Research On Registration Technology Of Vehicle And Ground Point Cloud Data

As a high-tech technology that is gradually developing and mature,3D laser scanning technology can acquire a large number of point cloud data with high precision and high resolution in a short time.It has strong real-time performance and is widely used in urban landmark information.Actual engineering measurement such as collection,road flatness detection,highway landmark object extraction,etc.According to the different carriers,the 3D laser scanning system can be divided into:Mobile Mapping System,Terrestrial Laser Scanning System,Airborne Laser Scanner System.The vehicle-mounted mobile measurement system has the characteristics of simple operation,high-intensity real-time,high measurement accuracy,high measurement efficiency,and continuous operation;But in the area of narrow,complex terrain,in case of GNSS lost lock for a long time is difficult to get satisfy the measurement precision of the whole surface data,so it is necessary to use the Terrestrial Laser Scanning System to assist the survey.In order to obtain the complete 3-D point cloud model of the fusion of vehicle and ground point cloud data,this paper makes a deep research on the mosaic technology of vehicle and ground point cloud data.The main research work is as follows:1.Studied the relevant theories of common coordinate transformation models and analyze their applicability.The commonly used coordinate transformation models include nonlinear seven-parameter model,rodrigue matrix model and quaternion model;the three coordinate transformation models can be used for large rotation angles,however,the nonlinear conversion accuracy affected by the seven-parameter model is selected initial value,the linearization procedure is cumbersome and complicated calculation;the rodrigue matrix model does not need to be linearized and can directly solve the parameters,can be directly calculated,but there are singular points;The quaternion model can be solved directly,the model is simple,the calculation speed is fast,and the calculation parameters have high precision.2.An improved ICP algorithm based on curvature extraction feature points is studied.Aiming at the problem that the traditional ICP algorithm requires high initial position of the point cloud,a rough stitching algorithm based on feature points is proposed,the rotation point R and the translation matrix T are solved by the feature points to implement coarse registration of point cloud data.Then,the point cloud index is created by using the octree to improve the point cloud neighborhood search speed,and extracting the feature points by calculating the curvature,calculating the Hausdorff distance of the curvature between each feature point,and retaining the corresponding point pair whose Hausdorff distance is less than the threshold,complete the fine registration of point cloud data,solve the traditional ICP algorithm search point recently long time consuming,error point to more problems.The registration results of traditional ICP algorithm and improved ICP algorithm are compared and analyzed,experiments show that the algorithm has high registration efficiency and can get accurate stitching effect.3.The integrated registration of vehicle and multi-station ground point cloud data is studied.The improved ICP algorithm based on curvature extraction feature points is used to registration the vehicle and ground point cloud data,and the rotation and translation parameters are calculated,due to the influence of error accumulation and transmission,the splicing result will have a large error.Therefore,taking the calculated parameters as the initial value,the coordinates of the common point pair in the unified coordinate system should be equal,and the converted common point should also be fitted to the same plane as the center of the two stations,and the rotation and the translation parameters are corrected to obtain the registration result that meets the accuracy requirements.Field collection of vehicle and ground point cloud data,after denoising treatment,splicing experiments are carried out to verify the feasibility and accuracy of the algorithm...