Automated Extraction Of Buildings And Facades Reconstructon From Mobile LiDAR Point Clouds

Laser scanning or light detection and ranging (LiDAR) provides an efficient solution for capturing spatial data in a fast, efficient, and highly reproducible way. It has been widely used in many fields, such as cultural heritage documentation, reverse engineering, three-dimensional (3D) object reconstruction, and digital elevation model (DEM) generation, as it can directly obtain the3D coordinates of objects. LiDAR can be divided into three categories, namely, airborne LiDAR, terrestrial LiDAR, and mobile LiDAR. Airborne LiDAR has been successfully used for digital elevation model (DEM) generation and reconstruction of building roofs. However, it has difficulties for capturing points of the facades of buildings. As mobile mapping technology has made a great progress, mobile LiDAR allows the rapid and cost-effective capturing of3D data from large street scenes including the dense points of building facades.In recent years, the processing of mobile LiDAR data has been focused mainly on objects extraction and facades reconstruction. To extract street-scene objects or detailed features of building facades, mobile LiDAR point clouds need to be classified into different categories (e.g., buildings, trees), which is a key step for accurate identification and3D reconstruction of street-scene objects. Moreover, facades reconstruction requires the detailed features of building facades like windows and facade footprints to be automatically recognized.On the one hand, mobile LiDAR systems capture high-accuracy, high-density points both at accuracies and resolutions, which beyonds those availablities through aerial Photogrammetry, and when using the terrestrial LiDAR is impractical. However, compared with advances in mobile LiDAR systems, automated algorithms and software tools for efficiently extracting3D street-scene objects of interest from mobile LiDAR point clouds rather fall behind, due to huge data volumes and complexity of urban street scenes, as well as the presence of occlusion. On the other hand, different from the approaches for airborne LiDAR data processing, methods for processing mobile LiDAR data have to deal with fully3D point clouds. Due to the non-unique correspondence between (X, Y) coordinates and Z coordinate, the algorithms for filtering and classifying airborne LiDAR data, such as triangulated irregular network (TIN)-based filtering method, are difficult in handling with the mobile LiDAR data because of data dimensionalities.This dissertation aims to fulfill two tasks, namely, the automated classification and buildings recognition from mobile LiDAR point clouds, the reconstruction of building facades with detailed features like facade footprints and windows from the point clouds of buildings.This dissertation proposes a novel method to generate the georeferenced feature image of mobile LiDAR data, which represents the spatial distribution of scanning points and preserves the local geometric features of street-scene objects. Then the approach based on the georeferenced feature image was proposed for automated extraction of buildings from mobile LiDAR data. The proposed approach consists of two steps:coarse classification of buildings in image space using image segmentation and contour extraction, accurate identification in3D space adopting profile analysis and eigenvalues analysis.Secondly, this dissertation focuses on facade footprints extraction, windows extraction from building facades, and geometric reconstruction of wire-frame building models. A coarse-to-fine approach using RANSAC planar segmentation was firstly proposed to automatically extract the facade footprints of building point clouds. The geometric features and semantic description of different planar patches of buildings were elaborated to distinguish facade walls from segmented planar patches. To extract detailed features from building facades, this dissertation presents a method combining facade raster images and facade TIN models to automatically extract rectangle windows in facade walls. Finally, the topology between different planar patches like facade walls and roofs were built to reconstruct the geometric wire-frame models of buildings.The proposed method of the georeferenced feature image generation transforms the extraction of street-scene objects like buildings from3D mobile LiDAR point clouds into the geometric features extraction from2D imagery space, thus simplifies the automated building extraction process. Four datasets captured by Optech’s Lynx Mobile Mapper system were selected for assessing the performance of the proposed methods of buildings recognition and facades reconstruction. Experimental results demonstrate that the proposed methods provide promising solutions for automatically extracting street-scene buildings and building facade details like windows and footprints from mobile LiDAR point clouds.