CALIBRATED MULTICHANNEL MONOCULAR RANGING FROM STANDARDIZED LICENSE PLATES WITH METROLOGY-EXACT VALIDATION

Can a single, ordinary camera provide reliable metric distance to a leading vehicle without relying on active sensors like radar or lidar? This paper investigates whether the standardized physical dimensions of U.S. license plates—which are legally regulated and present on every road vehicle—can serve as a passive, low-cost metric anchor for distance estimation.

Instead of relying on learned depth networks that often lack absolute scale, the authors use a geometric approach. The system treats the license plate as a target of known physical dimensions. It extracts three distinct features from the plate: the height of the serial characters, the width of the outer plate frame, and the span between mounting holes. By inverting the pinhole camera projection, each feature provides an independent distance estimate. The authors introduce a robust pipeline that includes:

• Consensus-Gated Fusion: A method that weights each channel by its inverse variance and automatically discards corrupted readings that deviate significantly from the median.
• Pose Correction: A 3D attitude recovery process that corrects for foreshortening caused by the plate's tilt and yaw relative to the camera.
• MetroSim: A metrology-exact simulation testbed that provides ground-truth distance data across all 51 U.S. jurisdictions, allowing for precise validation without the noise of on-road surveying.

• Accuracy: The system achieves a mean absolute percentage error of 5.79% (median 2.17%). The plate outline is the most reliable ruler, while the character height serves as a useful close-range refinement.
• Robustness: The multi-channel fusion ensures that if one feature is obscured (e.g., by an aftermarket frame or dirt), the system maintains a distance estimate using the remaining visible features.
• Independence: The method works across diverse vehicle classes and lighting conditions, providing a fault-tolerant complement to active sensors like radar, which have different failure modes.