Automatic Road Roughness Detection and Ranking using Deep Learning and Computer Vision

Abstract

Road roughness is considered one of the most important problems that government institutions face because it requires many complex issues to find the roughness of the street. It also requires a lot of expensive tools which, in turn, measure the roughness of the roads. so, in this research paper we create a new model study road roughness and rank the roughness of this road automatically without the need for any cost or human intervention. Our proposed model checks the roughness by capturing the imaging using a drone, then it processes and analyzes the images coming from the drone, using several models that work together. our model shows the pattern of roads from the captured image using Gray level Size Zone Matrix(GLSZM) features Zone Percentage (ZP) and Size Zone Non-Uniformity (SZN) and then take the spikes of its distributions then take these spike to get optimal value K for K-mean to segment the image, the result of first model enter to second model that make sorting for this images depending on GLSZM features (ZP and SZV) to improve the result of our model, after that the image enter to CNN to get the outcomes by classifying it into which category this roughness belongs. The best accuracy we achieved in our model reached 91.94%, which is a very high accuracy, and therefore by a large percentage all correctly captured images from the drone has accurate results.