Page 648 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

2.1. Pre-Processing
2.1.1 Sectioning: Point cloud data is divided to smaller sections, due to reduce entire processing time. All of divided sections are
x
equal and square. The area along X and Y axis are divided into a N × N grid using minimum and maximum value of ( ) . Every
y
single cell is called a section and there is an overlap between sections which are next to each other. Since the building border
may fall on the edge of section, the overlap is considered in order to not losing any information about that building. Here about
5 meters is considered for overlapping range. After sectioning, processing is implemented on each individual section (Figure 3).

Figure 3. Presentation of sectioning step

2.1.2 Noise Removal: As shown in Figure 4, according to elevation bar, outlier points have abnormal elevation compared to
inlier points and illustrate with green color. After dividing the data to smaller and equal section, to remove noisy points which
have abnormal elevation, Sparse Outlier Removal method is utilized [17].

Figure 4. Recognition of noisy points with abnormal elevation

Utilization the Sparse Outlier Removal method, k-nearest neighbor (k-NN) points are chosen around every point. After
that, mean ( ) and standard deviation ( ) are calculated for elevation reorganization (z) for noisy points according to Equation
(1) and (2) respectively. At the end, points whose elevation is within a certain range of the standard deviation (μ ± αS where α
is a coefficient for increasing effect of standard deviation) are determined as not noisy (NR) points, respect to Equation (3). Then,
all other points are deleted.

3