TY - GEN
T1 - Using artifical intelligence for automating pavement condition assessment
AU - Asian, O. D.
AU - Gultepe, E.
AU - Ramaji, I. J.
AU - Kermanshachi, S.
N1 - Publisher Copyright:
© International Conference on Smart Infrastructure and Construction 2019, ICSIC 2019: Driving Data-Informed Decision-Making.
PY - 2019
Y1 - 2019
N2 - The financial burden due to pavement damage on road networks is a major handicap to the economic development of a country. According to an ASCE report, this issue may cost as much as $67 billion per year. Regularly planned condition assessments and repairs of pavement can mitigate any derived costs and increase traffic safety. However, due to the large extent of civil infrastructure networks, required periodic inspections and assessments can be expensive and time-consuming. Further compounding the issue is that the majority of damage assessment mechanisms rely on human visual analysis, which can be prone to potential user bias and errors. In this study, we present a framework to automate roadway assessment by implementing a Convolutional Neural Network (CNN) that classifies various types of cracks in pavements. CNNs are a special type of deep artificial neural networks that demonstrate high accuracy and efficiency in image-based machine learning tasks. One of the main advantages of CNNs is that they can automatically learn the salient features of an image dataset without any prior knowledge or pre-processing by the user. Thus, the need for feature engineering is obviated and thereby eases the deployment of our assessment framework. Our framework was developed and tested on a balanced dataset containing 400 color images and consisting of four types of pavement damage: (1) longitudinal, (2) transverse, (3) alligator, and (4) pothole cracks. We apply image augmentation using a bundle of transformations to improve the crack classification accuracy of our CNN. The classification accuracy of the four types of cracks was found to be 76.2%. Demonstrating that the proposed CNN model can predict crack types without any user intervention at a good level of accuracy. To improve the robustness and accuracy of our assessment framework, we will analyze more types of cracks, using a larger dataset size in future studies.
AB - The financial burden due to pavement damage on road networks is a major handicap to the economic development of a country. According to an ASCE report, this issue may cost as much as $67 billion per year. Regularly planned condition assessments and repairs of pavement can mitigate any derived costs and increase traffic safety. However, due to the large extent of civil infrastructure networks, required periodic inspections and assessments can be expensive and time-consuming. Further compounding the issue is that the majority of damage assessment mechanisms rely on human visual analysis, which can be prone to potential user bias and errors. In this study, we present a framework to automate roadway assessment by implementing a Convolutional Neural Network (CNN) that classifies various types of cracks in pavements. CNNs are a special type of deep artificial neural networks that demonstrate high accuracy and efficiency in image-based machine learning tasks. One of the main advantages of CNNs is that they can automatically learn the salient features of an image dataset without any prior knowledge or pre-processing by the user. Thus, the need for feature engineering is obviated and thereby eases the deployment of our assessment framework. Our framework was developed and tested on a balanced dataset containing 400 color images and consisting of four types of pavement damage: (1) longitudinal, (2) transverse, (3) alligator, and (4) pothole cracks. We apply image augmentation using a bundle of transformations to improve the crack classification accuracy of our CNN. The classification accuracy of the four types of cracks was found to be 76.2%. Demonstrating that the proposed CNN model can predict crack types without any user intervention at a good level of accuracy. To improve the robustness and accuracy of our assessment framework, we will analyze more types of cracks, using a larger dataset size in future studies.
UR - http://www.scopus.com/inward/record.url?scp=85087419739&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85087419739&partnerID=8YFLogxK
U2 - 10.1680/icsic.64669.337
DO - 10.1680/icsic.64669.337
M3 - Conference contribution
AN - SCOPUS:85087419739
T3 - International Conference on Smart Infrastructure and Construction 2019, ICSIC 2019: Driving Data-Informed Decision-Making
SP - 337
EP - 341
BT - International Conference on Smart Infrastructure and Construction 2019, ICSIC 2019
A2 - DeJong, M.J.
A2 - Schooling, Jennifer M.
A2 - Viggiani, G.M.B.
PB - ICE Publishing
T2 - 2nd International Conference on Smart Infrastructure and Construction: Driving Data-Informed Decision-Making, ICSIC 2019
Y2 - 1 July 2019 through 3 July 2019
ER -