UAV-based Localization of Removable Urban Pavement Elements Through Deep Object Detection Methods

Abstract

We introduce a deep learning framework leveraging YOLOv8 architecture to automate the localization of Removable Urban Pavements (RUPs) using UAV imagery. The core idea behind RUPs is to provide pavements that can be quickly opened and closed using lightweight on-site equipment. This approach aims to efficiently restore the street’s original appearance and functionalities within a short timeframe, typically just a few hours. Our study explores the feasibility of autonomously localizing RUP elements, paving the way for robotic-driven replacement with prefabricated, fully functional components. Moreover, the integration of UAV data enhances safety and accessibility to challenging areas. Experimental results underscore the efficacy of our approach in achieving precise localization and thereby enabling proactive maintenance efforts.