KC, Kiran (University of Sydney), Worrall, Stewart (University of Sydney), Berrio Perez, Julie Stephany (University of Sydney), Balamurali, Mehala (University of Sydney)

CyberPotholes: Pothole defect detection using synthetic depth-maps

Scheduled for presentation during the Regular Session "Synthetic datasets in perception" (ThBT4), Thursday, September 26, 2024, 15:10−15:30, Salon 7

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Road infrastructure is vital globally but increasingly threatened by pothole defects, due primarily to extreme weather driven by climate change. New technologies like depth imaging cameras and Machine Learning (ML) object detection offer promising solutions for improved road surveying and maintenance. However, reliable ML methods need large training datasets. This study addresses this challenge by proposing a novel, automated pipeline for generating labelled synthetic multiclass pothole depth-map datasets, simulating data captured by a depth camera. This pipeline allows the creation of a large dataset, CyberPotholes, specifically designed to benchmark ML object detection techniques for pothole detection. Benchmark evaluations of the pipeline and an ML model trained on a smaller dataset yielded positive results, highlighting the potential of this approach for future development in automated road inspection systems.