Metodologia para detecção e georreferenciamento de placas de sinalização em vias rurais

Abstract

The use of the methodology for detection and georeferencing of road signs in rural roads aims to improve the management of road assets, especially the road signs. We trained the YOLO (You Only Look Once) method for detection and georeferencing of vertical traffic signs in videos generated from rural roadways. In this study, the objective was to identify the vertical traffic signs. For that, we applied through the training of convolutional neural networks (CNN) the georeferenced footage. The refinement of the network parameters was effectively defined: we used a dataset called “normal-brazilian-traffic-signs_dataset”; and, from that, we were able to identify and georeference the signs present in the analyzed videos. It is important to emphasize that the classification and accuracy depend on parameters, such as the definition of the confidence level. We observed that the accuracy increases significantly with the increase we adjusted the number of training epochs; among others. This enabled the automation of the process of identifying the signs from the videos of the filming of the roads. Thus, this study contributes to the decision making related to road asset management regarding signage and road safety.