Where We Left Off: Classical Ray Tracing
What we’ve seen so far is usually called classical ray tracing
• it was the method used by most early ray tracers
To refresh our memories, it most important features are
• trace eye rays through each pixel
• at intersection with nearest surface
– trace rays towards lights to check for shadow
– recursively trace reflected and transmitted rays
• supersample with multiple rays per pixel for antialiasing
Starting with this, we want to look more closely at two issues
• how to make sure we’re ray tracing efficiently?
• how to extend this to encompass more visual phenomena?
Ray Tracing Efficiently
The primary path to efficiency
• avoid tracing rays whenever possible
• and above all, avoid ray–surface intersection tests
A ray tracing system can be easily overcome with rays
• minimum 1 eye ray per pixel, many more with supersampling
• recursion depth kyields 2
k+1
−1rays traced per eye ray
– counting reflection & transmission rays but not shadow rays
Consider this example
• image resolution of 1024x768 = 786,432 pixels
• 3x3 supersampling = 7 million eye rays
• recursion depth 5 = 63*7 = 441 million
• each tested against 10,000 polygons
• 4.4 trillion intersection tests (ignoring shadow rays)
Spatial Data Structures
Probably the single most important efficiency improvement
• divide space into cells
• record what geometry lies in each cell
• first test rays against cell
• only check geometry within cell if the ray actually hits the cell
Several data structures in common practice
• hierarchical bounding volumes
• BSP trees
• octrees
• regular 3-D grids
Hierarchical Bounding Volumes
Begin by intersecting ray with the root volume
• if no intersection, ignore all child volumes
• otherwise, recursively test child volumes
We’ll only test objects whose bounding volume is actually hit by ray
• hopefully ignoring most of the scene
• but this depends a lot on the structure of the hierarchy
