[Elphel-support] Global Shutter

[Sebastian Pichelhofer]

On Mon, Nov 29, 2010 at 04:07, Abe Bachrach <abachrac at mit.edu> wrote:
> thanks for the quick response.
> In talking with one of the other people in lab, we realized that most of the
> experience that we had working with cameras with rolling shutters was with
> cheap webcams. We hoped that the readout on the elphel sensor *might* be
> fast enough that the rolling shutter effect would be insignificant.
> Unfortunately it doesn't sound like that is the case.
> - if the time gap is ~1/15 sec that means that if the camera is moving at
> 5m/s (~11mph) then camera would move 33cm during the readout period. 4X
> binning would make it a bit better, however the sensor would move ~8cm,
> which is still quite significant.

You mentioned you need only a small fraction of the 5 megapixels
resolution the sensor offers. If you reduce the WOI (Window of
Interest) you can increase the framerate and therefore the readout
time (more than with binning):

For example at 640x480 you can reach a max. of 126fps which would
reduce ERS artefacts to just a little more than 10% of the 33cm you
mentioned at 1/15s.

Regards Sebastian


> For more info on the rolling shutter distortion that I'm referring to, see:
> http://en.wikipedia.org/wiki/Rolling_shutter
> There are algorithms out there, that try to compensate for the rolling
> shutter distortions:
> http://mpac.ee.ntu.edu.tw/Exhibition/rolling-shutter.php
> http://www.cvl.isy.liu.se/research/rs-dataset/0382.pdf
> http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=05459408
> However it is much better to avoid them altogether by using a sensor with a
> global shutter. This is the approach taken by most people in the robotics
> and machine vision communities.
> As I said on the phone, I believe the Elphel platform would be VERY
> attractive for many people in the machine vision/robotics community, however
> a global shutter is a must for any of the applications that involve fast
> motion which many of them do.
> thanks!
> -=Abe
>
> On Sun, Nov 28, 2010 at 9:45 PM, Andrey Filippov <andrey at elphel.com> wrote:
>>
>> On Sun, Nov 28, 2010 at 7:30 PM, Abe Bachrach <abachrac at mit.edu> wrote:
>>>
>>> One other question for Andrey/someone else is:
>>> - How much time elapses between when the first and last row are read-out?
>>> from looking at the datasheet, it says that the maximum datarate is
>>> 96Mp/s, which would mean that for full resolution, the time gap would be at
>>> least 0.052488 seconds.
>>
>> Abe,
>>
>> It is somewhat longer than that because of the large "margins", the
>> average data rate of the sensor running at 96MHz is ~75MPix/sec. There are
>> formulae  in the datasheet that allow to calculate line readout time for
>> different ROI and decimation
>>>
>>> - is the sensor being run at the full 96MHz clock rate? Is that time gap
>>> number correct?
>>
>> 96MHz - yes, correct, but the "gap" is wrong - at full resolution readout
>> time (and so the delay between the first and last line exposure) is ~1/15
>> sec
>>>
>>> also,
>>> - How does the subsampling mode effect this. If we put the sensor in
>>> binning/skipping mode, and downsample by 4x, ideally, this would mean that
>>> there it takes 0.0032 seconds to read out a frame.
>>
>> Yes, that is correct. Just keep in mind that there is a large "dead" time
>> (horizontal blanking)  added to each scan line, but small on top and bottom
>> (vertical blanking)
>>
>> Andrey
>>
>
>