Overview / Usage

Today, the rapidly growth of cities has made it necessary to following up the performance of infrastructures and architectural works. In developing cities, it is of great importance for human life to follow the deformation of structures over time or materials worn out as a result of disasters. Today, many different methods have been developed for structural deformation evalutaions. In particular, the detection of cracks on material surfaces provides important inferences for performance evaluations. Manual analysis of performance in infrastructure and architectural works does not meet today's needs, both in terms of cost and due to subjective conclusions. Important developments in information technologies ensured that material analysis and audits are carried out objectively and reduced costs to a minimum. Thereby, surface performance analysis by processing images in a computer environment has allowed to obtain higher accuracy results with less cost. In this study, image processing methods were imposed on to analyze surface cracks in materials used in architectural structures and land roads. Surface cracks have been detected by using morphological image processing methods and Hough-line transformation. In order to evaluate the success of these two methods, machine learning methods were used. Comparisons were made using Naive Bayes, Random Forest, Knn and Multi-Layer Perceptron methods, which are machine learning methods for success evaluations. The results gathered as a result of the experiments showed that the Hough-line transform is more successful in extracting features. The highest accuracy was achieved with the Hough-line method, one of the image processing methods, and Random Forest, one of the machine learning methods. It has been successfully classified with an accuracy rate of 97% using Hough-line transformation from image processing methods and Random Forest from machine learning methods to detect cracks on surfaces.

Methodology / Approach

Image processing with hough-line, machine learning for surface detection