Damage Signature Generation of Revetment Surface along Urban Rivers Using UAV-Based Mapping

Ting Chen, Haiqing He, Dajun Li, Puyang An, Zhenyang HuiISPRS Int. J. Geo Inf 2020 · DOI

TL;DR

This paper introduces a novel UAV-based photogrammetry approach for automatically detecting damage signatures on urban river revetments. By generating high-resolution 3D point clouds and employing a self-adaptive, multiscale gradient operator, the method achieves significantly improved accuracy in identifying revetment damage compared to traditional visual inspections and other automated methods.

What Problem Does This Paper Address?

Urban river revetments are crucial for flood control and infrastructure protection, but they are susceptible to damage from various factors like erosion, land subsidence, and human interference. Traditional visual inspections for revetment damage are time-consuming, labor-intensive, and can be technically complex, especially for obtaining detailed information. While remote sensing technologies like UAV photogrammetry offer potential for high-resolution 3D reconstruction, existing methods often struggle with automated, accurate damage recognition, particularly in complex urban environments. This paper aims to bridge this gap by developing an efficient and accurate automated system for revetment damage signature generation using UAV-based mapping.

Methods

The core of this research lies in a two-pronged approach:

  1. High-Resolution 3D Surface Reconstruction: The study utilizes a low-cost UAV equipped with a consumer-grade camera to capture aerial images. These images are processed using a novel object space-based dense matching approach that combines region growing with Semi-Global Matching (SGM). This method reconstructs a finely detailed 3D surface of the revetment, aiming for accuracy and efficiency. Key steps include feature extraction and matching, incremental Structure-from-Motion (SfM), bundle adjustment, and the aforementioned region growing coupled with SGM.
  2. Self-Adaptive and Multiscale Gradient Operator (SMGO) for Damage Signature Generation: To identify damage, the generated 3D point clouds are transformed into a slope intensity image. A custom operator, SMGO, is then applied to this image. SMGO is designed to extract abnormal regions of varying sizes by considering omnidirectional gradients (horizontal, vertical, and diagonal) across multiple scales. This multiscale architecture allows for the detection of damage features that might be missed by single-scale or non-adaptive operators. Vegetation removal is also incorporated to mitigate its impact on the analysis.

Datasets

The study was conducted using data collected from concrete revetments along urban rivers in Nanchang City, China. Two sections of revetment, measuring 450m and 570m in length, were surveyed. The data acquisition involved a DJI Mavic Air UAV capturing high-resolution images. Ground Control Points (GCPs) were established and measured using RTK GPS to georeference the dense point clouds and validate the accuracy of the reconstruction. The paper does not introduce a new benchmark dataset but rather uses a real-world case study for evaluation.

Key Insights and Ideas

  • UAVs as a Cost-Effective Solution: The research demonstrates that low-cost UAVs can be effectively employed for detailed revetment monitoring, offering a more accessible alternative to expensive terrestrial laser scanning or complex aerial surveys.
  • Object Space-Based Dense Matching: Moving beyond traditional image-space matching, the object space-based approach is shown to improve efficiency and accuracy in dense point cloud generation, particularly for revetment surfaces.
  • Multiscale Gradient Operator for Robust Damage Detection: The SMGO's ability to analyze damage signatures at multiple scales is crucial for capturing diverse damage types (e.g., cracks vs. collapses) and ensuring comprehensive detection, overcoming limitations of simpler gradient methods.
  • Independence from Prior Data: The proposed method does not require pre-existing data of undamaged revetments, making it more practical for real-world applications where historical data might be scarce or incomplete.

Key Results

  • High Accuracy Reconstruction: The proposed dense matching method achieved RMSE values for check points that were comparable to or better than commercial software like Pix4Dmapper and Agisoft Metashape. Specifically, horizontal errors were less than 4 cm, and vertical errors were less than 6 cm.
  • Improved Damage Recognition Accuracy: The SMGO-based damage recognition method significantly outperformed both field visual inspection and a non-multiscale gradient operator (NMGO). For collapse detection, the proposed method achieved Precision, Recall, and F1_score values of approximately 92.85%. For crack detection, these metrics were around 89.18%, 91.67%, and 90.41%, respectively, consistently higher than the other methods.
  • Efficiency: The proposed method's time consumption was reported to be 87% of that of Pix4Dmapper and Agisoft Metashape.

Limitations and Future Work

  • Steep Slopes: The current approach, relying on nadir camera orientation, may face challenges in achieving high-precision surface reconstruction for very steep revetments.
  • Future Work: The authors suggest future research to optimize the approach by incorporating oblique photogrammetry and deep learning techniques to enhance damage recognition capabilities, especially for steep revetments. Further investigation into the application of the method on revetments with steeper slopes is also recommended.

Summary updated

BibTeX
@article{chen2020damage,
  author = {Ting Chen and Haiqing He and Dajun Li and Puyang An and Zhenyang Hui},
  title = {{Damage Signature Generation of Revetment Surface along Urban Rivers Using UAV-Based Mapping}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2020},
  doi = {10.3390/ijgi9040283},
}
Figure