Airborne LiDAR Point Cloud Data Processing And 3D Building Reconstruction

The airborne LiDAR (Light Detection and Ranging) is a kind of new promising technique in obtaining instantly 3D accurate information of the earth surface and objects. It represents a collection of laser technology, dynamic pose measurement technology and high precision differential GPS technology, and it can survey topographic information more detailed and accurate than those obtained by traditional photogrammetric methods. However, the data set obtained by LiDAR system is 3D discrete sub-randomly spatial distributed point cloud that represents points from both the ground surface and objects found on the ground surface. The research on data processing and application of LiDAR point cloud falls obviously behind the development of LiDAR hardware system. At present, how to process LiDAR point cloud data to extract topographic information and different object information from point cloud is a key problem of LiDAR research. The application research of LiDAR data in many fields such as topographic mapping, urban construction and forestry programming and so on is the active studying topic too. Therefore, this dissertation develops the research on airborne LiDAR point cloud data processing and 3D building reconstruction.In this thesis, the characteristics of airborne LiDAR system and data obtained by system are analyzed firstly. Then, for different typical objects, points'spatial distributed geometrical characteristics, intensity characteristics, echo characteristics and spectral characteristics are studied. Based on the above characteristics, this paper puts forward a series of filtering and classification methods of LiDAR point cloud data. And then, application research of building extraction and construction from LiDAR data is explored. The main contributions of this dissertation are as following:Firstly, TIN (Triangulated Irregular Network) filtering algorithm based on height jump and TIN filtering algorithm based on neighboring height difference and assistant plane are presented. In the TIN model of LiDAR point cloud, there exist some height jump rules between different typical object points and neighboring points. Through parameters of height difference and neighbor number, the TIN filtering algorithm based on height jump can separate non-ground points and ground points gradually. This method has better effect on filtering LiDAR ground point cloud and retaining continuous undulate topography at the same time. After TIN filtering algorithm based on height jump, TIN filtering algorithm based on neighboring height difference and assistant plane filters the non-ground points further. This algorithm projects the filtered ground points into assistant plane. The points on assistant plane and non-ground points are together constructed TIN model. By setting neighboring height difference parameter, the building point cloud can be extracted from non-ground point cloud.Secondly, a slope-based planar-fitting filtering algorithm of LiDAR data is proposed. The TIN model of original discrete LiDAR point cloud is built firstly. In TIN model, some points with similar slope can be regarded as they are in the same approximate plane. According to the parameter of slope, some point sets of approximate horizontal plane, inclined plane, vertical plane and continuous smooth curved surface can be filtered using the region growing method. This algorithm has preferable effects in feature extraction of continuous ground, building tops and building vertical walls and so on. Moreover, the building tops which are higher than surrounding points can be separated from many extracted planar point sets according to height difference.Thirdly, filtering and classification methods of LiDAR data based on echo intensity and spectral information are discussed. The process of compositing LiDAR point cloud data and spectral data is called SILC (Spectral Imagery LiDAR Composite). The spectral values of different qualitative objects are calibrated from spectral image data by supervised classification method. Using calibrated spectral value to classify the SILC data, the vegetation point cloud can be filtered effectively. Echo intensity data represents the laser reflectance characteristic of different qualitative objects. Influencing facors of echo intensity are the foundation of echo intensity application. Through clustering laser echo intensity value of LiDAR data using K-mean clustering method, different qualitative objects with distinct different reflectance characteristic can be separated.Fourthly, methods of building extraction and 3D reconstruction from LiDAR data are explored. From the above proposed filtering and classification approaches of LiDAR data in this thesis, the process of building extraction is summarized. Based on the extracted building top point cloud, a method of 3D building reconstruction from LiDAR data is presented. The process of this method includes clustering building top point cloud, fitting building top plane, deciding building outer boundary and every planar boundary, and calculating 3D coordinate of every corner point parameter to reconstruct 3D building model. This method is able to reconstruct effectively not only simple building model but also complicated irregular planar building model.This dissertation studies on the airborne LiDAR point cloud data processing and 3D building reconstruction. It has practical and theoretical significance on LiDAR point cloud data processing and application.