technique is that it may cause actual cracks that are oriented
on the main direction of the bridge to be eliminated. However,
the inclusion of this tool was deemed acceptable as the benefit
from the number of false detections eliminated outweighed
the possibility of eliminating actual changes. Figure 5 shows
an enlarged subset of the Albuquerque (ABQ) product derived
with and without the application of this directional tool. While
GEOBIA
techniques yielded favorable results, the high computa-
tional cost and difficulty in integrating complex segmentation
techniques into a near-real time workflow currently preclude
this method from being an effective approach for the initial
assessment of cracks affecting the condition of bridge and road
surfaces. Another significant drawback of the
GEOBIA
method
is that classification is very sensitive to subtle scene variations.
Figure 4. Examples of source images (left) and supervised classification results using six classes (right). Pixel groups
classified as cracks are shown in white with all other classes in black. This method resulted in high levels of false detection.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
February 2018
81
