Electronic Thesis and Dissertation Repository


Doctor of Philosophy


Mechanical and Materials Engineering


Prof. George K. Knopf


Over the last decade, 3D range scanning systems have improved considerably enabling the designers to capture large and complex domains such as building interiors. The captured point cloud is processed to extract specific Building Information Models, where the main research challenge is to simultaneously handle huge and cohesive point clouds representing multiple objects, occluded features and vast geometric diversity. These domain characteristics increase the data complexities and thus make it difficult to extract accurate information models from the captured point clouds.

The research work presented in this thesis improves the information extraction pipeline with the development of novel algorithms for consistent density scanning and information extraction automation for building interiors. A restricted density-based, scan planning methodology computes the number of scans to cover large linear domains while ensuring desired data density and reducing rigorous post-processing of data sets.

The research work further develops effective algorithms to transform the captured data into information models in terms of domain features (layouts), meaningful data clusters (segmented data) and specific shape attributes (occluded boundaries) having better practical utility. Initially, a direct point-based simplification and layout extraction algorithm is presented that can handle the cohesive point clouds by adaptive simplification and an accurate layout extraction approach without generating an intermediate model.

Further, three information extraction algorithms are presented that transforms point clouds into meaningful clusters. The novelty of these algorithms lies in the fact that they work directly on point clouds by exploiting their inherent characteristic. First a rapid data clustering algorithm is presented to quickly identify objects in the scanned scene using a robust hue, saturation and value (H S V) color model for better scene understanding.

A hierarchical clustering algorithm is developed to handle the vast geometric diversity ranging from planar walls to complex freeform objects. The shape adaptive parameters help to segment planar as well as complex interiors whereas combining color and geometry based segmentation criterion improves clustering reliability and identifies unique clusters from geometrically similar regions. Finally, a progressive scan line based, side-ratio constraint algorithm is presented to identify occluded boundary data points by investigating their spatial discontinuity.