Automatic classification and quantification of asphalt pavement cracks using deep learning-based object detection algorithms

Articles in Press

Document Type : Original Article

Authors

1 Department of Civil Engineering, Technical and Engineering Faculty, Eqbal Lahori Institute of Higher Education, Mashhad, Iran

2 Department of Computer Engineering, Technical and Engineering Faculty, Eqbal Lahori Institute of Higher Education, Mashhad, Iran

10.22034/road.2022.350483.2066

Abstract

Pavement crack detection is an important method in road maintenance and traffic safety.Traditionally, road condition assessment was done by field inspection, which is time-consuming and costly, and the results are dependent on the evaluator's opinion, It should be replaced by automated methods to reduce both workload and thus maintenance costs. This paper examines the performance of YOLOv3 and YOLOv5 algorithms for automatic crack detection.These models are able to determine the type, position and geometric characteristics of the crack accurately and at a high speed compared to other methods.For the purpose of modeling, the images taken from Mashhad surface roads have been used. These images were labeled for linear and surface crack options. Then, models were created using the v3 model and five v5 series algorithms and transfer learning and were evaluated in terms of accuracy and prediction speed. The accuracy of the models is between 77 and 98% and the prediction speed of the models is between 17.4 and 105 milliseconds, which indicates the optimal performance of the models.Finally, v5s model was used as the final model for predicting cracks in one of Mashhad roads due to its acceptable accuracy (92.8) and high prediction speed (23.9 milliseconds) compared to other models.Then, based on the outputs of the model, the maintenance approach was presented.

Keywords

Files

Share

How to cite

Statistics