Quantificação de Microesferas de Vidro para Sinalização Horizontal pela Granulometria por Correlação com Pré-Processamento por Rede Convolucional

Abstract

The horizontal pavement marking is an important safety feature that must be visible both day and night. The night visibility is assessed by retroreflectivity and depends on the number of glass microspheres on the painting. Thus, knowing the quantity and distribution of microspheres in the horizontal marking are indicative of its quality. This paper proposes to use correlation-based granulometry with pre-processing by U-Net convolutional network for counting the number of microspheres present in digital images. We evaluate 10 images containing 1468 microspheres. The outcomes are promising, reaching an average hit rate of 0.9334. Keywords— road demarcation, correlation-based granulometry, U-Net, microsphere quantification.