Detection and classification of asphalt pavement cracks using YOLOv5

Abstract

Automatic pavement crack detection is essential for assessing road maintenance and ensuring safe driving. Traditional crack detection has problems such as low efficiency and lack of complete detection. This study aims to solving the problems of traditional crack detection methods and using deep learning models. We proposed a method based on object detection algorithms for pavement crack detection and discussed the latest YOLOv5 series models for pavement crack detection while explaining the theoretical concepts. Finally, a crack detection model and effective pavement management is presented. The proposed model can determine the type, position and geometric characteristics of cracks accurately and at a higher speed in comparison with other methods. For this purpose, the images that had been taken from the asphalt of Mashhad roads were used to train and evaluate the model. Images were labeled for both linear crack and surface crack. Proposed model is developed using five YOLOv5 series algorithms and transfer learning and were evaluated for accuracy and speed of prediction. The models’ accuracy is between 77 to 98% and the prediction speed is between 17.4 to 105 milliseconds, which indicates the optimal performance of the models. The v5s model with 92.8% accuracy and a speed of 23.9 ms is selected as the final model for real prediction of cracks in one of the main thoroughfares of Mashhad. Based on the dimensions and the type of predicted crack and the use of the proposed decision tree, the maintenance approach for each part was determined.

Keywords

• pavement management
• crack
• machine learning
• deep learning
• yolov5