Key Technologies Of Laser Point Cloud And Image Fusion

Point cloud data acquired by laser scanning equipment of different platforms has become an important means for fast acquisition of 3D data in smart cities.However,due to the limitation of scanning mode and viewing angle,it is generally difficult to obtain complete point cloud of the object.The image point cloud can obtain the measurement data of the complex area by extracting the image data of a certain overlap rate and generating a dense three-dimensional dateset according to the computer vision.How to combine the point cloud obtained by these two different platforms with their respective superior information and complementary needs is the key research content of smart city construction in recent years.Aiming at the specific reasons for the lack of 3D laser point cloud field acquisition,combined with the characteristics of image point cloud,the method of laser and image point cloud fusion is studied.The main research contents are as follows:(1)For the point cloud,there are a lot of noise points affecting the fusion precision problem.The denoising work before the point cloud fusion obtained by the two sensors is studied.The common point cloud denoising algorithm based on statistical analysis is studied.The near-outlier point identification algorithm is added to remove the near noise points close to the main point cloud.(2)Aiming at the problem that the amount of point cloud before fusion is large and affects the efficiency of fusion,a method of comprehensively judging feature points based on curvature,distance and angle of normal vector is proposed,and the surface area,volume and model error of the model are reconstructed.Three aspects are compared with traditional simplified methods.(3)For the problem of laser point cloud cavity and missing,the registration method based on scale factor iteration(SICP)algorithm is used to complete the corresponding pair of missing points.Firstly,a rough registration method based on curvature and geometric features is proposed to provide a good initial position for the point cloud to be registered to prevent local optimum.Then,a dynamic iterative factor and a stepwise optimal solution scale factor are added to the SICP algorithm to reduce the iteration,and improve accuracy and effectiveness.The experimental results show that the near-outlier point identification algorithm is used to remove the noise around the subject point cloud and obtain the pure point cloud data.The point cloud reduction method based on the curvature,distance and normal vector angle.Compared with the traditional method,the point cloud coarse registration method can be maintained more;the point cloud coarse registration method based on curvature and geometric features can provide a better initial position for the point cloud to be registered,and the improved SICP algorithm can improve the registration accuracy and reduce the number of iterations.