Computer Vision and Pattern Recognition Technologies for Construction - Mobile and Pervasive Computing in Construction - page 191

Figure 10.2 Concrete column detection using edge and line features (Lukins and
Trucco, 2007)
information to get desired columns. Such methods can detect thin and stick-like
objects, whose texture is easily corrupted by image background (David, 2005).
Consequently, the sole reliance on edge information renders these methods
inadequate for complex scenes (Nikolaos et al ., 2006).
Zhu and Brilakis (2010a) recently combined both color/texture and boundary
information as concrete column detection cues. For each concrete column, it was
found that (1) the shape of one concrete column surface is dominated by a pair of
long vertical lines and (2) the texture and color patterns on the surface are uniform.
Based on these two findings, a concrete column detection method was proposed
(Figure 10.3). Under the method, long vertical lines are first detected using an edge
detection operator and the Hough transform. Each of the lines is then compared to
its neighboring ones. If they have similar length, they are regarded as a pair. For
each pair of vertical lines, a bounding rectangle is then constructed and its vertical/
horizontal edge ratio is calculated. If the vertical edges are significantly larger than
the horizontal, the rectangle is considered a column candidate. This rectangle's