PCA-Based algorithm for unsupervised bridge crack detection

Principal Component Principles (PCA) based algorithm to extract cracks in concrete bridge decks for the purpose of automating inspection is presented. PCA will be used to identify clusters using a database of bridge images. Results from three different PCA approaches are presented in this work. The first approach employs PCA by itself on raw data. In the second approach, a linear structure modeling is implemented prior to PCA processing in an effort to enhance the results since cracks can be detected as linear structures. Several convolution-processes with masks designed to identify linear structure in the data are used. In both cases, attempts to detect cracks in a global framework were used. The third approach, on the other hand, used local information (neighborhoods) instead of global. That is, each image is segmented into small blocks where each block is processed as an individual entity. Experimental results show enhancement in the local detection with linear modeling over the global.