Author(s): Rosa, Gonçalo J.M. ; Afonso, João M.S. ; Gaspar, Pedro Dinis ; Soares, Vasco N.G.J. ; Caldeira, J.M.L.P.
Date: 2024
Persistent ID: http://hdl.handle.net/10400.11/9081
Origin: Repositório Científico do Instituto Politécnico de Castelo Branco
Subject(s): Pedestrian crossings; Smart cities; Computer vision; Convolutional neural networks; Performance evaluation
Crosswalks play a fundamental role in road safety. However, over time, many suffer wear and tear that makes them difficult to see. This project presents a solution based on the use of computer vision techniques for identifying and classifying the level of wear on crosswalks. The proposed system uses a convolutional neural network (CNN) to analyze images of crosswalks, determining their wear status. The design includes a prototype system mounted on a vehicle, equipped with cameras and processing units to collect and analyze data in real time as the vehicle traverses traffic routes. The collected data are then transmitted to a web application for further analysis and reporting. The prototype was validated through extensive tests in a real urban environment, comparing its assessments with manual inspections conducted by experts. Results from these tests showed that the system could accurately classify crosswalk wear with a high degree of accuracy, demonstrating its potential for aiding maintenance authorities in efficiently prioritizing interventions.
