fps1000 - "low cost high speed camera"

[cloudscapes]

This popped up on a photography forum.

http://www.kickstarter.com/projects/1623255426/fps1000-the-low-cost-high-frame-rate-camera

The price is REALLY competitive. Only a few "consumer-priced" cameras will do 1000fps at low resolutions, this one claims tens times as fast for half a grand or so (there are different tiers). The choice of C mount is logical. Lots of lenses available, and they fit smaller sensors nicely. I have a few that I adapt to my M43 camera.





Boards and proto look good.

[ferrix]

I'm concerned with the lack of sharpness that is apparent in all the example videos.  He claims it's not an inherent limitation of the camera but just down to his photographic skills.  But since he designed the entire camera that doesn't really put me at ease.  Also, all the videos look soft in the same way.

[mikeselectricstuff]

I'm concerned with the lack of sharpness that is apparent in all the example videos.  He claims it's not an inherent limitation of the camera but just down to his photographic skills.  But since he designed the entire camera that doesn't really put me at ease.  Also, all the videos look soft in the same way.

Most cameras do a lot of post-processing to produce "nice" images - could be there isn't any of that (yet) and/or there is filtering to do noise reduction - remember thet exposure times are very short.

Looks like a nice piece of engineering - keep it simple. I would have liked to see a jog dial and external trigger in & strobe out though - I suppose these could be added as it looks like the UI software part is open.
Being able to select footage in-camera for save is definitely the way to go, as you usually only want a very small subset of captured data.

Quick reality check :
Flash memory is a Micron MT29F128G08AMCABH
Only skimmed the datasheet, but a quick reality check on write speeds :
350uS per page (8192+448 bytes), 2 planes per die, 4 dice per device, 2 devices = approx 395MbytesMBytes/sec. 
VGA 640x480x840fps=258Mpixels/sec
So in the right ballpark, probably with some compression

Looks like sensor is Onsemi Python

[mikeselectricstuff]

Some data on the sensors :
http://www.onsemi.com/pub_link/Collateral/PB_NOIP1SN1300A-D.PDF

[Kostas]

I'm concerned with the lack of sharpness that is apparent in all the example videos.  He claims it's not an inherent limitation of the camera but just down to his photographic skills.  But since he designed the entire camera that doesn't really put me at ease.  Also, all the videos look soft in the same way.

Most cameras do a lot of post-processing to produce "nice" images - could be there isn't any of that (yet) and/or there is filtering to do noise reduction - remember thet exposure times are very short....

The image looks rather soft, so the lens could be "soft". It seems that this camera takes C-mount lenses and the one fitted is probably a cctv camera lens, so nothing special, just cheap. Another factor is the aperture used to capture the video. Less light means wider aperture, which means shallower depth of field*, but also - usually - softer images. All lenses have their best sharpness at a specific aperture and there is an ideal range of apertures. Wide open apertures generally give softer images. Who knows, it could be a lack of skills combined with poor optics and wide aperture. The good thing is that there are definitely good lenses out there; the c-mount to eos adapter gives even more choices.

* So, a narrower in-focus range.

[mikeselectricstuff]

I'm concerned with the lack of sharpness that is apparent in all the example videos.  He claims it's not an inherent limitation of the camera but just down to his photographic skills.  But since he designed the entire camera that doesn't really put me at ease.  Also, all the videos look soft in the same way.

Most cameras do a lot of post-processing to produce "nice" images - could be there isn't any of that (yet) and/or there is filtering to do noise reduction - remember thet exposure times are very short....

The image looks rather soft, so the lens could be "soft". It seems that this camera takes C-mount lenses and the one fitted is probably a cctv camera lens, so nothing special, just cheap. Another factor is the aperture used to capture the video. Less light means wider aperture, which means shallower depth of field*, but also - usually - softer images. All lenses have their best sharpness at a specific aperture and there is an ideal range of apertures. Wide open apertures generally give softer images. Who knows, it could be a lack of skills combined with poor optics and wide aperture. The good thing is that there are definitely good lenses out there; the c-mount to eos adapter gives even more choices.

* So, a narrower in-focus range.

If you read the comments it seems these are not particularly optimal & he's promised to post more soon. 

[marshallh]

Surprised to see a piddly MachXO2, expected a ECP3
I have not seen such high density dram before.
Board layout is obviously done in Eagle (bleh)

Surprisingly simple unlike hte previous Edgertronic camera that was on kikckstarer (7 series xilinx, SoC CPU , DDR3 SODIMM etc)

[mikeselectricstuff]

Surprised to see a piddly MachXO2, expected a ECP3
I have not seen such high density dram before.

It's NAND flash, not DRAM.
The data rate isn't a huge issue in this application, it's more a case of sustaining that rate over a significant period. With this flash, it has multiple planes and dies, plus a large page size so as long as you have a moderate size fifo, the flash write time becomes acceptable.
I suspect they pre-erase pages before starting - it would be interesting to know if achievable rate drops in post-trigger mode.
 

