Exact and Adaptive Signed Distance Fields Computation for Rigid and Deformable Models on GPUs.

Area of Science:

• Computer Graphics
• Geometric Modeling
• Real-time Simulation

Background:

• Existing methods for real-time signed distance field construction often rely on approximate distances.
• Efficient and exact distance field generation is crucial for advanced simulations.
• GPU acceleration offers potential for significant performance improvements in geometric computations.

Purpose of the Study:

• To develop a GPU-based method for constructing exact adaptive distance fields.
• To improve the speed and accuracy of distance field generation for complex models.
• To enable high-fidelity real-time simulations of dynamic scenes.

Main Methods:

• Utilized a GPU to construct an exact adaptive distance field from an octree using Morton codes.
• Employed rectangle-swept spheres for bounding volume hierarchy (BVH) construction around triangulated models.
• Implemented a multi-BVH structure to enhance workload balance across GPU processors and an octree-based distance upper bound to optimize queries.

Main Results:

• Achieved construction speeds 35-64 times faster than serial CPU implementations.
• Enabled real-time simulation of fabric interacting with the Stanford Bunny at 20 frames per second.
• Demonstrated the effectiveness of the exact adaptive distance field for complex geometric interactions.

Conclusions:

• The proposed GPU-accelerated method provides an efficient and exact approach to adaptive distance field construction.
• This technique significantly advances the capabilities for real-time geometric simulations.
• The method offers a substantial performance gain over traditional CPU-based approaches.