Research On Registration Of LIDAR Point Data And Remote Sensing Images

In recent years, with the advances of related technology and the increasing needs of the community, light detecting and ranging (LIDAR) technology as a kind of real-time instrument of three-dimensional information acquisition, have been improved rapid rapidly. LiDAR system has effectively widen the range of data sources by fast collecting accurate and high-resolution digital surface models. LiDAR change the data acquiring mode from the traditional forward intersection to a continuous and automatic way, which make a progress on robotization of data processing. LiDAR data and aerial images both have advantages and limitations, and are complementary to each other. LiDAR system can acquire 3D point cloud of the target object directly but ignore semantic information (material, structures, etc.), whereas the surfaces extracted from aerial images contain abundant semantic information. On the other hand, high density point cloud acquisition by automatic matching of overlap images is still a hard problem, and depending on the matching result could lead to potential risks. Accordingly, it is a new key research in the field of photogrammetry and remote sensing that combining accurate 3D point cloud from LiDAR system and digital images to intellectualize ground object classification and extraction.The paper proposed a registration method by combination of LiDAR point-clouds data and aerial images, which overcomes limitations in dormain of photogrammetry and remote sensing, the registration has just been used among aerial images. In this thesis, we mainly focus on the registration between multi-resolution aerial images and LiDAR point-clouds data with different densities. At the same time, according to given resolution of aerial image and density of LiDAR data, we deeply study on the approaches of extracting matching metadata from two data sources for higher accurate registration. Following aspects about these methods are introduced in this paper including:The basic matching problem of LiDAR cloud data and RS images are disscussed, including making choice about matching metadata, matching models, methods and similarity determine parameters. Matching methods of Areo-born LiDAR cloud data and RS images are also disscussed from the angle of problems above when new requirements of existed methods using on this background are studied.Based on matching basic items choosing problem of Areo-born LiDAR cloud data with RS images, pre-processing methods for LiDAR clouds data such as data filtering, segmentation and classfying, are stated. Then Using the data of airborne LiDAR feature information such as 3d coordinate information, strength and echo information etc. to emphasize point clouds echo attributes based grid deleting method and extraction method based on lateral point clouds features' infinitesimal statistical of high-precision line feature matching metadata.This chapter aims to analyze the registration effects of different remote sensing image and LiDAR point clouds data using different registration transformation models, to find scale empirical formula for different registration models. However, registration results are influenced by so many factors, such as, single or multiple image matching methods, the choice of matching cell, lens distortion of the sensors and acquisition methods of images. Therefore, this chapter will first analyze the registration results by remote sensing images with different geometric resolution and LiDAR point clouds data with different density acquired by different sensors, different scales and different registration models, then scale analysis empirical formula of airborne LiDAR point clouds data and remote sensing images registration with different registration transformation models are obtained. Owing to registration scale analysis empirical formula has no existing formula to consult; this paper uses mathematical modeling approach, according to the registration results with experimental data, the introduction of position error parameters to assess registration accuracy, resolution and scale factor parameters determined by point cloud density, which are used to determine the initial scale analysis empirical model, then the mathematical model parameters are fitted based on experimental data to obtain the scale empirical formula with the minimum fitting error.The registration method of line features instead of point features for airborne LiDAR point cloud data and remote sensing image is presented, and the remote sensing images are not limited to that obtained by the same LiDAR equipment. For point clouds data with discrete features, the properties that point features can not accurately locate, high-precision line features are extracted in the point cloud data using grid excluding method of echo attributes or feature micro-element statistical method, on this basis, with any two points on the line features, the unknown parameters are introduced to represent any point on the line feature, then the combination of matching cell with image points and the line features of LiDAR point cloud data is established, and strict registration model with 2D-3D is established, the combination calculation of unknown parameters and the registration transformation parameters introduced to realize registration of airborne LiDAR point cloud data and remote sensing image. Then analysis the accuracy effects of single/multi-slice registration mode, camera lens distortion on this registration method, and validation of registration scale empirical formula of frame aerial images and airborne LiDAR point clouds data with matching cell using line features instead of point features based on collinearity equation strict registration model.