Title:  Pavement Crack Segmentation on Range Images using Deep Learning

Committee: 

Dr. Tsai, Advisor

Dr. Yezzi, Co-Advisor   

Dr. AlRegib, Chair

Dr. Xiuwei Zhang

Abstract: The objective of the proposed research is to develop an advanced crack detection algorithm that leverages the emerging machine learning and 3D line laser sensor technologies to achieve better resolution, accuracy, and robustness on automated pavement crack detection. Specifically, a novel crack segmentation algorithm is proposed to perform pixel-wise detection of cracks on pavement range images captured by 3D line laser sensors. The proposed algorithm leverages the power of deep learning to robustly recognize cracks from complex pavement surfaces. In addition, with the advantage of the range image, geometrical modeling approaches are developed to further improve the generalization and detail-preserving capabilities of the crack segmentation algorithm. The outcome of the proposed research will enable transportation agencies to detect cracks from pavement surface data with higher resolution, accuracy, and robustness. This will support the transportation agencies to apply optimal treatments at the right location and time for maintaining the transportation infrastructures cost-effectively.