Electronic Theses and Dissertations


Mohsen Maniat



Document Type


Degree Name

Doctor of Philosophy


Civil Engineering

Committee Chair

Charles Camp

Committee Member

Shahram Pezeshk

Committee Member

Adel Abdelnaby

Committee Member

Mihalis Gkolias


The need for developing an economical and efficient quality assessment system for pave-ment motivates this study to take advantage of available new technologies and provide a novel approach to address this need. In this study, the utility of using Google Street View (GSV) for evaluating the quality of pavement is investigated. GSV is a technology featured in Google Maps and Google Earth that provides interactive panoramas along many streets throughout the world. This technology provides a large data set of pavement images that can be used for pavement evaluation. Advanced deep learning algorithms are utilized to automate the pavement assessment process of these GSV images. These algorithms autonomously learn to find the important fea-tures in a data set to perform a particular task. A convolutional neural network (CNN) is one of the deep learning algorithms that has been shown to be very effective in learning from digital im-ages. Several CNNs are used in this study to perform image classification on GSV pavement im-ages. Training an effective CNN with many learning parameters requires a large image data set. To provide the required data for training a CNN, a large number of pavement images are ex-tracted from GSV are then divided to smaller image patches to form a larger data set. Each image patch is visually classified into different categories of pavement cracks based on the standard practice. A comparative study of pavement quality assessment is conducted between the results of the CNN classified images patches obtained from GSV and those from a sophisticated com-mercial visual inspection company. The result of the comparison indicates the feasibility and ef-fectiveness of using GSV images for pavement evaluation. An effective CNN is designed and trained on the image data set to automate the crack detection process. The trained network is then tested on a new data set. The results of this study show that the designed CNN is effective in classifying the pavement images into different defined crack categories.EP LEARNING-BASED VISUAL CRACK DETECTION USING GOOGLE STREET VIEW IMAGES


Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest