|M.Sc Student||Zeibak-Shini Reem|
|Subject||Change Detection via Terrestrial Laser Scanning|
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Sagi Filin|
|Full Thesis text|
Detection of changes has been a subject for research for many years, seeing applications
such as motion tracking, inventory-like comparison and deformation analysis as only a few examples. The emergence of terrestrial laser scanning as a tool for dense and accurate 3D mapping has also seen a growing number of applications in the context of change detection. These usually refer either to deformation analysis or to evaluation of differences between epochs. The methodologies that have been devised to facilitate those evaluations were either localized and applied to well-defined objects, or were based on some limiting assumptions such as enforcing similar scanner position between epoches.
This research presents a model for the detection of changes via terrestrial laser scanning.
The proposed model addresses such matters as varying scale across the scene, occlusion, and laser scanning artifacts. Making use of the relative 3D rigid-body-transformation relation between datasets and utilizing a polar data representation, the "change detection question" can be reformulated into a "point visibility" problem. This reformulation provides direct solution to problems related to occlusion, multi-scale analysis between and within scans, and varying scanner positions. It, therefore, enables to simultaneously detect major and minor changes in different ranges. The research then analyzes the implication of, ranging errors, scanning related artifacts and other inaccuracies related to the scanning geometry, on the detection of changes. These extensions are made while maintaining high computational efficiency of the core model.
Awareness to artifacts and error sources within and between scans allows providing a more thorough understanding about the limit of detection, and so distinguishing actual changes from ones relating to the acquisition system and pose parameters. This realization, therefore, leads to a more reliable and robust detection procedure. Among the different error sources, we focus on the effect of objects geometry, objects boundary, and their location and orientation relatively to the scanner. While similar analyses, relating to ranging uncertainty have been addressed in the past, here the association between different scans must come into effect. The research studies therefore how errors originating from individual scans affect the detection of mutual changes.
The model is tested on different scans, where changes appear in different size and resolution, either in urban areas or within open areas where objects are mostly unstructured. Results show that minute details can be identified by applying the proposed approach. They also demonstrate the important role of the scanner-aware processing steps to substantially increase the model reliability.