Project Team
Students
Vedat Veziroglu
Aerospace Engineering
Penn State Lehigh Valley
Faculty Mentors
Tracey Carbonetto
Penn State Lehigh Valley
Engineering
S. Ilgin Guler
Penn State University Park
Civil and Environmental Engineering
Project
Project Video
Project Abstract
Structural degradation of pavements due to repeated traffic loads and environment could manifest as rutting, cracking, or deformations. Among the visual indicators of damage, different types of cracking are often used to evaluate the health of a pavement. Manual inspection methods of condition evaluation are often time-consuming and include a degree of subjectivity that might lead to uncertain conclusions. Deep learning (DL) is an expanding field providing ways to automate laborious tasks and is well suited for the task of autonomously detecting cracks from images of pavement data with high accuracy. The literature on using DL to detect cracks from roadway imagery is limited. State-of-the-art DL models are based on Convolutional neural networks (CNNs) that perform either binary classification to label images as cracked versus non-cracked or a semantic segmentation that can efficiently locate the cracks in an image. In this study, both approaches have been explored. First, a regular CNN architecture for crack classification is tested. Owing to its simplicity, the basic CNN could not perform satisfactorily as expected. Next, an image segmentation architecture, U-Net is used to perform semantic segmentation using masked images of cracked pavement data. The model achieved an accuracy of 0.91 on the test data. Further, to check the robustness of the trained model, pavement images from Google Streetview were tested without further fine-tuning. Identifying diverse types and orientations of cracks is beyond the scope of this research, which will be explored in the future.
Evaluate this Project
Use this form link to provide feedback to the presenters, and add your project evaluation for award(s) consideration.