TL;DR
This paper introduces a novel method for reconstructing 3D surfaces from "quasi-scattered" data, commonly generated by coordinate measuring systems. By combining NURBS curve fitting, a new curvature-based resampling technique, and iterative projection optimization, the method achieves accurate and efficient surface reconstruction, outperforming existing approaches in terms of speed.
What Problem Does This Paper Address?
Accurately reconstructing 3D surfaces from measured data is crucial in fields like CAD, reverse engineering, and object shape detection. A common challenge arises from data acquired by 3D coordinate measuring systems (CMMs and PCMMs). This data, termed "quasi-scattered," is not perfectly gridded nor completely random; it's ordered by measurement sequence but has varying numbers of points per row. Existing surface fitting methods often struggle with this data structure, requiring extensive pre-processing or failing to leverage the inherent order. This paper aims to bridge this gap by developing an efficient and robust method specifically for quasi-scattered data.
Methods
The proposed method is a three-stage process:
- NURBS Curve Fitting (Row-wise): Each row of quasi-scattered data is first fitted with a Non-Uniform Rational B-Spline (NURBS) curve. This stage employs an iterative projection optimization scheme, merging data parameterization and point projection to speed up convergence. Newton iteration is used for efficient projection.
- Curvature-Based Resampling: A novel resampling technique is introduced to address the varying number of points per row. This method leverages the parameterization results from the curve fitting and the curve's curvature. It samples more points in regions of high curvature and fewer in gentler areas, ensuring an equal number of resampled points per row while preserving geometric properties. A Gaussian low-pass filter is used to smooth curvature variations.
- NURBS Surface Fitting (Hierarchical): The resampled data, now effectively gridded, is used to construct the final NURBS surface using a hierarchical fitting approach. This stage also utilizes iterative projection optimization, building upon the results of the curve fitting and resampling. Uniform weights are applied for simplicity and efficiency.
Datasets
The paper evaluates the proposed method using:
- Simulated Data: Six examples of surfaces are used, including a quadric surface, a hat surface, a shell surface, a posit surface, a pump surface, and a blade surface. These are generated from known mathematical equations.
- Real-Measured Data: Data from a pump surface (Example 5) and a blade surface (Example 6) are used, which were originally acquired by measurement machines.
The paper does not introduce new benchmark datasets but rather uses existing examples and real-world data to demonstrate the method's effectiveness.
Key Insights and Ideas
- Leveraging Quasi-Scattered Data Structure: The core insight is that the "quasi-scattered" nature of the data, while a challenge for traditional methods, can be exploited. The ordering within rows provides valuable information that can be used for efficient fitting.
- Curvature-Driven Resampling: The novel resampling approach is a key innovation. By focusing on curvature, it ensures that important geometric features are captured accurately, even when the original data density varies significantly. This is crucial for maintaining fitting accuracy.
- Integrated Parameterization and Projection: Merging parameterization and point projection within the iterative optimization loop significantly reduces computational cost and improves convergence speed. This "boning together" of these steps is a practical advantage.
- Robustness to Data Scatterness: The method demonstrates robustness to varying degrees of data scatterness, including simulated u-fluctuation and row-spacing variations, with minimal impact on fitting accuracy.
Key Results
- Efficiency: The proposed method is significantly faster than existing approaches, particularly for quasi-scattered data. For example, fitting a blade surface took only 17 seconds, compared to hours for some other methods.
- Accuracy: The method achieves high fitting accuracy, comparable to or better than existing methods, with average fitting errors in the order of 10⁻³ to 10⁻⁷ mm across various examples.
- Robustness: The method shows good robustness to variations in data distribution and even to missing data points, with the additional error introduced by resampling being negligible.
- Versatility: The method is demonstrated to work well on a variety of surface types, from simple quadrics to complex free-form shapes.
Limitations and Future Work
- Local Modification: The paper mentions that local modification (often used in shape design) is not considered in the proposed method. This could be a limitation for applications requiring fine-tuning of specific surface regions.
- Uniform Weights: Uniform weights are applied for simplicity. While this is efficient, it limits the ability to emphasize certain areas of the surface during fitting.
- Future Work: The paper doesn't explicitly detail future work, but the robustness and efficiency demonstrated suggest potential for broader applications in industrial metrology and reverse engineering. Further exploration of incorporating local modifications or adaptive weighting schemes could enhance its capabilities.
Summary updated