ray-box intersection

Question

I've searched everywhere for an equation that suits my needs but i couldn't find anything useful. I'm making a voxel ray-tracer so I need an equation that gets me the intersection of a ray with a box, and the box's normal.

I looked at inigo quilez's intersection code but I can't derive the equation from the code shown so i need the equation itself rather than code.

---

Say we have the ray's origin $O(-5, 1, 0)$ and the box position $B(0, 0, 0)$ with the ray's direction as a unit vector being $(1, 0, 0)$ and the box's side lengths being $(2, 2, 2)$

1. I need the equation to return the distance between the point where the ray hits the box, and the ray's origin

2. I need the box's normal where the intersection is

Answer 1

PBRT has code for axis-aligned bounding boxes.

https://www.pbr-book.org/3ed-2018/Shapes/Basic_Shape_Interface#Bounds3::IntersectP

They only have an intersection predicate for occlusion rays, but extending this to give you the normals and support non-axis-aligned bounding boxes is simple.

To acquire the normals, you create a normal in the proper direction for the hit slab. It loops through the x, y, and z directions, and since this is axis aligned, you return (1, 0, 0), (0, 1, 0), or (0, 0, 1). There is also a check to see if the hit values are reversed, in which case the normals are reversed, e.g. (-1, 0, 0).

If you want to support non-axis-aligned bounding boxes, it's probably easiest to apply the inverse of the box's transform to the intersection ray and then do the axis-aligned test as above. In the end, you apply the transform to the output normals (remember that transforming normals has different rules than transforming vectors).

Answer 2

Denote you ray as $\vec{P} + \vec{D}*t $ and $\vec{S}_h$ be the half size of the box (ie $ \vec{S}_h = (1,1,1)$). The following equation must be satisfied (with abusive notation): $ || \frac{\vec{P} + \vec{D}*t}{\vec{S}_h} - 1 ||_{L_{\infty}} = 1 $ where vector-division is defined as $\frac{\vec{a}}{\vec{b}} = (a_1/b_1,a_2/b_2,a_3/b_3,...)$ and subtraction $\vec{a} - b = (a_1-b,a_2-b,a_3-b,...)$. Note that the max norm is used. There are branchless solutions to finding both solutions to $t$ in this equation.

To get the normal after computing $t$: evaluate the vector inside the norm, find the basis vector corresponding to the component with largest absolute value and multiply it by the sign of the component.