Research on Airborne LiDAR Point Cloud and Image Registration Based on ICP Algorithm with Line Feature Constraints
The traditional ICP algorithm has drawbacks such as low computational efficiency and susceptibility to local optima.This paper added line feature constraints to the traditional ICP algorithm and proposed an ICP algorithm with line feature constraints.The method adopted a coarse to fine registration strategy.Firstly,the SIFT algorithm was used to pair feature points in the DSM images generated from point cloud and image data.After eliminating mismatches with the RSANC algorithm,it served as the initial parameters for data registration.Then building boundary lines were extracted as line features from the original point cloud data and image DSM data,and the line features were simplified,fitted,or orthogonalized.The endpoints were used to represent the feature lines,and the endpoints of the same named feature lines were added to the ICP algorithm as line feature constraints to complete the registration of point cloud data and image dense point cloud data.Compared with the classic ICP algorithm,the new method proposed in this article fully extracts point cloud and image features without the need for operations on all point cloud data.Only a small number of points(about 3%)need to be sampled to participate in registration calculation,which can complete the registration of point cloud and image 3D-3D.The computational efficiency is greatly improved,and under the constraint of online features,it avoids the disadvantage of traditional algorithms easily falling into local optima.Through data experiments,the improved algorithm significantly increased the average point spacing of the painful ICP algorithm registration results from 0.462 m to 0.342 m,resulting in a significant improvement in accuracy.
NETL
NSTL
万方数据
