Hybrid automated road crack segmentation using morphological operations and boundary tracing

Abstract

Cracks in the road surface are one of the early indicators of structural damage that has an impact on safety and infrastructure maintenance costs. Accurate early detection is a challenge in complex visual conditions such as uneven lighting and varied asphalt textures. This study proposes an efficient and fully automated hybrid segmentation method to detect cracks in road surface imagery. This method consists of several main stages: image enhancement using contrast limited adaptive histogram equalization (CLAHE), initial segmentation through a combination of Otsu's thresholding, adaptive Gaussian thresholding, and Canny edge detection, followed by mask enhancement with morphological operations (closing, opening, and erosion). The DeepCrack dataset is used as a source of test data. The evaluation results showed high performance with detection accuracy reaching 95.82%. These findings show that the proposed method is not only precise and sensitive, but also adaptive to visual variation without the need for manual training or parameters. A major novelty lies in the integration of three classic segmentation methods in one morphology pipeline that is computationally lightweight yet competitive, making it potential for real-world applications of automated inspection systems.