Efficient Algorithm for Crack Detection in Sewer Images from Closed-Circuit Television Inspections

Abstract

This paper presents a new algorithm for automated crack detection in sewer inspection closed-circuit television (CCTV) images. Cracks often have a long and thin rectangular shape with a darker appearance relative to other components in the image; therefore, they typically manifest as edges. The proposed algorithm exploits previous information on the visual characteristics of crack features in typical CCTV images to efficiently identify actual cracks and filter out background noise. The algorithm consists of three main steps. The first preprocessing step prepares the CCTV image for crack detection by identifying a set of candidate crack fragments using the Sobel method to detect horizontal and vertical edges separately. The Hough transform is then used to identify and remove the edges associated with information labels typically found in CCTV images. The second step applies a set of morphological operations to enhance candidate crack segments by filling the gaps between closely adjacent and aligned edges. The enhancement step results in merging crack fragments that potentially represent segments of the same crack curve. In the third step, two filters are defined based on previous knowledge of the visual characteristics of cracks, and then applied to remove noise edges and extract a set of real crack segments. We tested the proposed algorithm on a set of CCTV videos obtained from the cities of Regina and Calgary in Canada. The experimental results demonstrated the efficiency of the proposed algorithm, and showed its robustness in detecting various patterns of sewer cracks.