Research On Registration Method Of Point Cloud Data And Optical Image Based On Feature

With the popularity of multi-source sensors in the field of surveying and mapping,point cloud data and optical images have been widely used.Point cloud data compared with optical images can provide high precision 3D spatial information,but compared with optical images,the surface texture information of ground objects is deficient,which will cause certain difficulty in data processing.Therefore,how to coordinate point cloud and optical image,and how to establish mapping relationship between two kinds of data with the aim of realizing complementary advantages have been became research hotspots.Point cloud and optical images have great differences in spatial dimension,data structure and features,which means that the registration is difficult.Most of the existing registration methods have complex algorithms and limited adaptive scope.The registration method based on point cloud depth map has good applicability in both registration,but the difference between point cloud depth map and optical image is significant,so it is difficult to register.To solve the above problems,the registration method of point cloud data and optical image was studied in this paper.Based on the improved RIFT feature algorithm,indirect registration of point cloud data and optical image was realized,and the external image elements were iteratively optimized to achieve direct registration and true color mapping of point cloud data of two different sources.The main work contents are as follows:(1)Under the features-based registration scheme,to solve the problem of unsatisfactory local mapping of image registration,a thin plate spline model is introduced to replace the rigid registration model to improve the RIFT algorithm,so as to achieve better mapping transformation of the image to be registered both locally and globally,the correct matching rate is improved by about 5%.(2)The point cloud data and optical image registration are realized based on the improved RIFT feature algorithm.Firstly,point cloud depth image is generated by pinhole imaging model.Then,the improved RIFT feature algorithm is used to extract corner points and edge points as registration elements,and Euclidean distance is used as similarity measure to realize the registration of point cloud depth map and image,and then the indirect registration of point cloud data and optical image is realized.(3)Based on the space resection method,the exterior orientation elements of the image are optimized.Using the feature points obtained by indirect registration,under the condition of the minimum mean distance constraint of feature points,the iterative optimization of external pixels of the image is realized based on the space rear intersection method,and then the direct registration and RGB true color mapping of point cloud data are realized based on the collinear equation of photogrammetry.In order to verify the effectiveness of the proposed method,registration experiments are carried out with airborne point cloud data and aerial survey images,and the results of registration and parameter optimization are analyzed from the aspects of visual accuracy and coordinate accuracy.The analysis shows that the proposed method is effective and reliable due to the high data registration accuracy.