Road Deterioration detection: A Machine Learning-Based System for Automated Pavement Crack Identification and Analysis

Deterioration of road surfaces, influenced by external factors such as vehicular traffic intensity, adverse climatic conditions, and suboptimal design, manifests in defects including potholes, distress, and irregularities. This study endeavors to design, construct, and assess an automated framework to detect and analyze pavement distress utilizing machine learning methodologies. Specifically, the investigation delves into the adoption of object detection paradigms to discern and classify diverse pavement crack patterns. Critical focal areas of the study encompass the procurement of trustworthy datasets, the refinement of object detection algorithms tailored for pavement distress identification, intricate crack typology classification, and a rigorous assessment of their severity and dimensions. A diverse array of pavement distress imagery was sourced from both publicly accessible repositories and firsthand mobile device captures. Several object detections models, notably the You Only Look Once version 5 (YOLOv5), the avant-garde You Only Look Once version 8 (YOLOv8) introduced in 2023, and CenterNet, were rigorously trained and assessed with the amassed dataset. Performance evaluation metrics underscored the YOLOv5 model's superiority over CenterNet. Uniquely, this research is distinguished by its groundbreaking use of the YOLOv8 model for pavement crack detection, thereby pioneering the application of this state-of-the-art model in such domains.