Research On Registration Method Of Optical Image And 3D Laser Point Cloud

Automatic 3D model reconstruction is an important research direction in the field of digital photogrammetry and 3D laser scanning.The 3D reconstruction technology of optical images have been very mature,but it still has its limitations.The geometric accuracy of reconstruction is lower than that of 3D point cloud obtained by laser scanner.It is of great significance to study the fusion of optical image and 3D laser point cloud.In order to realize the fusion of these two kinds of source data,the key is to unify the optical image and laser point cloud into the same coordinate system,that is registration.The coarse registration of optical images and 3D laser point clouds is mainly achieved by feature extraction and matching.However,manual extraction of control points is time-consuming and laborious,especially for 3D point clouds with unclear features,feature extraction is difficult.ICP(iterative close point)algorithm and mutual information algorithm are most commonly used in optical image and 3D laser point precise registration algorithm.These algorithms have good effect on optical image with small distortion difference,and poor registration effect when optical image distortion difference is large.Three dimensional reconstruction of optical image is to generate sparse point cloud by matching multiple optical images with a certain degree of overlap and aerial triangulation.Dense point cloud can be generated by dense matching,and then 3D model with texture information can be generated through bundle adjustment and texture mapping.In this paper,the sparse point cloud or dense point cloud generated by the images is referred to as the photography point cloud.The 3D reconstruction technology combining optical image and laser scanning point cloud needs to realize the registration of photography point cloud and laser scanning point cloud first.Due to the inconsistency of scale between the photographic point cloud and the 3D laser point cloud,and the large difference in the local density of the point cloud,it is difficult to register the photographic point cloud and the 3D laser point cloud.This paper aims to solve the problem of optical image and laser scanning point cloud registration,and studies the key steps involved.The main research work is as follows:(1)The method of automatic coarse registration of sparse point cloud and dense point cloud based on image generation and laser scanning point cloud is studied.A method of coarse registration between image and laser scanning point cloud based on the improved4 pcs algorithm is proposed.Aiming at the high accuracy of the image tiepoints,the number of points is small,the precision of dense point cloud is low,and the number of points is large.The sparse point cloud and dense point cloud are combined with the laser scanning point cloud for coarse registration,and the sparse point cloud is given higher weight when determining the maximum four point congruent sets.On the experimental surface,the coarse registration algorithm can achieve good results.(2)Research the non-rigid registration method in the fine registration of the sparse point cloud generated by the image and the 3D laser point cloud.he research uses the4 PCS coarse registration method based on scale improvement as the rigid registration method,and the point-point constrained beam adjustment model as the non-rigid registration to realize the rigid-non-rigid alternate registration.On the basis of the point-point constrained adjustment model,a point-surface constrained bundle adjustment model is proposed,which improves the precision of the optical image and the 3D laser point cloud with large changes in surface curvature.