raytracer

What

raytracer is an exercise in linear algebra used to render delicious-looking objects. It supports diffuse and specular lighting, shadows, antialiasing, and recursive rendering for reflections. The only catch is: you can only render spheres.

Try it here!

Current Goals

• Dare I say... polygons?! Normals would be easy, though I am foreseeing great troubles if I don't reduce the intersection-search range. Perhaps a convex 'sleeve' for complex objects that work as a stepping stone. If the larger sleeve isn't intersected, a polygon within the sleeve definitely is not. How to make the sleeves... Spheres? Rastering onto screen?
• Also, a good stepping stone (conceptually) to a 2d rendering engine for games. (like hex, hopefully)

---

Why

It's interesting. Also, it's a great example of how essential linalg is in "the real world".

---

How

Implementation

Mostly a combination of a very long flight and javascript. Positions are vectors, colors are (almost) vectors, material are vectors, everything is vectors!

It starts with a camera and a virtual screen in front of it. Depending on the viewing resolution and multisampling size, a ray is sent from the camera to roughly at where each pixel would correspond to in the virtual screen.

If the ray intersects an object, determined by running a quadratic equation on the 3d limits of the objects and the direction of the ray, the color found at the closest intersection determines the color that will be displayed on the pixel that corresponds to that ray. (or the virtual pixel it was sent through mapped to the real pixel we see)

The color at the intersection is determined in a lengthy process: First we cast a ray at it from every light in the scene. Occluded rays drop shadows (essentially darken the color at that location). Non-occluded rays add color to the location based on the angle of the ray falling on the object and the material properties of the object the location is on. Finally, we reflect the original ray that was cast from the camera off of the current location, and if it intersects another object, we repeat the color-determination process recursively until we reach a predetermined depth. (Unfortunately, a hall of mirrors would fry my poor laptop.) We add these colors based on the reflectivity property of the reflected object's material.

Made by following the theory on Avik Das's excellent 3d renderer workshop.

Usage

As it is, you have to edit the file raytracer.js. Due to local development constraints with modules, I had to put everything in one file. :(

Specifically, you can adjust the elements as follows:

• Camera location and FOV:
  • camPosition is a global-scope Vector that determines camera positon.
  • The relative direction of the vectors from the camera to imagePlane's four "corners" will determine FOV.
• Canvas size:
  • screenW and screenH. If the image appears stretched, make sure the distances between imagePlane vectors are proportional to the height and width of the viewport/canvas.
• Lights
  • Lights are pointLight objects with Vector positions and Color diffuseIntensity/specularIntensity properties. They are currently generated in raySphCollisionTest with examples.
• Spheres
  • Spheres are Sphere objects with Vector positions and Material material properties (shininess etc.). They are currently generated in raySphCollisionTest with examples.

Finally, index.html will display a local page with the render on a canvas.