By Nidhi DhullReviewed by Susha Cheriyedath, M.Sc.Nov 4 2024A recent article published in the Alexandria Engineering Journal proposed using monocular vision and image-enhanced fractal science based on computer vision and image processing techniques for non-contact detection of underwater concrete cracks. A four-level structural health condition was established to measure underwater cracks and assess safety.
Study: Application of computer vision techniques to damage detection in underwater concrete structures. Image Credit: Erin Donalson/Shutterstock.com
Background
Concrete is widely used in underwater engineering. However, cracks in such structures can significantly alter stress distribution, thus destroying the structures’. Therefore, detecting and repairing cracks is a major issue in concrete engineering construction.
Current underwater concrete crack detection is mainly based on manual visual inspection and underwater probing by divers. These methods have several limitations, such as fragmentation of inspection results, blind inspection areas, low efficiency, and high safety risks. Moreover, the inspection depth is limited by the depth of diving operations.
The rapid development of underwater robots is increasing inspection using cameras carried by these robots. However, images obtained through the cameras cannot be used directly for inspection as obtaining high-quality underwater images is difficult. Thus, the difficulty of acquiring underwater crack images, light scattering, and color bias hinders effective underwater structure inspection.
Therefore, this study proposed a non-contact detection study of underwater concrete cracks using monocular vision and fractal science based on image enhancement.
Methods
The ground concrete crack dataset was collected through a concrete axial compression test. Four concrete specimens and some standard cubic ones were fabricated for the compression test. The loading process was divided into preloading (50 KN) and final loading stages (3000 KN).
Subsequently, the water environment was simulated to generate a new dataset for underwater crack image target detection and segmentation algorithms. An underwater image enhancement algorithm based on an encoder structure was designed to improve the performance of crack detection and segmentation algorithms.
In computer vision and image measurement, mathematical geometric models of camera imaging were developed to solve the coordinate transformation relationship between spatial target points in each coordinate system, thereby transforming between two-dimensional image coordinates and three-dimensional spatial positions.
Different crack segmentation methods were used for underwater image processing according to the characteristics of the following cases: fine linear and wide linear cracks. Notably, the Prewitt operator was used for fine crack image segmentation. Alternatively, the optimized maximum between-class variance method (Otsu) was used to handle more common and damaging wide cracks.
Gamma correction was used to reduce the brightness of the enhanced image, improving detection accuracy. Additionally, a decoder structure-based underwater image restoration network was designed to restore the color shift generated by the attenuation of underwater images and enhance image clarity and contrast. Convolution was employed to extract features with high-frequency information from the input images. This extracted information was represented through high-dimensional vectors mapped to image blocks, which consisted of three modules: shrinkage, nonlinear mapping, and expansion.
The developed system was employed to classify cracks using the fractal theory. The Fraclab toolbox in MATLAB R2019a was used to calculate and analyze the fractal dimension of cracks.
Results and Discussion
The thorough analysis revealed that increasing turbidity of the water body significantly influenced the crack measurement accuracy and efficient measurement distance for underwater concrete structures. Notably, underwater filtering, noise reduction, and image enhancement methods could enhance the recognition accuracy and restore information on cracks to recognize and measure cracks in the underwater structure.
However, the proposed method had limited processing ability for real information. The recognition accuracy error of crack width in the case of light turbidity was about 2% at 0.5 m, 9% at 0.8 m, and 16% at 1.2 m. Alternatively, for heavy turbidity, the recognition accuracy error of crack width increased to about 10% at 0.5 m, 16% at 0.8 m, and 28% at 1.2 m.
Therefore, the proposed monocular vision and image enhancement-based fractal science for underwater concrete crack detection applied to a closer distance for the turbid underwater environment.
Notably, the fractal dimension of the crack increased with increasing water turbidity, worsening the detection accuracy. However, in such cases, the fractal dimension was equivalent to a safety factor in judging the state of the structure. Thus, though the developed system could not fully reflect the safety state of the structure, it could still be used as a safety indicator, assisting measurement and analysis.
Conclusion
Overall, the researchers proposed an image-processing method based on the fractal theory to detect and quantify cracks in underwater concrete structures. An underwater environment was simulated to study the applicability of this method based on computer vision. Notably, the applied image enhancement techniques solved the challenge of acquiring images of underwater structures at short distances.
The developed algorithm exhibited good experimental results for concrete crack detection under turbid water bodies. Therefore, the system was practical and effective in a specific environment. It can be an efficient solution to enhance the visualization and monitoring of underwater structures. However, capturing images for longer distances needs further improvement.
Journal Reference
Cui, B., Wang, C., Li, Y., Li, H., & Li, C. (2024). Application of computer vision techniques to damage detection in underwater concrete structures. Alexandria Engineering Journal, 104, 745–752. DOI: 10.1016/j.aej.2024.08.020, https://www.sciencedirect.com/science/article/pii/S1110016824009050
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.