LiDAR point cloud no-initial-value registration method based on planar feature constraint

Abstract

The invention discloses a LiDAR point cloud no-initial-value registration method based on planar feature constraint. According to the invention, expression of plane features in a three-dimensional space is realized by adopting a four-parameter method; parameters of homonymous plane features of the registered adjacent observation stations are equal to serve as constraint conditions, a three-dimensional space similarity transformation objective function based on dual quaternion description under plane feature constraint is constructed according to the least square criterion, and non-initial-value solving of the registration parameters is achieved through extreme value analysis of the objective function. The four-parameter expression of the plane features provides a simpler and more effective mode for the comparison of homonymous features; compared with a vector algebra, the spatial similarity transformation model based on dual quaternion description is simpler in expression form and fewer in additional constraint conditions in the registration process; compared with an iteration method, the algorithm has the advantages that the parameters are directly solved on the premise that the initial values of the parameters of the space similarity transformation model do not need to be determined in advance, the stability of the algorithm is better, and the reliability is higher.

Description

technical field [0001] The invention specifically relates to a LiDAR point cloud registration method without an initial value based on plane feature constraints. Background technique [0002] LiDAR, with its fast, efficient, high-precision and other excellent characteristics, provides a reliable data guarantee for the comprehensive, true, fast and accurate reproduction of geospatial entities and their environmental information. However, due to the fact that the spatial complexity of geographic entities is generally high, and the visual area of ​​LiDAR sensors is usually narrow, in order to obtain surface feature data that can fully characterize geographic entities and their surrounding objects, spatial 3D data based on ground LiDAR technology The data acquisition method usually needs to aim at the object along multiple different viewing directions, collect LiDAR point cloud data that can describe the surface characteristics of the target geographic entity and its surrounding...

AI Technical Summary

Problems solved by technology

However, the iterative method itself has the following deficiencies: 1) It is necessary to determine the initial value of the transformation parameter in advance, and realize the linearization of the nonlinear objective function model accordingly. Improper selection of the initial value may result in incorrect results or incorrect function. No convergence; 2) Constrained by the linearization process, this type of method is essentially only suitable for small corner coordinate transformations. In the registration process of LiDAR point clouds, if the similar transformation model parameters between the two coordinate systems cannot be provided The initial value, or the initial value of the similarity transformation model parameters provided is not ideal, and the correctness and reliability of the calculation results will be seriously affected

However, using vector algebra to describe straight line features and plane features in 3D space, the diversity of its expression forms directly leads to the diversity and complexity of LiDAR point cloud registration model construction based on line / plane feature constraints. phenomenon is more evident in the design of analytical registration models

The proposal of Plücker linear coordinates overcomes the shortcomings of linear features in the LiDAR point cloud registration process to a certain extent. It describes the straight line in three-dimensional space through six parameters such as the direction vector of the straight line and the moment of the straight line. The meaning is clear and the form is more concise. However, the implementation of the existing algorithm still adopts the basic idea of ​​the iterative method. Compared with the iterative algorithm based on the description of vector algebra, although the algorithm is less dependent on the selection of the initial value of the iteration, the algorithm running results The correlation between the choice of the initial value and the initial value cannot be explained theoretically

Images