Quote Originally Posted by Con View Post
If an object is moving across the screen, such as a warthog driving by, then wouldn't the position data between the front of the hog in one frame and the back of the hog in the next be almost the same? The only major change would come from the front and back of the hog where it's compared with the background. I'm not sure how you would pull objects out of the scene. On the plus side, that new SSDO method works using depth and position data doesn't it?
Yep, this seems to be the problem iv'e been coming up against. Seems the way I thought it was implemented isn't quite right. I initially thought I could compare the positions of two frames at each pixel, but as you say, doing this means the difference between frames would be greatest at the boundry of a moving object and wouldn't give a per object velocity.

What I actually need to do is get the velocity of each vertex when the objects are initially drawn. However I can't think of a feasible way to do this without some major editing of vertex shaders and the like. So an accumulation buffer motion blur will have to do. Position information will still be available for other post process usage though .