Automatic Registration Of Image Point Clouds With LiDAR Data Based On Plücker Coordinates

The three-dimensional registration of LiDAR point clouds and aerial image point clouds is the basis of multi-source data fusion.The traditional ICP algorithm requires the registration point cloud to be an inclusion relationship,and it is time-consuming to re-search the corresponding points when calculating registration parameters in each iteration.In addition,the Euclidean distance is considered as the nearest point distance,which exists a certain degree of mismatch.As a kind of registration primitive,linear feature is rich in urban areas,not affected by occlusion and has stronger geometric constraints.It does not need all points on corresponding lines to be corresponding points,nor all corresponding lines to participate in parameter calculation when performing registration,which can improve the registration efficiency and achieve high registration accuracy between two point clouds.Most of the existing linear registration algorithms manually extract corresponding features,and are unable to automatically establish the index structure of corresponding lines,which results in low registration efficiency.In addition,the registration geometric model only uses the orientation information of the linear features,without considering the position information and scale difference,resulting in low registration accuracy.Aiming at the above problems,this paper proposes a new automatic registration method for aerial images and LiDAR point clouds,directly performing threedimensional registration between dense-matching point clouds and LiDAR point clouds,and the research is carried out in two aspects: automatic extraction and matching of linear feature in point clouds and construction of registration geometric model.The main work and innovations are as follows:1.An automatic linear feature extraction algorithm for point clouds based on plane boundary coding is proposed.Index structure of point feature is established by KD tree and KNN,and the plane feature of point clouds is fitted by region growing algorithm under the constraints of he normal vectors.On this basis,the planar fitting points are projected to construct binary images,with the image boundaries encoded by Freeman normalized differential code,and the linear features can be fitted from the encoded boundaries by Hough transformation.The experimental results show that the proposed method has low extraction errors,with higher efficiency than the traditional plane intersection method.The extracted lines are easy to establish corresponding feature index structure,which is beneficial to improve the accuracy and the efficiency of feature matching.2.A linear feature index algorithm of point clouds based on asymmetric boundary is proposed.The asymmetric boundaries are classified according to the Freeman normalized differential code,with the angle and length ratio between intersecting lines used as the constraint criterions to automatically obtain the corresponding lines.The experimental results show that the encoded asymmetric boundary has a certain direction with a unique starting point,not affected by the rotation and translation,thus realizing the automatic matching of the corresponding linear features in different point clouds.3.A geometric registration model with the scaling coefficient based on Plücker coordinates is proposed.The corresponding linear features are described by the Plücker coordinates,making full use of the direction,position and scale information.A collinearity condition equation of feature lines is established based on screw motion,which the scaling coefficient is uniformly and iteratively calculated with the rotation and translation parameters under the least squares adjustment method.The results of the contrast experiment show that the proposed geometric registration model has stronger constraints and higher registration accuracy,which provides a new idea and method for the registration of urban aerial images and LiDAR point clouds.