UAV Photogrammetry-Based 3D Road Distress Detection

Yumin Tan, Yunxin LiISPRS Int. J. Geo Inf 2019 · DOI

TL;DR

This paper introduces a novel, low-cost method for automatically detecting and measuring road distress (like potholes and bulges) using 3D models generated from Unmanned Aerial Vehicle (UAV) photogrammetry. The approach achieves centimeter-level accuracy in height measurements, demonstrating its potential for practical road maintenance.

What Problem Does This Paper Address?

Maintaining road infrastructure is crucial for safety and efficiency. Traditional manual road inspections are time-consuming, labor-intensive, and subjective. While automated methods exist, they often struggle with comprehensive coverage, can disrupt traffic, or lack detailed 3D information. This paper addresses the need for a highly efficient, low-cost, and accurate method to detect and quantify road surface distress, particularly focusing on unevenness like bulges and cavities, by leveraging the flexibility and accessibility of UAVs.

Methods

The core of this research lies in a multi-step process:

  1. 3D Model Reconstruction: Unmanned Aerial Vehicles (UAVs) capture overlapping oblique images of the road. These images are then processed using Pix4Dmapper photogrammetry software to create a dense 3D point cloud model of the road surface.
  2. Pavement Extraction: A region growing algorithm is employed to segment the road surface from its surroundings. This involves meshing the point cloud, filtering out noise (like tree leaves), and iteratively growing a "road" region based on elevation similarity.
  3. Distress Detection: For the extracted road surface, a local plane fitting approach is used within sliding windows. Points deviating significantly from these local reference planes are identified as potential distress. A deviation threshold (DT) is crucial here, and a region growing algorithm is re-applied to delineate the full extent of the distress area.
  4. Feature Extraction: Once distress regions are identified, their geometric features are extracted. This includes calculating the area using the Graham scan algorithm for border determination, and defining length and width based on the principal axes of inertia of the distress point cloud.

Datasets

The study utilized data collected from eight road segments in Fuling district, Chongqing, China. These segments varied in length, width, and surface type (all asphalt). Field measurements of distress dimensions (length, width, and height/depth) were also collected for comparison. No new benchmark datasets were introduced; rather, existing road infrastructure was used for data acquisition.

Key Insights and Ideas

  • UAVs as a Cost-Effective Solution: The paper strongly advocates for UAVs as a practical and affordable alternative to traditional survey vehicles for road condition monitoring, highlighting their flexibility and ease of use.
  • 3D Models Unlock Detailed Analysis: Moving beyond 2D imagery, the 3D models provide crucial geometric information (height/depth) that is essential for accurately characterizing distress severity.
  • Region Growing for Robust Segmentation: The use of region growing for both pavement extraction and distress identification proves effective in handling complex road environments and identifying irregular distress shapes.
  • Centimeter-Level Height Accuracy: The achieved accuracy in height/depth measurement (around 1 cm error) is a significant finding, demonstrating the method's suitability for practical engineering applications where precise measurements are needed.

Key Results

The proposed method successfully detected various road distress types, including cavities (potholes, subsidence) and bulges (piling up, corrugation), across the eight tested road segments. The height/depth error for detected distresses was consistently around 1 cm, with a 95% upper bound of 2.12 cm for all samples. This level of precision is deemed reliable for engineering practice. The method also accurately identified the precise locations and boundaries of distress regions.

Limitations and Future Work

  • Model Accuracy Dependency: The accuracy of the final distress detection is inherently tied to the accuracy of the initial 3D model generated by Pix4Dmapper. Factors like image resolution and camera calibration play a significant role.
  • GCPs for Absolute Accuracy: While the method achieves high relative accuracy, the use of Ground Control Points (GCPs) is acknowledged as beneficial for improving absolute accuracy, though it adds to field work.
  • Distress Definition: The paper notes that the definition of "length" and "width" for irregular distress shapes can be ambiguous, and their measurements are based on principal axes of inertia.
  • Future Work: The authors suggest improving accuracy and efficiency. They also propose leveraging existing stable landmarks (like milestones) as natural GCPs to enhance model accuracy without extensive field work.

Summary updated

BibTeX
@article{tan2019uav,
  author = {Yumin Tan and Yunxin Li},
  title = {{UAV Photogrammetry-Based 3D Road Distress Detection}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2019},
  doi = {10.3390/ijgi8090409},
}
Figure