Fusion Processing Of LiDAR Data And CCD Images

In recent years, the LiDAR technique based on active measurement principle has been got more and more attentions of various fields, and the board application future has been material-lized with its unique advantages. With the unceasing enhancement of its hardware integration, many terrestrial LiDAR systems are equipped with CCD cameras, can obtain point clouds data and CCD images in measuring. At present, the CCD images mostly have been just used for texture mapping. In fact, if the informations of images can be made full use to realize the fusion processing of terrestrial LiDAR data and CCD images, it will have great significance to give full play to the spatial information acquisition of LiDAR.Through analyzing the characteristics of two kinds of data, this paper starts by the many complementariness of them to describe the object, focuses on the fusion processing of terrestrial LiDAR point clouds and CCD images, and presents some new ideas for solving problems of the point clouds data processing combining the existing classic methods. The contributions of this paper can be detailed as follows:1. The data acquisition principle of the LiDAR technique and CCD cameras, the correspon-ding coordinate transformation and the registration of point clouds and images are expounded. According to the relationship between two kinds of data, the point clouds is separated by the image grid so as to establish a simple topological relationships of point clouds, and bring convenience to the mapping of subsequent fusion.2. The extraction of edge features in images and the surface approximation of point clouds are introduced. Then in the light of the difficulties in edge features extraction of point clouds and the uncertainty of defects boundary, gray moment for edge detection and least squares support vector machine are combining to realize the edge extraction of point clouds and the Fragmentary area repairing on the edge. In view of that the structure in real world of the target are complex and the actual edges are varied, through listing and analyzing the position relations, most parts of the situations are artificially divided to make sure that the proposed methods are more general.3. Through analyzing the shortage of classic ICP automatic registration method, the SIFT registration method is used to improve the efficiency of the classical algorithms:based on the separation relation, the candidate matching set of point clouds is obtained by the matching points in the images. To prevent mismatch, the RANSAC is used to inspect. Then according to the least-square principle, the initial transform parameters are calculated. Eventually the automatic ICP algorithm iteration is used to finish the registration of multi-view point clouds.4. The display principle of point clouds based on discrete points is studied, and the multiresolution shows method is proposed by using real-time interpolation. Through analyzing the visibility and re-sampling position of point clouds in the process, the MLS is used to fit the surface of local point clouds and interpolate, and the new points are textured to realize the multi-resolution display of point clouds with the real vision like images. Finally an experimental platform is developed in vc++ environment, and some operation images are depicted to validate the methods in this paper.