Abstract

Pavement cracks are a major form of road infrastructure degradation, necessitating efficient and accurate detection for timely maintenance. Existing inspection methods rely either on labor-intensive manual surveys or automated systems constrained by high hardware costs and GPS dependency, limiting their flexibility for continuous surface assessment. This paper introduces a dual-channel crack detection model that integrates continuous pavement texture reconstruction with deep segmentation and high-precision boundary refinement algorithms, enabling on-site implementation and accuracy enhancement for longitudinal crack detection. A feature-based image stitching algorithm is developed to reconstruct continuous pavement textures from high-resolution images, enabling GPS-free crack localization. The proposed method further combines the YOLOv8-seg model with adaptive morphological operations to achieve pixel-level crack reconstruction. Comparative experiments reveal that the hybrid approach achieves superior segmentation performance with finer boundary delineation and improved branch recovery compared to the baseline model. The paper provides a practical solution for automated inspection and high-fidelity reconstruction of longitudinal cracks, effectively supporting pavement maintenance planning.

Download paper here

Download preprint here

This work was supported by the National Key Research and Development Program of China (grant number 2022YFC3801100); and Shenzhen Science and Technology Program (grant number SGDX 20240115110503006).

Financial Sources: