Automated Road Crack Detection: A Comparative Analysis of Edge Detection Techniques for Enhanced Accuracy

Abstract

In the modern era, road networks are vital conduits connecting diverse regions, thus necessitating continuous maintenance to ensure efficient transportation. Surface distresses, notably cracks, pose a ubiquitous challenge, affecting road safety and longevity. To address this, automated road crack detection has gained momentum, driven by image processing techniques and deep learning. This study investigates the impact of various image processing methods on road crack detection, aiming to augment conventional human visual inspection. Geometric model-based and deep learning approaches represent two prominent paradigms for crack detection. Geometric model-based techniques employ fundamental image processes, including segmentation and morphological operations, enhancing detection reliability through pre-processing. Thresholding and filtering-based methods, though widely used, tend to neglect critical geometric and photometric characteristics of cracks. However, when combined with geometric models, they yield more accurate results. Image preprocessing, involving grayscale conversion and Canny edge detection, is a critical initial step in the crack detection process. The embedded Gaussian filter mitigates Canny’s noise sensitivity. Empirical analysis compared edge detection methods, with Roberts emerging as the most effective. Further improvements were proposed, including morphological operations. This study underscores the significance of image processing techniques and edge detection methods in automating road crack detection. The proposed modification offers a promising avenue for enhanced accuracy in detecting road cracks, contributing to developing efficient road monitoring systems. These advancements are vital for maintaining safer and more durable road networks, with potential applications in transportation infrastructure management.