Image Processing-Based Analysis of Wall Crack Using Machine Learning Approaches

Sum, Kah Zheng (2021) Image Processing-Based Analysis of Wall Crack Using Machine Learning Approaches. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Sum Kah Zhen.pdf Restricted to Registered users only Download (13MB)

Abstract

The need for manual labour in inspecting building wall cracks by trained inspectors causes inconvenience and inefficiency, thus unnecessary time and monetary costs. The need for trained inspectors is also a cause for concerns including the lack of trained inspectors and subjectivity of judgements among inspectors. While previous researches have successfully used machine learning algorithms for crack detection as well as image processing for crack width estimation, little attempt has been found on integrating both methods into a complete solution. The proposed solution is able to use pretrained Convolutional Neural Network (CNN) model, specifically VGG16 to predict wall cracks. Besides, image processing is able to estimate wall cracks width well in pixel units. Adaptive thresholding is used for binarization while median blurring, morphological closing, morphological opening and shape analysis are used for noise removal. Both methods are integrated into a full solution where only predicted cracks are estimated. A simple deployment method using Python open source libraries has also been suggested. The solution is able to detect most cracks and is able to estimate thin and medium cracks well. However, the solution does not include conversion from estimations in pixels to real life physical units. Future works include collecting real life physical measurements data, further improving noise removal efficiency and collecting more wall crack images focusing on challenging noises which has exposed the limitation of the solution proposed.