Cloud Features And Intensity

As an important technology to promote the development of automobile intelligence,autonomous driving has already become a hot research topic.In the technical system of high-level autonomous driving,localization and mapping can output high-precision vehicle position and attitude data and high-quality environmental maps,which provide important basic information for subsequent planning,control and other links,and have a decisive impact on the safety and efficiency of the overall autonomous driving task.At present,there are a variety of independent localization or mapping methods,each of which has outstanding advantages but has obvious shortcomings.The multi-sensor fusion scheme can fully combine the advantages of each sensor and enhance the overall accuracy and robustness by using complementarity,so it has become a key research direction.Among all kinds of sensors,Li DAR has a high precision of environmental measurement and is not easily affected by illumination or other conditions.Therefore,this paper studies the Li DAR based multi-sensor fusion localization and mapping method.Due to the mainstream methods mainly rely on the structural features of Li DAR point cloud,in the scene with unbalanced feature distribution and large range and multiple interference,the odometry and loop closure detection will be directly affected,resulting in large local or global errors,which significantly reduces the accuracy of overall localization and mapping.Therefore,in view of the specific problems existing in localization and mapping methods,this paper studies the Li DAR odometry,error correction based on multi-sensor and multi-sensor fusion framework,proposes a multi-sensor fusion localization and mapping method based on Li DAR point cloud features and intensity,and makes use of rich and targeted scene data for experimental verification.The core content can be divided into the following four parts:(1)Li DAR Odometry Combining Point Cloud Features and Intensity InformationOdometry is an important basis for localization and mapping methods.Aiming at the problem of accuracy decline of mainstream odometry in the scene of uneven point cloud density and feature imbalance,this paper proposes a Li DAR odometry combining point cloud features and intensity information.Firstly,the scale and basic errors of the original Li DAR data are reduced by point cloud IV preprocessing,and the corrected Li DAR point cloud is obtained.Then the structural features of the spatial point cloud were extracted based on the curvature data to form a stable line-plane feature set.On this basis,the Li DAR reflection intensity information is fully mined,and a point cloud registration method combining the feature and intensity is proposed.The accuracy of Li DAR odometry in the feature unbalance scene is improved by using the double information of the feature and intensity.(2)Correction of Local and Global Errors in Localization and MappingError correction is very important to the accuracy and robustness of localization and mapping.Aiming at the problems of poor robustness of Li DAR odometry in fast vehicle movement and large error in long distance driving,this paper establishes an error correction module of localization and mapping based on a variety of sensor information.Firstly,the high frequency position attitude measurement of IMU is converted into the prior information of Li DAR key frame by pre-integration,which corrects the local error of odometry and enhances its robustness in fast motion.Secondly,a loop closure detection method based on double-layer descriptor is proposed,which improves the quality of point cloud association in large range and high dynamic environment and corrects the global localization error by searching and matching from rough to fine.Finally,the quality test of vehicle GNSS data is carried out,and the position observation provided by it is used to improve the global accuracy of localization and mapping.(3)Multi-sensor Fusion Localization and Mapping Combined with Odometry and Correction ModuleThe combination of odometry and correction module and the fusion of various sensors improve the global accuracy and robustness of localization and mapping methods.Firstly,based on the concept of keyframe,the vehicle pose data and the environment feature point cloud are stored and managed.Through the optimization and adjustment of keyframe,the corresponding information in the odometry and correction module is further combined.Secondly,the sensor information of Li DAR,IMU and GNSS is converted into the corresponding observation node by using the framework of factor graph,which is connected with the status node of vehicle,and the overall multisensor fusion factor graph is constructed.Finally,the global pose information of the vehicle was optimized by updating various constraints in the factor graph,and the global point cloud map of the environment was established by using the modified local map,and the high-precision localization and mapping results were obtained.(4)Experimental Test and Result Analysis of Localization and MappingFinally,this paper built a test platform based on ROS,and carried out targeted experimental tests according to the sequence of scenes from half open to open with increasing difficulty.By comparing with true value and open source method,the proposed method is evaluated comprehensively.The multi-scene chart data and joint analysis show that the odometry,local and global error correction module combining the point cloud features and intensity information has achieved good practical results.The overall localization and mapping method has maintained high accuracy and robustness in a variety of scenes,and can output stable vehicle positioning data and high-quality environmental point cloud map.