Flexible and stone pavements distress detection and measurement by deep learning and low-cost detection devices

In this article is presented a robust and real-time low-cost automated method for detecting and measuring the various distress types of flexible and stone road pavements. The distress detection, classification and measurement are based on the applications of deep learning approach and YOLOv3 algorithm. A dataset for road pavements damage detection with approximately 9,150 images and 15,585 bounding boxes of flexible and stone road pavements damage was first created and then used in the neural networks training phase. The values reached by the metrics used in the research to evaluate the object detection performance (Loss, Precision, Recall, RMSE) prove that the proposed model detects the pavement distresses with high accuracy and precision. The validation of the method was performed by an error analysis obtained by comparing for some case studies the pavement distresses detected with the suggested method and the real ones. The correct detection rate in the pavement distress detection ranges from 91.0 % to 97.3% depending on the pavement and distress types. The effectiveness of the proposed technique in detecting and measuring flexible and stone pavements distress sheds light on new opportunities for carrying out preliminary and exhaustive inspections of flexible and stone pavements using low-cost detection devices and artificial intelligence techniques.

Automated summary

This article presents a robust and real-time low-cost automated method for detecting and measuring various distress types of road pavements using deep learning and YOLOv3 algorithm. The proposed technique achieves high accuracy and precision in detecting pavement distress and sheds light on new opportunities for using low-cost detection devices and artificial intelligence techniques in carrying out inspections of road pavements.