Real-time refraction through multiple layers using image-space method with inline ray tracing

Sammanfattning: Although modern hardware allows us to use ray tracing in real-time, ray traced refractions are still demanding due to its need to generate multiple rays recursively in order to fully refract through a multi-layered object. With the recent release of inline ray tracing, it is possible to traverse and register ray hits inside a fragment shader. This allows for convenient implementation of ray tracing into past rasterization algorithms and possibly improve on their limitations such as more physically accurate simulation of certain natural phenomena. Image-space refractions is a rasterization algorithm used to simulate refractions in real-time. This algorithm is limited in that it shows difficulties when refracting through concave objects containing more than two layers. This paper implements a new refraction method which combines image-space refractions with inline ray tracing into a Hybrid method in order to remove the limitations caused by image-space refractions and generate physically accurate refractions close to the ray traced results but with less rays used. The Hybrid method is evaluated using an NVIDIA GeForce RTX 2070 SUPER graphics card with screen resolution set to 1920x1080 and compared with three other refraction methods which are (1) ray traced refractions, (2) image-space refractions, and (3) a state-of-the-art refraction method implemented in Avalanche Studios APEX engine. Both visual and rendering speed comparisons are made between the four methods on three different 3D models of varying amount of surface layers and complexity, a 4-layered concave drinking glass, a 2-layered convex sphere, and a concave pillar with a lot of variety in amount of layers. The results reveal that at most, three rays can be saved per pixel using the Hybrid method compared to using fully ray traced refractions while in worst case only one ray is saved, both while still achieving physically accurate refractions. The amount of rays saved using the Hybrid method, cause great improvement in rendering speed compared to ray traced refractions. While the image-space and the state-of- the-art refraction methods are a lot faster, they produce less physically accurate visuals for concave objects containing more than two layers, compared to the Hybrid and ray traced methods.