Detecção inteligente de falhas em pavimentações asfálticas com redes neurais convolucionais regionais

Abstract

In this paper we addressed the automatic road damage inspection as a Computer Vision detection problem in benefit of solutions to help smart cities im prove traffic quality and security. To do so, we considered an experimental scenario with realistic data from three different countries and four configurations of YOLO networks. When compared to related work from literature, our results have signifi cant improvements in prediction time using a lower number of parameters, yielding an experimental mAP of 0.53. We also evaluated our solution in a case study with images from Brazil that highlights several practical challenges that need to be taken into account when proposing automatic detection models for such problem.