Evaluación de daños en pavimento flexible usando fotogrametría terrestre y redes neuronales

In Colombia, road deterioration is assessed by means of road inventories and visual inspections. For this assessment, the Instituto Nacional de Vías (Colombia's National Road Institute) (abbreviated INVIAS in Spanish) uses the Vision Inspection de Zones et Itinéraires Á Risque (VIZIR) and Pavement Index Condition (PCI) methods. These two methods serve to determine the severity of damages in flexible and rigid pavements. However, they can be tedious and subjective and require an experienced evaluator, hence the need to develop new methods for road condition assessment. In this paper, we present a methodology to evaluate flexible pavement deterioration using terrestrial photogrammetry techniques and neural networks. The proposed methodology consists of six stages: (i) image capture, (ii) image preprocessing, (iii) segmentation via edge detection techniques, (iv) characteristic extraction, (v) classification using neural networks, and (vi) assessment of deteriorated areas. It is verified using real images of three different pavement distresses: longitudinal cracking, crocodile cracking, and pothole. As classifier, we use a multilayer neural network with a (12 12 3) configuration and trained using the Levenberg–Marquardt algorithm for backpropagation. The results show a classifier’s accuracy of 96 %, a sensitivity of 93.33 %, and a Cohen's Kappa coefficient of 93.67 %. Thus, our proposed methodology could pave the way for the development of an automated system to assess road deterioration, which may, in turn, reduce time and costs when designing road infrastructure maintenance plans.