Ray Casting versus Rasterization

Two ways to make an image: shoot rays from the eye into the scene, or project geometry onto the screen and fill it in.

Two dual approaches

Rasterization asks, for each triangle, which pixels it covers. Ray casting asks, for each pixel, what the scene contains along its ray. They solve the same visibility problem from opposite directions, and each fits different data and hardware.

Rasterization

Kronos motion — training from sim

Rasterization is the backbone of real-time graphics. GPUs are built to project and fill triangles extremely fast, making it ideal for surface geometry like machine models and isosurfaces. Cost scales with geometry and pixels, and it handles opaque scenes with a depth buffer very efficiently.

Ray casting and ray tracing

Ray casting shoots a ray per pixel and finds what it hits. It is the natural method for volume rendering, where the ray integrates through a field rather than stopping at a surface. Ray tracing extends this with secondary rays for reflection, refraction, and accurate shadows, at higher cost.

Rasterization: fast for surfaces, hardware-accelerated, needs tricks for global effects.
Ray casting: natural for volumes and implicit surfaces, cost scales with samples per ray.
Modern GPUs support both and often combine them.

Choosing

Use rasterization for the machine's surface geometry; use ray casting for volumetric fields such as density or flux (see Volume Rendering). Hybrid renderers rasterize surfaces and cast rays through volumes in the same scene.

Kronos use

Surface models render by rasterization for interactivity, while simulated 3D fields use ray-cast volume rendering when interior structure matters.