Scholarlink

<p>Pavement surface monitoring is a critical aspect of ensuring road longevity and minimizing maintenance costs by addressing damages at an early stage. Traditional manual inspection methods are time-consuming and prone to errors, while sensor-based approaches often lack the precision required for real-time detection. This study proposes an automated, real-time solution for detecting and assessing the severity of asphalt pavement distress using YOLOv8 (You Only Look Once), a state-of-the-art convolutional neural network (CNN) model. The dataset, prepared using Roboflow, comprises 29,000 high-resolution images, including 12,000 images of potholes and 17,000 images of various crack types. After rigorous preprocessing, annotation, and augmentation, the YOLOv8 model was trained to identify pavement distresses such as potholes and alligator, longitudinal, and block cracks. The model achieved an accuracy of 88.87%, with an error rate of 11.12% in calculating distress areas. Severity classification was performed based on dimensions and depth, following guidelines from the Federal Highway Administration (FHWA). For example, the model accurately detected a low-severity pothole with dimensions of 128.1&#xa0;mm × 39.4&#xa0;mm and a depth of 5.0&#xa0;mm. Precision-recall analysis showed high efficiency for detecting alligator cracking (99.5% precision) and an overall model performance of 82.4% at mAP50. The proposed framework provides a robust tool for transportation agencies, including the Ministry of Road Transport and Highways (MORTH) and the National Highway Authority of India (NHAI), to enhance road maintenance and ensure safety. Future enhancements will focus on extending the dataset to diverse environmental conditions and improving the precision of bounding box dimensions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Detection and Severity Assessment of Asphalt Pavement Distress Using YOLOv8: A Deep Learning Approach

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Ravi Kumar,
  • Smita Tung

摘要

Pavement surface monitoring is a critical aspect of ensuring road longevity and minimizing maintenance costs by addressing damages at an early stage. Traditional manual inspection methods are time-consuming and prone to errors, while sensor-based approaches often lack the precision required for real-time detection. This study proposes an automated, real-time solution for detecting and assessing the severity of asphalt pavement distress using YOLOv8 (You Only Look Once), a state-of-the-art convolutional neural network (CNN) model. The dataset, prepared using Roboflow, comprises 29,000 high-resolution images, including 12,000 images of potholes and 17,000 images of various crack types. After rigorous preprocessing, annotation, and augmentation, the YOLOv8 model was trained to identify pavement distresses such as potholes and alligator, longitudinal, and block cracks. The model achieved an accuracy of 88.87%, with an error rate of 11.12% in calculating distress areas. Severity classification was performed based on dimensions and depth, following guidelines from the Federal Highway Administration (FHWA). For example, the model accurately detected a low-severity pothole with dimensions of 128.1 mm × 39.4 mm and a depth of 5.0 mm. Precision-recall analysis showed high efficiency for detecting alligator cracking (99.5% precision) and an overall model performance of 82.4% at mAP50. The proposed framework provides a robust tool for transportation agencies, including the Ministry of Road Transport and Highways (MORTH) and the National Highway Authority of India (NHAI), to enhance road maintenance and ensure safety. Future enhancements will focus on extending the dataset to diverse environmental conditions and improving the precision of bounding box dimensions.