TL;DR
This paper presents a novel method for precise 3D positioning and orientation of mobile laser scanning systems (MLSS) in GNSS-denied subway environments. By integrating dead reckoning with sparse control point data and a four-parameter transformation, the system achieves millimeter-level accuracy, significantly improving inspection efficiency and enabling accurate defect detection.
What Problem Does This Paper Address?
Subway tunnels require frequent and accurate inspections to ensure safety and operational integrity. However, these environments are characterized by a complete lack of GNSS signals, making traditional positioning methods ineffective. Existing inspection methods are often time-consuming and may not provide the required accuracy for precise defect localization. This paper addresses the critical challenge of obtaining high-accuracy absolute 3D positions and orientations for MLSS within these GNSS-denied tunnels, enabling efficient data collection and reliable analysis.
Methods
The core of the proposed method lies in an integrated navigation system that combines an Inertial Measurement Unit (IMU) and an odometer for dead reckoning (DR).
- Initial Alignment: A static initial alignment process is performed to determine the MLSS's initial attitude relative to a navigation coordinate system (NCS). This involves coarse and fine alignment steps, utilizing a Kalman filter to estimate misalignment angles.
- Forward/Backward Dead Reckoning (DR): Once initial alignment is complete, the system employs both forward and backward DR algorithms. These algorithms use IMU and odometer data to continuously calculate the MLSS's position and attitude over time, even without external references.
- Sparse Point-Assisted Correction: The key innovation is the use of sparse control points strategically placed along the tunnel. As the MLSS passes these points, its laser scanner captures their local coordinates.
- Four-Parameter Transformation: A four-parameter transformation method is then applied. This method uses the captured control point data to correct accumulated errors in the DR-derived positions and attitudes. Crucially, it transforms the MLSS's state information directly from the navigation coordinate system (NCS) to a local coordinate system (LCS), eliminating the need for intermediate geodetic transformations. This allows for direct fusion of point cloud data into the LCS.
Datasets
The method was evaluated using data collected from two real-world subway tunnel inspection scenarios:
- Beijing-Zhangjiakou High-Speed Railway Tunnel: This experiment was used to demonstrate the detailed operation process and principle of the proposed method.
- Wuhan Subway Lines 2 and 6: Data from these lines were used to verify the reliability and accuracy of the system in actual working conditions over approximately one year.
The paper does not introduce new benchmark datasets but rather utilizes existing infrastructure and control points for validation.
Key Insights and Ideas
- GNSS Independence is Achievable: The research demonstrates that high-accuracy absolute positioning in GNSS-denied environments like subway tunnels is feasible without relying on external satellite signals.
- Sparse Control Points are Sufficient: A surprisingly small number of control points, even with large intervals (up to 480m), can effectively correct DR errors and achieve millimeter-level accuracy. This significantly reduces the effort required for control point placement and measurement.
- Direct LCS Transformation: The four-parameter transformation method directly mapping from NCS to LCS is a significant simplification, reducing computational load and data processing time compared to traditional multi-step transformations.
- Efficiency Gains: The proposed method drastically reduces inspection time, allowing for the complete inspection of a 5-6 km tunnel in under 4 hours, a substantial improvement over traditional methods.
Key Results
- Beijing-Zhangjiakou High-Speed Railway Tunnel: With a control point interval of 120 m, the 3D coordinate accuracy of the point clouds reached 8 mm.
- Wuhan Subway Lines:
- With a 60 m control point interval, 3D accuracy was 4 mm.
- With a 120 m interval, accuracy was 6 mm.
- With a 240 m interval, accuracy was 7 mm.
- With a 480 m interval, accuracy was 8 mm.
- Inspection Time: A 5-6 km tunnel inspection can be completed in less than 4 hours, including data acquisition, processing, and analysis.
- Accuracy Meets Requirements: The achieved accuracies (down to 4 mm) meet the stringent requirements for subway tunnel inspection and defect detection.
Limitations and Future Work
- Elevation Accuracy Divergence: The paper notes that the INS's inherent error divergence in the elevation direction is a limitation. While control points improve this, further algorithmic optimization for elevation accuracy is suggested.
- Filtering and Feature Recognition: The authors suggest that further research into point cloud filtering and feature recognition algorithms could enhance the extraction of tunnel diseases during post-processing, further reducing workload.
- Algorithm Optimization: The paper implies that the elevation direction's accuracy could be further improved, suggesting potential for future algorithmic enhancements.
Summary updated