Research On Quality Improvement Of Mobile LiDAR Point Clouds

A Mobile LiDAR system is a combination of a precise GNSS/IMU integrated navigation and positioning device(a POS),some laser scanners,some cameras and other sensors.The MLS provide a safe and time-efficient way to collect 3D point clouds.It has become an advanced technology for acquisiting and updating geo-spatial information.This technology can be applied to many transportation-related fields,such as road information inventory and production of high accuracy driving maps for intelligent driving.The accuracy and completeness of mobile LiDAR point clouds directly affect the object extraction,recognition and subsequent applications.If the laser scanner has been calibrated before data acquisition,the error of point cloud is mainly caused by the POS error.Due to the error of GNSS signal occlusion or multipath effect,POS error is inevitable.The relative accuracy of point clouds can be improved by the constrain and optimization of overlapped multi-stip point clouds.This method is called strip adjustment(SA).To improve the absolute precision,GCPs should be used in the adjustment,which is called ground control adjustment(GCA).The adjustment is to optimize the positions and attitudes.How to establish a pose model need to be studied.The model describes observation equations and the relationship between adjacent poses.To repair the incompleteness of local areas caused by obstructions or field angle restrictions,terrestrial LiDAR point clouds are used.The registration method for uniting the coordinate system of terrestrial LiDAR and mobile LiDAR point clouds need to be studied.The registration among local areas of overlapped multi-strip mobile LiDAR point clouds can also generate correspondences for establishing observation equations of the SA.To improve the accuracy and completeness of mobile LiDAR point clouds,this paper studies the point cloud registration,pose model establishment,SA and GCA:1.Automatic registration of point clouds using a genetic algorithm(GA):In traditional automatic registration,coarse registration is first applied based on feature matching,and then ICP is used to finish fine registration.The registration based on GA does not depend on initial transformation parameters,so it is more applicable.However,the accuracy and efficiency of existing GA registration is low and need to be improved.In the modified method,the rough position obtained by the built-in GPS of the terrestrial laser scanner is used as priori information to determine the optimal search space and a maximum registration model called Normalized Sum of Matching Scores(NSMS)is presented.The root mean square errors(RMSE)of the modified GA registration between terrestrial LiDAR and mobile LiDAR point clouds can be achieved 5cm.The proposed method can also be used for registration between terrestrial LiDAR point clouds or mobile LiDAR point clouds.2.A pose model with time-variant fitting of pose errors:According the geometric positioning model of Mobile LiDAR,the errors and its influence of point clouds are analyzed and the observation equations of poses are established.The cubic spline function with time stamps is used to fit the pose errors,and the optimization objective function that minimizes the sum of squared distances of the correspondences is estimated to unknown parameters of the model.The abbreviation of the model is time-variant pose model(TPM).The assumption of TPM is that the pose errors are small in a short period of time,which is regarded as continuous change.3.Automatic SA:The method is divided into 3 steps:preprocessing of point clouds,correspondence estimation,and solving the TPM.The preprocessing is to prepare for correspondence estimation,to reduce the number of points,and to eliminate points with a little contribution to correspondence estimation,and then to improve the efficiency.According to the characteristics of overlapping multi-strip point clouds,a two-layer registration strategy is proposed to obtain correspondences.The accuracy for SA is measured by the distance RMSE of correspondences.The relative accuracy is 3-5cm if there is enough overlap.4.Sparse GCA(SGCA):An adjustment method only using GCPs is proposed for a single strip.A SA method with additional GCPs is presented for overlapped multi-strip point clouds.The absolute precision is improved and its value is about 7cm.The average space distance of GCPs should be about 50m for adjustment using only GCPs and 100m for SA with additional GCPs.The test point clouds are collected at slow speed(20-40km/h).The average space distance can increased properly at high speed.