The Phantom Camera is a high-speed camera when you need specs like 1920 x 1200 resolution at 1,380 fps. The Columbia Team used it in the granular flow studies. We found that the sales rep was very generous in allowing in-lab testing of a camera before deciding whether to purchase or not.
"It's becoming clear that the camera we plan to deploy in the centrifuge needs to have rather higher resolution, frame rates and sensitivity if we are to resolve granular motions in the detail we'd really like. The prosumer JVC has max 600fps and modest compressed res, while the machine vision cameras we've looked at have >mega-pixel res but not much higher max fps. So, I've started looking at Phantom cameras as a higher-speed option. They are very expensive - think around $50k each - but they can provide >>1000fps at >megapixel res. Here's an example: the Miro LC320S."
The Columbia team identified the following pros and cons and ultimately decided to use it
1) The sensor. One of the best i've ever worked with. Amazingly fast and sensitive. Large sensel sizes (10 micron).
2) Frame rates. Up to 8,300 fps at VGA resolutions. Much faster at lower resolutions. Fast enough to study an enormous range of problems.
3) Lens flexibility. Interchangeable mount plates (F-mount, C-mount, and many others). Lens options almost unlimited provided that the image circle is >1".
4) Capture software. It's full featured, very mature, and very free
1) Record times (video goes to RAM as a raw uninterpolated uncompressed cine file). Assuming 12 GB of RAM:
- 2 minutes with full frame at 30 fps
- 4.5 minutes witn 1 megapixel at 30 fps
- 2.7 s with full frame at 1380 fps
2) Cost. This is not a cheap camera.
3) Windows only SDK and control software. As far as I can tell, all SDKs are built on Windows DLLs, which means that it may be very difficult if not impossible to build autonomous control applications on any platform other than Windows. There appear to be no cross-platform cine file readers available. However, the spec for cine files (raw, packed, and interpolated) is well documented and pretty simple (see attached PDF). It should be super easy to write a cross-platform reader in Matlab, C/C++, or Python.