Doesn't look like this article covers this, but you can anti-alias your scene using a method similar to what is covered for soft shadows. Instead of shooting rays back to a random point on a light surface though, you move the camera slightly (randomly) on the XY plane and shoot multiple rays per pixel.
Not only does this AA, but it also naturally offers depth of field blur.. The more you allow the camera to move with each ray, the more depth of field blur you get.
And then if you not only shoot rays from different camera locations, but also spread them out over time, AND you move objects in the scene, you get motion blur :). The greater the delta t you allow, the more motion blur you get.
Raytracing is really damn cool.. Too bad its still too performance intensive for real-time applications.
I think some pretty smart people (Neoptica) are working on this problem - aided by the the general purpose programmability of the latest generation of GPUs.
The nice thing about raytracing is that it scales so well with parallel processing. The more pipelines, the more simultaneous rays you can process at one time. Synchronization issues are minimal because essentially what you are doing is waiting for pixel data to return from all of the rays shot from the camera. Order doesnt really matter. Sounds like a perfect problem for multi-pipelined GPGPU programming.
Hadn't heard of Neoptica before, but they do look like they have some really smart people there. Raytracing itself is an old concept, but I'm sure there are creative ways to optimize and extend the concept yet to be done.
Not only does this AA, but it also naturally offers depth of field blur.. The more you allow the camera to move with each ray, the more depth of field blur you get.
And then if you not only shoot rays from different camera locations, but also spread them out over time, AND you move objects in the scene, you get motion blur :). The greater the delta t you allow, the more motion blur you get.
Raytracing is really damn cool.. Too bad its still too performance intensive for real-time applications.