[jeremy]

This is interesting timing, given that the new iPhone can do 240fps at 720p. Obviously not in the same league as this device, but I think this might be the start of something bigger. I suspect we will see a massive drop in high frame rate cameras in the coming years.

Great find, thanks for sharing.

[mikeselectricstuff]

Like thermal imagers, high-speed cameras are all about the sensor.
There is the additional issue of dealing with the datastream, but that's getting cheaper by the day.
Using flash is an interesting variation, and avoids dealing with fiddly DDR3 memory interfaces.

There is the potential issue of endurance - this chip is rated for 60K write/erase cycles. This corresponds to about 2 months of continuous recording, which is probably an acceptable compromise on a low-cost unit like this, though it may be a bit of a disadvantage for long post-trigger setups where you want to capture the period before an infrequent event by continuously recording.

 I would however hope they do wear levelling.

There have been very few generally available sensors - most specialist high-speed camera companies roll their own (possibly adapting existing sensor designs for fast readout)

This is a new range, AFAIK the first new high speed sensors to appear on the market for many years, and are cheap compared to others previosuly available - prices from  Findchips , at MOQ of around 40x

   PYTHON 300  (VGA)   $58.49
   PYTHON 500  (SVGA)  $72.49
       PYTHON 1300  (SXGA) $117.44
NOIP1SN2000A-QDI  (presumably the  2560x2048)    Eu230   



 

[Marco]

I'd rather see something like this as a daughterboard for Parallella or one of the Zedboards.

[mikeselectricstuff]

I know someone was experimenting with the Cypress FX3 to stream output from a high-speed sensor to USB3  to do a really cheap high-speed camera- I think they were fairly close to getting it running, but ISTR it was a lower resolution sensor than these.
 

[ferrix]

If you read the comments it seems these are not particularly optimal & he's promised to post more soon.

Yep I even mentioned that in my original post   It's not that I *dis*believe it, but I think it's important to actually see it proved before chucking money in.  On the other hand he's already funded so I guess it doesn't matter to those folks either way.

[jeremy]

I just realised, isn't the f437 export controlled because of the encryption accelerator? I wonder why they chose that over the 427.

[mikeselectricstuff]

I just realised, isn't the f437 export controlled because of the encryption accelerator? I wonder why they chose that over the 427.

Cold just be what they had /could get at the time of building prototypes

[jeremy]

Has it on the spec table though...

[mikeselectricstuff]

Has it on the spec table though...

They are UK based and ST are European, so not probably subject to dumb US regs.
I don't know if there are any dumb Eu regs that apply to this part.

[mikeselectricstuff]

With that amount of memory available, it would be cool to have a segmented memory mode so it could capture footage before/during or after multiple trigger events
 

[jeremy]

I tripped on regs trying to get similar parts (4x9 series) from the UK to AU. Took element14 a week or two to even work out what to do. One e14 lady over the phone flat out told me I couldn't buy them. I hope these chaps have thought about that.

This happened around the time that a lot of stm32 parts had their docs updated, noting that they were now suitable for defence applications. Maybe some agreements or regulations changed around that time.

[cloudscapes]

Just noticed the different tiers have differently sized sensors as well as the differences in frame rate on the graph. 1/4", 1/2" and 1". They're smaller than you might think, these are not diagonal measurements like a monitor might be measured. 1" lies between compact cameras and m43/apc-c. There are a few enthusiast compacts with 1" sensors. The sony RX100 is one if I recall.

Larger sensor means narrower depth of field with the same lens accros the board, so althoguh the more expensive tiers will give you better framerates, you might have more difficulty focusing. On the flipside, larger sensor means better low-light performance, so you might be able to stop the lens down for a wider DoF.

One day, I want to learn a bit on sensors to be able to stick one (a large-ish one, full frame idealy) inside an old SLR metal body!

[Rasz]

I know someone was experimenting with the Cypress FX3 to stream output from a high-speed sensor to USB3  to do a really cheap high-speed camera- I think they were fairly close to getting it running, but ISTR it was a lower resolution sensor than these.

tesla500?
that was a loong time ago

[Bassman59]

Some data on the sensors :
http://www.onsemi.com/pub_link/Collateral/PB_NOIP1SN1300A-D.PDF

I think that part is a follow-on/replacement for the older IBIS5-B-1300 device, which dates back to Fill Factory, which was bought by Cypress (and then the principals left to form CMOSIS), then Cypress sold their entire imaging division to ON.

It's easy enough to drive.

[mikeselectricstuff]

I know someone was experimenting with the Cypress FX3 to stream output from a high-speed sensor to USB3  to do a really cheap high-speed camera- I think they were fairly close to getting it running, but ISTR it was a lower resolution sensor than these.

tesla500?
that was a loong time ago

Yes, him, but quite a bit more recently - I think it stalled pending an issue with the Cypress  firmware. And probably not helped by him getting  a big CNC machine to play with

[PointyOintment]

Has it on the spec table though...

They are UK based and ST are European, so not probably subject to dumb US regs.
I don't know if there are any dumb Eu regs that apply to this part.

What about the Wassenaar Arrangement?