Segmentation and reconstruction of polyhedral buildings from aerial LIDAR point cloud
Abstract
Aerial LiDAR technology directly generates 3D coordinates of real world objects in a very short interval of time. This thesis presents a series of steps to segment and reconstruct buildings in 3D from these data. A 1D bidirectional labeling algorithm that filters the raw LiDAR data into points representing ground (terrain), and non-ground (trees, cars, buildings etc.) is first discussed. In the next step, the non-ground part of the data is segmented to represent individual buildings. A modified convex hull algorithm is introduced to determine the boundary of the point set representing each building. A hierarchical least squares process that represents the building boundary points as a series of straight lines is presented. In the third step, the building point cloud is further segmented into roof planes, using eigenvalue analysis and fuzzy unsupervised clustering techniques. Finally, 3D polyhedral building models are reconstructed using topologic constraints. The research presents a framework and its implementation to the 3D building modeling problem from a computational geometry, pattern recognition and topology perspective.
Degree
Ph.D.
Advisors
Shan, Purdue University.
Subject Area
Geographic information science|Architectural|Civil engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.