For part of my final project in CSE 872: Advanced Graphics, I did a variety of implementations of Real-Time Motion Blur. It was a fun project, and this blog might see me post more implementations in the future (I've been playing around with a velocity buffer based one), but for now:
Technique 1: An Accumulation Buffer
This is a really easy technique that essentially is just the averaging together of several frames, causing motion from dynamic objects and camera motion to become blurred. The reality is that is really just impractical, but converges to a correct solution if a small enough time step and enough samples are used. But in a real-time implementation, this will almost never be feasible. As a result, under-sampling known as "ghosting" can be seen. Check out this shot from my project:
Technique 2: Motion Blur as a Post Process
This is a technique presented in the excellent GPU Gems 3, where world space positions are reconstructed from the depth buffer during post processing. The difference between the current frame's position and the previous frame's position will produce a velocity vector that can then be used to blur the image. However, this technique only works for static geometry (camera motion), but the results are good:
Technique 3: Just Fake It with a Radial Blur already
This one I actually used in a game project recently that I will probably be posting about in the near future. This is something that Real-Time Rendering points out about Assassin's Creed- the effectiveness of a simple radial blur looks very favorable for certain instances of forward motion. We use it for the boos functionality on the vehicle in my game, and Assassin's Creed uses it for chase sequences. Here's a shot of my demo scene:
So that sums up my current round of motion blur techniques for my final. I might talk about the Depth of Field component of my project, but then again maybe not.