Answers are grounded in this paper and cite their sources.
Ensuring road quality is essential for traffic safety, vehicle longevity, and passenger comfort. Recently, cost-effective road quality measurement systems based on vehicle vibrations have emerged, often relying on supervised machine learning models. However, these models require high-quality labeled data, which is typically unavailable at the necessary resolution. Manual labeling is impractical due to the large data volume, leading to lower spatial resolution and inflexible datasets that cannot easily adapt to evolving requirements. This study presents a novel Machine In the Loop (MIL) annotation method based on unsupervised learning for automated pavement quality assessment, an area where previously only manual labeling methods were used. The proposed approach utilizes vehicle-mounted sensors to collect vibration data, which is then processed through feature extraction, frequency-domain transformations, and statistical segmentation. A Gaussian Mixture Model clustering algorithm categorizes road sections by quality, significantly reducing reliance on manual labeling. To assess the effectiveness of MIL annotation, a supervised Random Forest classifier is trained on the labeled dataset, achieving a classification accuracy of 91%. The methodology is validated through extensive real-world testing across diverse road conditions using multiple vehicle types. The results demonstrate that the novel MIL annotation with a subsequent supervised learning approach enables scalable, cost-effective, and high-resolution road quality assessment with minimal human intervention. This study highlights the feasibility of integrating unsupervised learning-based annotation into road management workflows, offering a flexible and adaptive solution for infrastructure monitoring and maintenance planning. • MIL annotation automates pavement assessment, reducing manual labeling needs. • IMU sensor data is processed via feature extraction and statistical segmentation. • GMM clustering enables high-resolution, adaptable road quality categorization. • A Random Forest classifier trained on MIL data achieves 91% classification accuracy. • The method is scalable, cost-effective, and validated across diverse road conditions.
No takes yet. Share an insight, caveat, or question.
Nagy et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: