Page 654 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

(a) (b) (c)
Figure 16. Removing piece of roof. (a) Top view of remaining points of walls; (b) and (c) are two samples of wall
segmented

After implementation of RANSAC algorithm to segment facades, all of walls are determined and are shown in Figure
17.

Figure 17. Segmented walls
3.3. Accuracy Assessments
The proposed algorithm was assessed in an urban environment where various challengeable buildings, ranging from
residential to business, located inside it. So, the method successfully detected 520 buildings properly among 533 (True
Positive and False Positive) that this shows high performance of the algorithm in the building detection. One advantage of
the proposed algorithm was that extracted no non-building objects as the building class (False Negative). Based the two Eq.
4&5, the precision and Recall calculated 98% and 99% respectively.

Precision = True Positive / (True Positive + False Positive) (4)

Recall = True Positive / (True Positive + False Negative) (5)

In order to assessing the building modelling, the Root Mean Square Error was selected that it was equal of about 0.05.

4. Discussion
From this section, we are going to assess our proposed algorithm with previous works. In terms of datasets, ALS and
MLS point clouds have been used in process of building modelling. This is because of that each of which provide some useful
information about buildings, leading to extracting and modelling of both roof and walls. But, previous studies rarely modeled
walls and roofs together and just focused on one of them. In a similar way, there are examples of researches which just
detected buildings and do not proposed the model of buildings components.

9