A modified pothole detection approach to capture their width

Abstract

Detecting potholes is an ongoing problem that requires extensive investigations as potholes significantly deteriorate urban beauty and can cause serious harm to commuters. Multiple techniques have been developed over the years to address that issue. However, there has been no formal way to define the size, particularly the width, of a pothole. This work presents An's mathematical definition for the width of an arbitrary 2D pothole at a point inside. Then, the width of each pothole is calculated by taking the supremum of all width values at all points in a given subset of that pothole. We also extend the detection pipeline by Koch and Brilakis in 2011 to prove the practicality of these definitions. The obtained experimental results show that our modification improves the pothole segmentation. Consequently, it gives a more accurate estimation of pothole widths.