Ouyang et al. (2025) A deep learning method for identifying waterlogging depth on urban roadways from surveillance camera images

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2025
• Date: 2025-12-12
• Authors: Mingyu Ouyang, Bowei Zeng, Guoru Huang
• DOI: 10.1016/j.rsase.2025.101827

Research Groups

• State Key Laboratory of Subtropical Building and Urban Science, South China University of Technology
• School of Civil Engineering and Transportation, South China University of Technology
• Guangdong Engineering Technology Research Center of Safety and Greenization for Water Conservancy Project

Short Summary

This paper introduces a deep learning method that integrates Cascade Mask R-CNN with ellipse detection to precisely identify waterlogging depth on urban roadways from surveillance camera images, achieving high accuracy and low absolute errors compared to manual measurements.

Objective

• To develop a rapid and precise method for detecting waterlogging depth on urban roads using surveillance camera images, addressing the limitations of traditional, labor-intensive, and expensive monitoring methods.

Study Configuration

• Spatial Scale: Urban roadways, specifically validated at a representative surveillance video site in Dongying City, China.
• Temporal Scale: Real-time or near real-time monitoring, enabling timely management of urban flooding events.

Methodology and Data

• Models used: Cascade Mask R-CNN (for vehicle wheel detection and segmentation), ellipse detection algorithms, minimum area rectangle detection algorithms.
• Data sources: Surveillance camera images/video from urban roadways. Validation involved 246 samples compared with manually measured depths.

Main Results

• The proposed model achieved an average bounding box precision and segmentation precision exceeding 97 % on the validation dataset.
• For 246 validation samples, the absolute errors in waterlogging depth measurements were consistently below 0.1 m when compared with manually measured depths.
• The method successfully utilizes detected and segmented vehicle wheels as reference objects for calculating waterlogging depth.

Contributions

• Proposes a novel deep learning-based method for real-time waterlogging depth identification on urban roads using readily available surveillance camera images.
• Integrates Cascade Mask R-CNN with geometric algorithms (ellipse and minimum area rectangle detection) for precise wheel segmentation and subsequent depth calculation.
• Demonstrates high accuracy (over 97 % precision) and low absolute errors (below 0.1 m) in waterlogging depth measurement, offering a significant improvement over traditional monitoring methods.

Funding

Not specified in the provided text.

Citation

@article{Ouyang2025deep,
  author = {Ouyang, Mingyu and Zeng, Bowei and Huang, Guoru},
  title = {A deep learning method for identifying waterlogging depth on urban roadways from surveillance camera images},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2025},
  doi = {10.1016/j.rsase.2025.101827},
  url = {https://doi.org/10.1016/j.rsase.2025.101827}
}