Well,
I know that Elphel exists, but even with their new fpga+dsp board,
they can not handle my pixel rate, which is 160 MP/s, at 14 or 16 bits
per pixel. With this, you can have ~ 6.6MP @ 24fps (FullHD is 2MP). The
elphel guys are also rather focused on CMOS sensors, which have some
very bad features - like rolling shutter, low sensitivity, high noise,
low dynamic range. Also getting a fairly big cmos is impossible, usually
they come in 1/3" or 1/4" optical format, while I want to use 2/3" to
4/3" sizes. Then there is RAW.. some guys call RAW also what it is not..
i want real raw - which is 320MB/s at top.. and there is no way this can
be handled by the Elphel core.
I have also experimented with reverse engineering a DSLR, taking out
RAW video data and skipping the h264 compression.. but that had
drawbacks that in video mode the sensor might be outputting less bits
(actually 8 bit RGB would be sufficient for h264 encoding) and there is
again a rolling shutter which I hate in every video I see, it is very
limiting the range of scenes/moves you can do.
The closest camera to my idea is the acam-dII.. 7000e (FullHD only,
2/3"), it has the smallest sensor with which I am only starting, the
largest then being 4x larger in area and MP.
The questionable middle-part of the camera depends on how much
processing you want to do. Either you can dump RAW from sensor head to
storage or connection (simplest scenario), or you want to have a
preview, in correct colors.. so you need to put there additional logic.
For helper functions like focus, AE, overexpose indication you again
need more logic.. and you can never tell depending on simulation how
much exactly (and it varies also by the level of parallelism, because of
slow processing or slow interfaces).
A modular system is the only (low-cost) way to go I suppose.
Daniel
On 08/18/2011 09:45 PM, Jose Hevia wrote:
The modules would be exactly the medium which solves my problem, a real
problem - making a digital cinema camera.
Now we are talking turkey.
I think this is a very good idea, I really wish I could play with my
digital camera better making automatic panoramas or shots or
programmed focus (like in cinema) easily without having to pay an arm
and a leg to do so because manufacturers make it "professional only"
features.
E.g I will love to be able to program the camera zoom to make
automatic photos with different focus and then extract the 3d of the
scene assigning a dept to every point in the picture based on the
picture that maximizes sharpness of the point. This is already done in
microscopy but is far from automatic.
Ephel cameras are expensive just for playing, I could buy a Sony cam
and destroy part of the housing and solder it and it will work for
less than 300euros/dollars.
Currently I know what is on one end (specific CCD chip) and what would be
somewhere in the middle (specific LCD) - but I do not know exactly what is
on the end and how many "middles" there will be. To accelerate this
development I want to use modular design - but not specific to imaging as I
do not want to throw out modules after design or which did not fit the job.
What specific things do you want your camera to do?
Comparing to Arduino it is nice, but it is made for people who can not
design a simple pcb nor connect the atmega to parallel port for programming
(I am widely using the ATmega mcu's, always on custom board.. first was very
experimental, the consequent ones were always perfect).
What I want is a set of modules which can replace designing a complex board
- when you take into account a reasonably big FPGA, it will be in BGA.. so a
breakout board is good.. but it will be a mess connecting all the devices by
wires.. so I grouped all the signals into few ports, and added out-of-band
communication (adc, phy's have usually i2c or spi for configuration) which
again does not need any wiring - true plug and play (develop).
I am not asking the question because I know - it mainly replaces an
application specific prototype PCB design and few rework cycles to get it
right (so it saves resources for the designer - both time and money).
What I do not know, if I had missed some specific aspect of a modular
system (currently I have signals and mechanic form factor selected and I am
rather looking if there is also something else interested in designing or
using it).
You should talk to these people:
http://cinema.elphel.com/
Maybe you can ask them what do they need, then when you finish it you
have something that works, because for a camera you need optics and
mechanical design and stuff that is not that easy to do.
The opengraphics team have lots of experience designing graphic
processing cards, and I think they are looking for a viable business
model, modifying the graphic cards for video processing could be it.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
