|Ph.D Student||Avrahami Yair|
|Subject||Automation in Roof Extraction within a Digital|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Emeritus Yerach Doytsher|
|Dr. Yuri Raizman|
|Full Thesis text - in Hebrew|
In this research, an approach and detailed algorithms for extracting roofs were developed. The approach assumes that most roofs are composed of several spatial polygons and can be obtained by extracting all these polygons or when the roof’s model is known by obtaining only some of them. In order to enable automation in roofs extraction, algorithms for spatial polygon extraction were developed. The major part of the algorithms deal with the matching problem between polygons, extracted from two images. Enlarging these algorithms led to the development of two detailed algorithms. The first group of algorithms focus on mapping from panchromatic aerial images taken under “difficult” conditions. First, the operator needs to point, in the left image, within a typical radiometric area of the roof. The next stages include: extracting the bounding polygon in the left image, calculating the approximated height and transforming the manual pointer to the right image, extracting the bounding polygon in the right image, simplifying the polygons, extracting the precise height and final mapping. The algorithms’ efficiency was tested and based on images at scale of 1:42,000 and scanned at 14µm, the obtained results are satisfactory with an accuracy of 1.08m and 0.52m for the planimetric and altimetric measurements, accordingly.
The second group of algorithms focus on a semi-automatic interactive model-based approach for automation in modular roofs extraction. The approach is composed of a manual and an automatic phase. In the manual, the operator chooses an Expanded Parameterized Model from predefined library models and an Interactive Option for Extraction of the roof. Then, the operator needs to point according to the guidelines of the chosen option. In the automatic phase, the selected spatial polygons are extracted, the parameters of the model are calculated and the roof is reconstructed. The algorithms’ efficiency was tested and based on colored images at a scale of 1:5,200 and scanned at 14µm, the obtained results are satisfactory with an accuracy of 0.28m and 0.31m for the planimetric and altimetric measurements, accordingly.
The results show that the approach can be implemented on many types of roofs and under diverse photographic conditions.
The improvement of both algorithms is expressed by three aspects: Decreasing the number of the measurements, an improvement of the measurement process speed and simplifying the measuring environment. The tests showed that the algorithms are efficiently and a cost-effective solution in comparison to the traditional photogrammetric process, and achieve standard accuracy requirements.