Real-time GPU surface curvature estimation on deforming meshes and volumetric data sets

Area of Science:

• Computer Graphics
• Geometric Modeling
• Scientific Computing

Background:

• Surface curvature estimation is vital for computer graphics applications like texture synthesis and shape representation.
• Current methods primarily rely on CPU algorithms, limiting real-time performance for dynamic geometry residing in GPU memory.

Purpose of the Study:

• To develop a GPU-accelerated algorithm for real-time curvature estimation on arbitrary triangular meshes.
• To address the performance bottleneck of transferring GPU-resident geometry to the CPU for curvature calculations.

Main Methods:

• A novel GPU algorithm for real-time curvature estimation on triangular meshes.
• Implementation using a graphics pipeline and CUDA.
• Demonstration with curvature-based nonphotorealistic rendering (NPR) feature lines and ambient occlusion approximation.

Main Results:

• The GPU algorithm achieves real-time curvature estimation on arbitrary triangular meshes.
• Demonstrated applications include NPR feature lines, ambient occlusion approximation, and curvature computation on volumetric data.
• The GPU implementation is up to ~18x faster than a multithreaded CPU benchmark.

Conclusions:

• The proposed GPU algorithm offers a significant performance improvement for surface curvature estimation.
• Enables efficient real-time curvature analysis for dynamic and GPU-resident geometric data.
• Facilitates advanced graphics techniques and volumetric data processing on the GPU.