管廊作为常见的城市基础设施, 其安全性对于城市居民极其重要。该研究针对现阶段管廊监测系统效率低、成本高的问题, 提出了一种基于计算机视觉技术的高效沉降监测方法。该方法创新性地使用特征平面叠加的方法, 采用基于组合缩放因子的空间监测方案, 利用不同的像素精度及灵活的截面叠加实现实时沉降监测。以北京市慈云寺桥地下管廊为例对该方法进行验证。实验结果显示：随着监测距离的增加, 监测误差能够收敛于特定常数值, 该方法可以通过校准的方式实现有效的高精度、高性能监测。环境因素定量监测结果也显示, 即使大部分监测截面处在不同温湿度、气压、粉尘的影响下, 该方法的监测结果仍具有较高的稳定性。

Objective: The condition of utility corridors, a critical component of urban infrastructure, is crucial for the public safety. However, underground utility corridors often have long routes and traverse complex geological areas, making structural inspections extremely difficult. Disturbances such as ground deformation, loads from the overlying strata and surrounding buildings, and nearby construction activities can cause uneven settlement, leading to severe cracking and leakage. Existing settlement sensing devices, such as stress-strain sensors, fiber-optic sensors, and inspection cameras, are often expensive and complex to install. This study proposes a more efficient and simplified vision-based method for radial monitoring of utility corridor sections using multiple feature planes. Methods: This study employed a template matching method to track target movements across multiple planes. By tracking predefined targets and detecting circles within the region of interest using the Hough circle transform, spatial changes were recorded. The template matching algorithm determined the spatial position of detection targets in consecutive frames for each monitoring section. The matching algorithm generated a similarity index for the detection target, and by integrating all detection results, a similarity matrix could be obtained. This matrix helped detect target positions across frames by mapping indices of the extrema and scaling factors to the original frames. The proposed method then adjusted and integrated these scaling factors to achieve real-time settlement detection of multiple radial sections. The underground utility corridor beneath the Ciyunsi Bridge in Beijing was used as a case study to validate this method. Results: The experimental results yield the following major findings: (1) Detection errors increase as sections move further from the camera but stabilize over time. After five days, errors for sections at 15 m and 30 m converge faster, reaching -5 mm and -8 mm, respectively. (2) Clearance convergence errors can mirror settlement trends, with smaller errors near the camera and larger for sections further away. Yet, all errors converge to specific boundary values. Sections at 45 m and 60 m, which have larger errors, converge to -12 mm and -16 mm, respectively. (3) Environmental factors have minimal impact on errors, particularly in sections close to the camera. Temperature and humidity have a greater impact on the 45-m section, but the correlation coefficient is still low, indicating a limited effect on errors. The correlation between radial convergence errors and environmental factors is similarly low, showing that environmental impacts are minimal. Conclusions: This study introduces a reliable detection technique leveraging computer vision detection technology by overlaying multiple detection sections and independently adjusting scaling factors. By harnessing radial space features in utility corridors and overlaying independent detection sections, the method enhances the data collection efficiency of the detection equipment. Additionally, overlaying scene pixel points improves data storage efficiency.