TL;DR
This paper introduces a novel method for calibrating out-of-focus projectors in structured light 3D measurement systems. By incorporating lens distortion correction on the projection plane and employing a nonlinear optimization approach, the method achieves higher accuracy compared to existing techniques, particularly when the projector is significantly defocused.
What Problem Does This Paper Address?
Structured light 3D measurement systems rely heavily on accurate calibration of both cameras and projectors. While camera calibration is well-studied, projector calibration, especially for out-of-focus projectors, presents significant challenges. Out-of-focus projectors, commonly used in binary defocusing techniques for faster scanning, introduce distortions that degrade measurement accuracy. Traditional calibration methods often assume the projector is in focus, leading to significant errors when this assumption is violated. This paper aims to address this gap by developing a calibration method that explicitly accounts for the distortions introduced by an out-of-focus projector.
Methods
The proposed method consists of two main stages: coarse calibration and final calibration.
- Coarse Calibration: This stage provides good initial values for the final calibration.
- Focal Plane Calibration: The projector is first treated as an inverse camera and calibrated using a pinhole camera model when it's focused on its focal plane. The resulting intrinsic and extrinsic parameters serve as initial estimates.
- Projection Plane Distortion Estimation: The projector is then adjusted to focus on the projection plane. Lens distortion on this plane is estimated using the pinhole camera model.
- Final Calibration: This stage refines the parameters using a nonlinear optimization algorithm.
- Distortion Correction: A high-order radial and tangential lens distortion model is applied to the projection plane to correct for distortions that arise when the projector is out of focus.
- Nonlinear Optimization: A re-projection mathematical model, incorporating the estimated distortions, is used. The objective is to minimize the sum of re-projection errors of reference points on the projector image plane. The Levenberg-Marquardt method is employed for this optimization.
A key aspect of the method is the "Phase-Domain Invariant Mapping" which establishes a unique pixel-to-pixel correspondence between the camera's CCD and the projector's DMD by utilizing phase information from both horizontal and vertical fringe patterns. This mapping is crucial for obtaining accurate calibration points on the projector.
Datasets
The paper describes experiments using a calibration board with a 7x21 array of white circles. This board was used to capture calibration images in various poses. The experimental system included a DLP projector and a camera. No specific public datasets were mentioned; rather, custom calibration data was generated for the experiments.
Key Insights and Ideas
- Out-of-focus projectors have significant distortions: The paper experimentally demonstrates that projectors exhibit noticeable distortions when operating outside their focal plane, and these distortions increase with the degree of defocusing.
- Distortion correction on the projection plane is crucial: Unlike traditional methods that might assume a perfect pinhole model, this work highlights the importance of explicitly modeling and correcting lens distortion on the projection plane for accurate out-of-focus calibration.
- Two-stage calibration for robustness: The proposed two-stage approach (coarse calibration for initial values, followed by nonlinear optimization) provides a robust way to find accurate parameters, especially when dealing with complex distortions.
- Phase-domain mapping is key for projector points: Establishing a reliable mapping between camera pixels and projector pixels using phase information is essential for obtaining accurate calibration data for the projector.
Key Results
- The proposed method achieved significantly reduced re-projection errors for the out-of-focus projector compared to a conventional method [29]. Specifically, re-projection errors decreased by 27.01%.
- Measurement accuracy was improved by 34.96%, and uncertainty was reduced by 26.17% when measuring distances between points on the calibration board.
- Experiments with a planar board showed that the fitting error remained relatively stable across different defocusing degrees with the proposed method, whereas the conventional method showed a rapid increase in error.
- Measurements of a hemisphere also demonstrated superior performance of the proposed method, with lower fitting residuals and less error variation across defocusing degrees compared to the method in [29]. For example, at defocusing degree 5, the fitting residual was 0.0574 ± 0.0685 mm with the proposed method, versus 0.0952 ± 0.0936 mm with the conventional method.
Limitations and Future Work
- Computational Cost: The paper acknowledges that the proposed method's computing time is somewhat longer compared to traditional calibration methods.
- Future Work: While not explicitly detailed, the success of this method suggests future work could focus on further optimizing the computational efficiency, exploring more advanced distortion models, or applying this calibration technique to other types of structured light systems. The authors' conclusion implies that the method is robust across various defocusing amounts, suggesting its applicability in a wide range of scenarios.
Summary updated