AI-based Road Inspection Framework Using Drones with GPS-less Navigation

The inspection of roads is an essential aspect of infrastructure maintenance in our country. Yet, conventional inspection methods often entail significant time and financial investments. Drones present an innovative and superior alternative for inspecting roads, offering swifter, safer, and more cost-efficient solutions. In this paper, we devise and deploy a low-cost framework for the inspection of roads using drones and machine learning. In our approach, we employ both an infrared (IR) camera in tandem with a high-resolution optical camera, as relying solely on optical cameras proves inadequate. While optical cameras excel in surface damage inspection of bridges and roads, IR cameras often yield valuable insights into the underlying structural issues. To enable autonomous drone navigation and the capture of images of the road structure when it identifies potential problems, our drone inspection system is outfitted with a minicomputer running sophisticated artificial intelligence (AI) algorithms. Leveraging these advanced AI algorithms, the drone autonomously performs inspection procedures without human intervention. The outcomes of these experiments demonstrated the system's capability to detect potholes with an average accuracy of 84.6% using the visible light camera and an impressive 95.1 % using the IR camera.

I. Introduction

A developed country's highway network spans thousands of centerline kilometers, comprising asphalt, concrete, or composite pavements that vary in age, condition, and performance. To effectively manage these diverse road systems, various maintenance programs have emerged. These programs aim to monitor ongoing performance, predict future conditions, aid investment planning, and identify necessary rehabilitation and maintenance measures. One such initiative is the Long-Term Pavement Performance Program (LTPP) established by the United States Department of Transportation. This active program focuses on comprehensive data collection, storage, analysis, and product development for both the United States and Canada [1]. Central to the LTPP is the assessment of pavement surface conditions, emphasizing the need for accurate measurements of distresses such as cracks, potholes, and other critical indicators [2].