Abstract
An active solution method for homography, derived from four laser lines, is proposed to recover pavement cracks captured by a camera to their real dimensions. The system, comprising a camera and four laser projectors, reconstructs projection points in the camera coordinate system. Laser lines are optimized using Plücker matrices to model projection points on the pavement. This allows for the active determination of the image-pavement homography. Experimental results demonstrate sub-millimeter recovery accuracy, with a minimum error of 0.78 mm, highlighting the system's potential for automated vision-based pavement inspection.
Introduction
Maintaining road infrastructure is a significant logistical challenge. Vision-based inspection using vehicle-mounted cameras offers a cost-effective solution for detecting pavement cracks. However, translating 2D image pixels into real-world dimensions requires solving the homography between the camera and the road plane.
In dynamic environments, this homography matrix is not constant. Vehicle vibrations and changes in load cause the camera to tilt and shift relative to the pavement, rendering static calibration ineffective. While 3D laser scanners provide high precision, they are often prohibitively expensive and sensitive to environmental conditions. This research proposes an "active" solution using four laser lines to dynamically calculate the homography for every frame.
Methodology
The proposed measurement system consists of an industrial camera and four line-laser projectors mounted on a rigid frame. The recovery process follows a multi-step geometric pipeline:
Calibration and Optimization
The laser lines are first calibrated within the camera’s coordinate system using a checkerboard target at various positions. The system employs Plücker matrices to represent these lines, simplifying the mathematical projection onto planes.
Active Homography Solution
During operation, the lasers project four points onto the pavement. A 2D-2D homography has eight degrees of freedom; since each laser point provides two constraints, four points represent the mathematical minimum required to solve for the homography matrix ($H$) in real-time.
Crack Recovery
Once $H$ is determined for a specific frame, detected cracks in the image are mathematically warped back to their true shape and size on the pavement plane. An optimization function is applied to refine the laser line parameters, which was found to reduce measurement errors by an average of 20.33% compared to basic initialization.
Experimental Results
The system was validated using a 150mm × 150mm checkerboard target and real pavement samples.
- Accuracy: The system achieved a minimum recovery error of 0.78 mm at a measurement distance of 700 mm and a relative angle of 8°.
- Robustness: The optimized method maintained high precision even as the camera angle increased to 16°, significantly outperforming non-optimized initialization methods.
- Optimal Configuration: Results indicated that a moderate distance (700mm) and a slight tilt (8°) provided the best accuracy, likely due to the optimal distribution of laser points within the camera's field of view.
Conclusion
The active homography solution provides a robust and efficient method for recovering real-world dimensions of pavement cracks from 2D images. While the system demonstrates high precision, future work will address challenges such as sunlight interference and non-planar road surfaces to enhance reliability in diverse field conditions.
Summary updated