[論文レビュー] Deep Learning Approaches in Pavement Distress Identification: A Review

Sizhe Guan, Haolan Liu+2 more|arXiv (Cornell University)|Aug 1, 2023

Infrastructure Maintenance and Monitoring被引用数 7

ひとこと要約

A comprehensive review of recent deep learning methods for pavement distress detection using UAV-based 2D imagery, covering data collection, algorithms, datasets, and challenges.

ABSTRACT

This paper presents a comprehensive review of recent advancements in image processing and deep learning techniques for pavement distress detection and classification, a critical aspect in modern pavement management systems. The conventional manual inspection process conducted by human experts is gradually being superseded by automated solutions, leveraging machine learning and deep learning algorithms to enhance efficiency and accuracy. The ability of these algorithms to discern patterns and make predictions based on extensive datasets has revolutionized the domain of pavement distress identification. The paper investigates the integration of unmanned aerial vehicles (UAVs) for data collection, offering unique advantages such as aerial perspectives and efficient coverage of large areas. By capturing high-resolution images, UAVs provide valuable data that can be processed using deep learning algorithms to detect and classify various pavement distresses effectively. While the primary focus is on 2D image processing, the paper also acknowledges the challenges associated with 3D images, such as sensor limitations and computational requirements. Understanding these challenges is crucial for further advancements in the field. The findings of this review significantly contribute to the evolution of pavement distress detection, fostering the development of efficient pavement management systems. As automated approaches continue to mature, the implementation of deep learning techniques holds great promise in ensuring safer and more durable road infrastructure for the benefit of society.

研究の動機と目的

• Survey the state-of-the-art in AI applications for pavement distress evaluation with a focus on 2D UAV-based imagery.
• Explain data acquisition, preprocessing, and model evaluation workflows for pavement distress detection.
• Compare traditional image processing methods with learning-based approaches in pavement distress tasks.
• Discuss UAV data collection challenges, 3D data considerations, and IoT/AI integration for dataset quality.
• Identify public datasets, open-source codes, and future research directions in UAV-based pavement distress detection.

提案手法

• categorize methods into traditional feature-based vs learning-based (DL) approaches for distress detection
• discuss UAV-based data acquisition and image quality considerations, including flight parameter tuning
• review preprocessing, feature extraction, representation, and classification pipelines
• highlight 2D image processing focus, with notes on 3D data and alternative sensing (GPR, acceleration, sound)
• summarize datasets, evaluation techniques, and availability of open-source implementations
• present challenges and future research directions in UAV-based pavement distress identification

実験結果

リサーチクエスチョン

• RQ1 What are the recent deep learning approaches used for pavement distress evaluation on UAV-based 2D images?
• RQ2 How is UAV-based data acquisition utilized and what are its advantages and challenges for pavement distress detection?
• RQ3 What public datasets and open-source resources exist for UAV-based pavement distress research, and how are models evaluated?
• RQ4 What are the main challenges in extending 2D DL pavement distress methods to 3D data and other sensing modalities?
• RQ5 What future directions are most promising for improving automated pavement distress detection and management?

主な発見

• UAVs provide high-mobility, cost-effective data collection that supports detecting cracks, potholes, and rutting.
• DL-based pavement distress evaluation benefits from dedicated data acquisition pipelines and preprocessing to improve model performance.
• Examples show notable accuracies in related image processing tasks, including a 92.14% accuracy for highway image line detection using the Hough transform and 0.9 average accuracy for 3D-GPR distress classification (on referenced studies).
• Traditional image processing and feature extraction are being complemented by DL approaches that learn discriminative representations from 2D UAV imagery.
• Image quality, camera flight settings, and overlaps critically influence UAV-derived datasets and downstream model performance.
• There is a growing emphasis on integrating AI with IoT technologies to enhance dataset quality and reduce subjectivity in measurements.