The HaToy Evaluation Scene
The scene in the images is USAAR's test scene 'HaToy', which exhibits a range of different motions (and object parallax) to illustrate the camera array working. Each of the viewpoints in the video is from a different camera and ALSO from a different time. This is "5D" video, where all the rays captured have different timestamps. That's why each view is from a slightly different perspective but also from a slightly different time. The rotating CD on the left looks different in each view.
The objects in the scene are:
- Very fast motion (spintop, CD drive). CD drive is controlled to 2400 rpm, which is 40 rps, so that when sampling with 40 Hz you can only see parts of the surface.
- Slow but complex motion (mobile)
- Slow but not translatory motion (trains) with occlusion
Conclusions:
- A rig offers the possibility to "virtually" increase the frame rate (of course at the cost of reducing the angular resolution at a certain point in time). While for static scenes it doesn't matter when you expose the cameras, for moving scenes there might be content that's visible only at certain instances of time. In those cases higher temporal resolution could reveal "more important" information than higher angular resolution. For those scenes it is "advantageous" (actually not "required") to create a virtually higher temporal sampling frequency by sub-framing.
- This also explains why we used this strange scheme of 2D bit reversal permutations: You want to have adjacent cameras (those who are "angular neighbours" and hence have the biggest content overlap) to span all motion phases. So in our case for e.g. 4 sub-frames each 2x2 sub-array covers exactly the four different motion phases. Hence we (in the overlapping area of the scene) can see the same content shot at different times.
- The CD drive is an excellent example: It's spinning with one round per frame, i.e. 0.25 rounds per sub-frame if four sub-frames are used. And indeed you should be able to detect a roughly 90° rotation of the coloured CD surface in each of the four cameras of a 2x2 sub-array.
- Using this set up, we discover that you can virtually increase the frame rate up to 160Hz temporal sampling frequency, at the cost of reducing angular resolution.