Abstract
Traditional B-spline surface reconstruction often struggles with sparse and non-uniformly distributed data, leading to high computational costs and memory overhead. This paper presents the Key-Point Interpolation (KPI) method, a two-stage algorithm that integrates Kriging and parameter perturbation to efficiently reconstruct smooth, dynamic surfaces. The method was validated using meteorological data from Shanxi Province, demonstrating high precision and significant data compression.
Introduction
In fields such as reverse engineering, meteorology, and geographic information systems, the ability to transform scattered data points into continuous, smooth surfaces is critical. B-splines are widely used for this purpose due to their global smoothness and local controllability. However, standard B-spline algorithms typically require data to be organized in regular grids. Real-world datasets, such as LiDAR scans or weather station readings, are frequently sparse and non-uniform. Attempting to fit B-splines to such data often results in an excessive number of control points, causing computational bottlenecks. This study introduces the KPI method to address these challenges, enabling efficient surface reconstruction from sparse, dynamic datasets.
Methodology
The KPI method operates in two primary stages: dataset generation with parameter optimization and control point solving.
Dataset Generation and Parameter Perturbation
The initial step involves defining the reconstruction domain using Constrained Delaunay Triangulation (CDT). To manage the complexity of sparse data, the authors implemented a parameter perturbation method. This optimization technique merges adjacent data parameters ($u$ and $v$ coordinates), effectively reducing the number of variables in the system while preserving the topological integrity of the mesh.
Following perturbation, Kriging—a geostatistical interpolation technique—is employed to estimate values at unobserved locations. This process generates a temporary gridded dataset that serves as a framework for the subsequent B-spline fitting.
Control Point Solving
The core of the KPI approach is a multi-level iterative solver. The algorithm distinguishes between two types of data:
- Key Points: These are the original, high-fidelity sparse data points that the final surface must interpolate exactly.
- Non-Key Points: These are the synthetic points generated via Kriging, used primarily to guide the surface's global trend and ensure smoothness.
By solving for control points across three hierarchical levels—fitting points, then curves, and finally surfaces—the method reconstructs a 3D dynamic surface that accurately reflects temporal changes.
Experimental Validation
The researchers tested the KPI method using a real-world meteorological dataset from Shanxi Province, China. The dataset comprised hourly temperature readings from 98 weather stations over a 48-hour period (October 15–17, 2017). To maintain efficiency, the complex geographical boundary of the province (originally 6,102 points) was simplified into a 49-vertex polygon.
Results and Discussion
The KPI method demonstrated exceptional performance in both accuracy and efficiency:
- High Precision: The reconstructed dynamic surface achieved a maximum interpolation error of $9.95 \times 10^{-14}$, indicating near-perfect fidelity to the original station data.
- Data Compression: The parameter perturbation method significantly reduced the model's complexity. The number of $u$ parameters was decreased from 124 to 66, and $v$ parameters from 120 to 68, leading to faster computation and lower memory usage.
- Surface Quality: The resulting surfaces were smooth and physically consistent, capturing the dynamic evolution of temperature gradients across the province.
The distinction between "Key" and "Non-Key" points proved vital. It allowed the model to leverage the predictive power of Kriging for global trends without inheriting its local inaccuracies, ensuring that the final B-spline surface remained anchored to the actual observed data.
Conclusion and Future Work
The KPI method provides a robust framework for reconstructing dynamic surfaces from sparse, non-uniform data. By combining the smoothness of B-splines with the predictive capabilities of Kriging and the efficiency of parameter perturbation, the authors have created a tool suitable for high-precision meteorological and engineering applications. Future research may focus on extending this logic to higher-dimensional data (e.g., 4D medical imaging) or integrating multiple variables, such as humidity and wind speed, into a single multi-variate dynamic model.
Summary updated