Advancing Road Maintenance with EfficientDet-based Pothole Monitoring

Abstract

Effective road maintenance is crucial for ensuring safe and efficient transportation but is often compromised by the widespread occurrence of potholes. This study introduces a novel approach using an EfficientDet-based model for sophisticated pothole monitoring. Potholes pose a significant hazard that requires proactive detection and timely resolution. Traditional detection methods frequently fall short in terms of accuracy and real-time capability. Addressing these limitations, our research leverages the EfficientDet architecture, known for its optimal balance of accuracy and computational efficiency, to enhance the detection and monitoring of potholes. We utilized a carefully curated dataset from Kaggle, which includes 1,500 training images, 1,000 validation images, and 500 test images, encompassing a variety of real-world pothole scenarios. This diversity enables the model to generalize effectively across different conditions. Our experimental evaluations demonstrate that the EfficientDet-based model achieves an impressive average precision of 0.90 and a robust recall of 0.92, highlighting its capacity for accurate and swift pothole detection-an essential component for improving road maintenance. Moreover, we provide a comparative analysis with five contemporary pothole detection algorithms: YOLOv5, RetinaNet, CenterNet, SSD, and Faster R-CNN, among which EfficientDet consistently shows superior performance in terms of precision, recall, F1-Score, and average precision. These findings highlight the significant advancements in road safety, infrastructure management, and resource optimization. By adopting sophisticated deep learning techniques like EfficientDet, we promote a transformative improvement in road maintenance practices, paving the way for more resilient, safe, and disruptionminimized transportation networks.