Page 653 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

In the building extraction, buildings were detected in data by using DSM that is shown in Figure 13-a. Due to the
georeferencing of DSM and point clouds data, after specifying buildings on DSM, location of buildings was recognized on
point clouds. Figure 13-b shows buildings model.

(a) (b)
Figure 13. (a) DSM of a sample of created section; (b) Buildings model of a sample of created section

We used ALS point clouds to identify buildings from all objects and just could achieve information about roof of
buildings. For achieving information about facades MLS point clouds had been utilized. Figure 14-a shows a sample building
with white color in point clouds data. Collecting data from facades of a building is shown in Figure 14-b.

(a) (b)
Figure 14. Selecting data from MLS point clouds. (a) A part of the created section; (b) Facades of a building

For removing piece of roof data from MLS point cloud an elevation feature was utilized and regard to graph in Figure
15-c best threshold was selected.

(a) (b) (c)
Figure 15. (a) Top view of building; (b) Implementation of elevation feature; (c) Graph of elevation feature to select
threshold

Remaining points of facades are shown in Figure 16-a. Ransac algorithm was used to segment facades. By using this
algorithm each wall was determined. Figure 16-b and Figure 16-c shows two sample of walls which were determined by
Ransac algorithm.
8