Dissertation
Development of pavement crack analysis algorithms using deep learning techniques to analyze UAV images
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Summer 2022
DOI: 10.25820/etd.006618
Abstract
The pavement condition is periodically evaluated by publics agencies, where the collected data are used to determine how to distribute limited funds to maintain pavements. Because the traditional pavement condition survey method using cameras mounted on a vehicle suffers from the lack of repeatability with a high cost, unmanned aerial vehicle (UAV) is considered as a promising tool to evaluate pavement conditions at a significantly lower cost. Inspired by current achievements of deep learning technology in various computer vision applications, this thesis presents automated crack detection algorithms using deep learning for UAV images of roads.
In order to develop pavement extraction and crack segmentation algorithms, U-Net was selected as a base model, and contracting encoder path of the U-Net model was modified with four different feature extraction deep learning models (VGG 16, Inception V3, ResNet 50, DenseNet 169). A tile-based pavement crack analysis system was then developed and applied to measure percent cracking and crack widths from the asphalt pavement sections. The developed crack analysis system can help public agencies manage pavements in a systematic and cost-effective manner.
Details
- Title: Subtitle
- Development of pavement crack analysis algorithms using deep learning techniques to analyze UAV images
- Creators
- Byungkyu Moon
- Contributors
- Hosin “David” Lee (Advisor)M. Asghar Bhatti (Committee Member)Paul Hanley (Committee Member)Salam Rahmatalla (Committee Member)Punam Saha (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Civil and Environmental Engineering
- Date degree season
- Summer 2022
- Publisher
- University of Iowa
- DOI
- 10.25820/etd.006618
- Number of pages
- xi, 127 pages
- Copyright
- Copyright 2022 Byungkyu Moon
- Language
- English
- Description illustrations
- illustrations (some color), tables
- Description bibliographic
- Includes bibliographical references (pages 99-107).
- Public Abstract (ETD)
- Pavement condition surveys are conducted by employing a vehicle equipped with data collection devices such as cameras, sensors, GIS, and accelerometers. However, this traditional data collect method suffers from the lack of repeatability, requiring very high manpower, time, and cost. To overcome such limitations, unmanned aerial vehicle (UAV) is considered as a promising tool to evaluate pavement conditions at a lower cost. Inspired by current achievements of deep learning technology in various computer vision tasks, such as object classification and sematic segmentation, this thesis proposes integrated automated crack detection and condition assessment system using deep learning for UAV image. As sustainability is increasingly emphasized, the main objective of this thesis is to develop automated crack detection algorithm for analyzing UAV images using the state-of-theart machine learning techniques by performing the following tasks: 1) developing a pavement extraction algorithm which can identify pavements from a drone image, 2) developing a crack detection algorithm from the extracted pavement images and 3) developing a crack quantification algorithm based on the detected cracks. A total of sixteen deep learning models were created, and top four models for pavement extraction and crack segmentation modules showed very high performance. A tile-based pavement crack analysis system was then developed and applied to measure percent cracking and crack widths from the asphalt pavement test sections. Proposed system showed high performance on detecting hairline cracks and distinguishing cracks from other objects and background. It can be expected that proposed system can provide strategic recommendations to the public agencies on managing pavements in a more systematical and cost-efficient manner.
- Academic Unit
- Civil and Environmental Engineering
- Record Identifier
- 9984285050702771
Metrics
44 File views/ downloads
27 Record Views