augmented reality - Occlusion of real-world objects using three.js


Keywords:three.js 


Question: 

I’m using three.js inside an experimental augmented-reality web browser. (The browser is called Argon. Essentially, Argon uses Qualcomm’s Vuforia AR SDK to track images and objects in the phone camera. Argon sends the tracking information into Javascript, where it uses transparent web pages with three.js to create 3D graphics on top of the phone video feed.) My question is about three.js, however.

The data Argon sends into the web page allows me to align the 3D camera with the physical phone camera and draw 3D graphics such that they appear to align with the real world as expected. I would also like to have some of the things in the physical world occlude the 3D graphics (I have 3D models of the physical objects, because I’ve set the scene up or because they are prepared objects like boxes that are being tracked by Vuforia).

I’m wondering if folks have suggestions on the best way to accomplish this occlusion with three.js. Thanks.


1 Answer: 

EDIT: it appears that the next version of three.js (R71) will have a simpler way to do this, so if you can use the dev branch (or just wait), you can do this much more easily. See this post: three.js transparent object occlusion


MY ORIGINAL ANSWER (without using the new features in R71):

I think the best way to do this is (to avoid extra work by creating new rendering passes for example) to modify the WebGL renderer (src/renderers/WebGLRenderer.js) and add support for a new kind of object, perhaps call them “occlusionObjects”.

If you look in the renderer, you will see two current object lists, opaqueObjects and transparentObjects. The renderer sorts the renderable objects into these two lists, so that it can render the opaque objects first, and then the transparent objects after them. What you need to do is store all of your new objects into the occlusionObjects list rather than those two. You will see that the opaque and transparent objects are sorted based on their material properties. I think here, you may want to add a property to an object you want to be an occluder (“myObject.occluder = true”, perhaps), and just pull those objects out.

Once you have the three lists, look what the render() function does with these object lists. You’ll see a couple of places with rendering calls like this:

renderObjects( opaqueObjects, camera, lights, fog, true, material );

Add something like this before that line, to turn off writing into the color buffers, render the occlusion objects into the depth buffer only, and then turn color buffer writes back on before you render the remaining objects.

context.colorMask( false, false, false, false);
renderObjects( occluderObjects, camera, lights, fog, true, material );
context.colorMask(true, true, true, true);

You’ll need to do this in a couple of places, but it should work.

Now you can just mark any objects in your scene as “occluder = true” and they will only render into the depth buffer, allowing the video to show through and occluding any opaque or transparent objects rendered behind them.