|Ph.D Student||Abo Akel Nizar|
|Subject||Automatic Building Extraction Using LiDAR Data|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Emeritus Yerach Doytsher|
|Professor Sagi Filin|
|Full Thesis text|
The reconstruction of buildings from points cloud offers promising prospects for rapid generation of large-scale 3D models, e.g., for city modeling. Such reconstruction requires knowledge on a variety of parameters that refer both to the points cloud and to the modeled building. The separation of the laser points describing the buildings from the rest of the data is one concern; the generation of a building model is another one. From the subset, one needs to learn the roof parts, and then convert them into an actual building model that complies with topological and geometrical rules.
The complexity of this task has led many researchers to use external information, mostly in the form of detailed ground plans to detect the subset of the points cloud and to provide first approximation of the building shape. This information is however not available everywhere and generally cannot be taken for granted. Moreover, most of the existing researches follow the assumption that buildings consist only of planar parts. This assumption makes limitations in reconstructing complex buildings specifically when they consist of curved surfaces.
In this work, a new reconstruction model that autonomously detects buildings within point clouds and reconstructs their shape is presented. Its main features are:
? An innovative filtering strategy that was developed to classify the laser data into: terrain and off-terrain.
? Safeguards that were developed to reduce the chance of misclassification within the building detection phase to a minimum.
? Reconstruction which involves aggregation of point sets into individual faces, and inference on the building shapes from these aggregates.
? Imposition of geometric constraints on the reconstruction to generate realistic models of the buildings.
? Extension of common models to support general shape, curved surfaces, reconstruction and representation of these using Non-Uniform Rational B-Spline (NURBS) surfaces.
Results show that high levels of detection and reconstruction of actual building shapes is achieved at medium point densities and even higher for denser ones.