Automatic analysis of road pavement surface imagery
- Oliveira, H.O.; Correia, P.L.;
"Automatic analysis of road pavement surface imagery
International Road Federation - World Meeting - IRF2010
This paper proposes an automatic analysis system capable of processing imagery acquired using a laser road imaging system, aiming to support the visual distress data analysis made by an inspector. The analysis consists of two main steps: crack regions detection and crack type classification. The first step processes images (of size 4015 x 3786 pixels), covering one entire lane, to identify those regions containing crack pixels. For this purpose, an image processing technique based on dynamic thresholding and region binary entropy computation is used. The classification step resorts to a connected components algorithm to indentify crack ‘objects’, which are then classified using a pattern recognition system developed for this purpose, according to a subset of the crack distress types identified in the Portuguese Distress Catalogue. Finally, images are labeled as containing longitudinal, transversal, miscellaneous or no cracks. The proposed automatic system is evaluated over an image database composed of real flexible pavement surface images acquired during a survey, using a set of well-know statistical pattern recognition metrics and exploiting the availability of ground truth data manually provided (human labeling) for the entire image database. Promising results are obtained in both crack detection and classification.