Continuous crack detection using the combination of dynamic mode decomposition and connected component-based filtering method

Concrete surface cracks are one of the earliest indicators used to assess structural damage, serviceability and durability. This study proposes a novel method to detect concrete cracks and evaluate the crack features. The proposed approach introduces the dynamic mode decomposition (DMD) algorithm to identify individual cracks in each frame of a video based on the analysis of the DMD spectrum. Since cracks contain different types of noise, a connected component-based filtering method is proposed to discard the noise of the bubbles and blobs. The inspected cracks can be used to evaluate crack features, including area, orientation and perimeter. The proposed crack detection approach is then applied to inspect the crack development during a cubic concrete sample test under axial loading. The detected results demonstrate that the proposed method can correctly identify most cracks. Because video frames generate enormous data volumes that are not conducive to quick detection, dimensionality reduction based on normalized singular values is also discussed and compared with the proposed method. After the cracks are identified, the crack features of the area, orientation and perimeter are determined over time, which helps to assess crack properties and to understand structural performance. The accumulated database of crack characteristics will provide useful information of lifetime predictions.