Home // SMART 2025, The Fourteenth International Conference on Smart Cities, Systems, Devices and Technologies // View article

Smart City Road Maintenance: A LiDAR and AI-Driven Approach for Detecting and Mapping Road Defects

Authors:
Giovanni Nardini
Roberto Nucera
Alessandro Ulleri
Stefano Cordiner
Eugenio Martinelli
Arianna Mencattini
Iulian Gabriel Coltea

Keywords: cities; road maintenance; LiDAR; AI; computer vision.

Abstract:
This work is focused on the development of an integrated system designed to detect, map, and analyze road surface defects, contributing to Smart City infrastructure maintenance. The system is installed on vehicles and leverages a multi-sensor approach, combining Light Detection and Range (LiDAR) point clouds, visual information from Red-Green-Blue (RGB) cameras, inertial data and Global Navigation Satellite Systems (GNSS) coordinates. Road defects such as potholes and alligator cracks are detected in RGB images by a custom deep learning model based on instance segmentation. The scene understanding is committed to a second Artificial Intelligence (AI) model based on semantic segmentation in order to perceive objects locations and the overall structure of the road. Afterward, all results are processed together and translated into the 3D domain of LiDAR data. This can be done through a proper camera calibration procedure and LiDARCamera data alignment with the estimation of intrinsic and extrinsic parameters. Then, AI segmentation results are projected to 3D point clouds in order to isolate the detected items from the rest of the point cloud and obtain three-dimensional models of each of them, enabling measurements like the affected surface extension, depth and volumes. GNSS and inertial data are fused together to obtain the correct orientation and location of the system, enabling geographic positioning of all detected items on the map. Results are displayed on a map-based portal, enabling easy access to near real-time defect data. This approach advances road monitoring by automating the mapping and analysis of surface conditions, enhancing urban infrastructure management. In addition, the strengths of this approach are the possibility of deploying the pipeline in edge devices enabling real-time computation, the use of pre-existing training datasets based on RGB images alone, and good accuracy on the geographical localization and estimation of defect measurements.

Pages: 1 to 4

Copyright: Copyright (c) IARIA, 2025

Publication date: April 6, 2025

Published in: conference

ISSN: 2308-3727

ISBN: 978-1-68558-251-7

Location: Valencia, Spain

Dates: from April 6, 2025 to April 10, 2025