Abstract
Calibrating multiple LiDAR sensors mounted on fixed infrastructure is a significant challenge for smart city applications. Traditional calibration methods often rely on sensor motion or complex environmental geometries. This paper introduces an automated "plug-and-play" solution using two non-parallel poles wrapped in retro-reflective tape. By exploiting the high-intensity returns from the tape, the system identifies poles and determines extrinsic parameters through a least-squares optimization process. Results show that this method achieves high precision with zero manual initialization, performing competitively against established planar-based techniques.
Introduction
As Cellular Vehicle-to-Everything (C-V2X) technology matures, urban infrastructure is increasingly equipped with LiDAR sensors to provide a comprehensive "bird's-eye view" for autonomous vehicles. These sensors, often mounted on lampposts, help eliminate blind spots and improve safety. However, creating a seamless 3D map requires precise extrinsic calibration between multiple sensors.
Existing calibration techniques typically require either sensor motion (impossible for fixed infrastructure) or complex manual setups involving checkerboards or perfectly orthogonal walls. The proposed method addresses these limitations by providing a simple, motion-free calibration process that works even in feature-poor environments like open harbors or docks.
Methodology: The LiDAR-Pole Model
The core of the proposed approach is a four-step pipeline designed for simplicity and robustness:
- Pole Extraction: Retro-reflective tape is wrapped around two poles. Because this material reflects LiDAR pulses much more intensely than natural surroundings, the poles can be isolated by simply filtering for high-intensity values in the point cloud.
- Line Representation: The isolated points for each pole are mathematically modeled as 3D line equations ($r = p + \lambda n$).
- Optimization: The calibration is framed as a least-squares problem, seeking the rotation ($R$) and translation ($t$) that best aligns the pole lines detected by different LiDAR units.
- Correspondence Resolution: To resolve the mathematical ambiguity inherent in aligning 3D lines, the Iterative Closest Point (ICP) algorithm is used to select the most accurate transformation from the potential candidates.
Experimental Evaluation and Results
The researchers validated the method using both synthetic simulations and real-world experiments. In simulations, the algorithm demonstrated high accuracy with an average calibration error ($e_{rt}$) of 0.168.
For real-world testing, a dual-LiDAR system (Velodyne and Hesai) was deployed in a corridor. Performance was measured using Rényi’s Quadratic Entropy (RQE), which quantifies the "sharpness" of the merged point cloud. The proposed method achieved an RQE score of 0.0822, nearly matching the 0.0824 achieved by labor-intensive planar surface methods, but without the need for complex geometric environments.
Conclusion
The use of retro-reflective tape transforms calibration from a complex geometric problem into a straightforward intensity-based detection task. By requiring only two non-parallel poles, this method offers a flexible and highly accurate alternative for infrastructure-based LiDAR systems. While the method requires overlapping fields of view and stable intensity data, it represents a significant step toward automated, low-maintenance sensor networks for smart cities.
Summary updated