Supervised strategies for cracks detection in images of road pavement flexible surfaces

The detection of cracks and other degradations in road pavement surfaces is traditionally done by experts using visual inspection, while driving along the surveyed road. An automatic cracks detection system based on road pavement images, as proposed here, can speed up the process and reduce results' subjectivity. The paper confronts six supervised classification strategies, three parametric and three non-parametric. The analysis is done on data resulting from dividing the image into a set of non-overlapping windows. Dealing with supervised classification strategies, a technique for automatic selection of training images from an image database is proposed as the initial step, after which a human expert should select the image windows containing cracks. The selected classification strategies work with a 2D feature space. Classifiers are evaluated using a set of well-know metrics, indicating that a better performance can be achieved using parametric classification strategies.