Screen-Space Voxel Ray-Casting GPU Pipeline Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing voxel rendering technologies face performance issues when rendering high-resolution virtual reality experiences at high framerates and lack optimized stereoscopic rendering capabilities, particularly for commercial devices targeting 2160×1200 pixels at 90 Hz, and are unable to handle rendering primitives other than voxels effectively.
Innovation Solution
A method for screen-space voxel ray-casting using a GPU pipeline that optimizes rendering by sorting primitives from front to back, storing information in per-pixel data structures, and employing a two-pass convolution screen space kernel to select intersection candidates, enabling efficient rendering of voxels and polygons as discrete points and continuous surfaces, respectively, along with real-time modeling and stereoscopic rendering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If sparse voxel octrees rendering is used, then volumetric rendering capability is achieved, but rendering performance deteriorates for high-resolution virtual reality experiences at high framerates
Solution Approach 1:
The rendering process is segmented into two distinct passes: a first pass that processes all primitives (polygons and voxels) through rasterization to populate per-pixel data structures, and a second pass that performs convolution-based voxel ray-casting only for pixels requiring volumetric rendering. This segmentation allows the system to maintain volumetric rendering capability while avoiding the performance penalty of applying voxel processing to all pixels.
Solution Approach 2:
The patent applies voxel ray-casting selectively rather than universally. By using a convolution kernel to identify only those pixels that require volumetric rendering (those where voxels are the closest primitives), the system performs partial action—applying the computationally intensive voxel processing only where necessary, thus maintaining performance while achieving the desired volumetric rendering effect.
2Area of stationary object
If voxel ray-casting is performed on the whole screen, then complete volumetric rendering coverage is achieved, but computational complexity increases
Solution Approach 1:
The patent extracts and isolates the voxel ray-casting operation from the general rendering pipeline. By separating the identification of voxel candidates (through convolution) from the actual ray-casting computation, and by limiting ray-casting to only those pixels identified in the first pass, the system reduces overall computational complexity while maintaining complete coverage where needed.
Solution Approach 2:
The first pass performs preliminary work by rasterizing all primitives and populating per-pixel data structures with information about closest primitives. This preliminary action prepares the data in advance, allowing the second pass to perform convolution and voxel ray-casting efficiently without redundant computations, thus reducing overall computational complexity.
3Manufacturing precision
If primitives are rendered as continuous surfaces, then visual quality is improved, but pixel fill rate optimization is reduced
Solution Approach 1:
The patent applies different rendering strategies to different types of primitives based on their local characteristics. Polygons are rendered as continuous surfaces with full shading information to maintain high visual quality, while voxels are rendered as discrete points with size and shading stored in per-pixel data structures to optimize pixel fill rate. This local differentiation allows each primitive type to be rendered in the most efficient manner for its specific requirements.
Data Source
AI summary
A method for screen-space voxel ray-casting of scenes on a whole screen. The method comprises providing a graphics card; rendering seamlessly any one of the following list of primitives containing at least polygons, voxels; using a GPU pipeline of the graphics card for rasterization; rendering of the primitives sorted from front to back; thereby optimizing a 2D pixel fill rate by in case the primitives are voxels, rendering the voxels as discrete points and storing size and shading information into a per-pixel data structure; storing information about closest primitives into the per-pixel data structure as neighboring 2D pixel data structures; in case the primitives are polygons, rendering the polygons as continuous surfaces and storing shading information into the per-pixel data structure. The method further comprises performing the screen-space voxel ray-casting of scenes on the whole screen by means of the GPU pipeline, thereby selecting intersection candidates from the neighboring 2D pixel data structures; thereby optimizing the screen space ray-casting of voxels by selecting the intersection candidates using a two-pass convolution screen space kernel of fixed size; and identifying the kernel size per frame, based on the distance to the closest voxel from the camera.