Intelligent Pavement Condition Rating System for Cycle Routes and Greenways

Syed M. Haider Shah, Waqar Shahid Qureshi, Gerard O' Dea, Ihsan Ullah

Abstract: This study introduces an intelligent framework for assessing cycling infrastructure, addressing the limitations of traditional pavement evaluation methods. At the core of the system is the CRSI, a 1-to-5 rating scale specifically designed to evaluate cycle routes based on critical factors like surface quality, vegetation encroachment, and drainage. A dataset of over 40,000 frames, extracted from videos captured using handlebar-mounted GoPro cameras and annotated by experts, forms the foundation of the system. Four deep learning (DL) models LeNet, AlexNet, EfficientNet-B2, and Swin Transformer-Tiny were trained and evaluated for Cycle Route Surface Index (CRSI) classification. Among all models, Swin Transformer-Tiny performed the best, achieving an impressive accuracy of 99.90%. To further test its robustness, we evaluated the system on four new videos, from which four separate frame sets were generated. Among these, Swin Transformer-Tiny again delivered the highest accuracy, reaching