A vision-based approach for autonomous crack width measurement with flexible kernel

Abstract A quantitative description of cracks is the primary key for examining the safety of structures. Vision-based crack detection has been recently developed to avoid visual inspection, which demands massive labor. For the analysis of a crack based on digital images, several width measurement methods have been investigated. However, the verification of the methods was carried out with irregular surfaces of which the ground truths of the widths were subjective. In this study, a vision-based approach for crack width measurement is proposed, implemented and validated by comparison of the widths which are numerically calculated and analytically featured, respectively. The main skeleton of cracks is extracted by pruning based on the partition of the boundaries of the cracks. A flexible kernel is proposed to calculate the propagation direction of the cracks, and its angular resolution is investigated in terms of the size of the kernel. For calculating the angular resolution, an algorithm for generating the possible skeleton patterns for crack propagation is developed. The verification of the width measurement method shows the existence of the proper range for the coefficient used in determining the size of the kernel. After the verification, the method is implemented for the crack width measurement using the coefficient range selected based on the results of the verification. The flexible kernel with the coefficient in the range provides improved results compared to constant kernels or kernels with the coefficients out of the range.