[libdc] Addressing problems with Videre and other non fully IIDC-compliant cameras.

[Jeremy L. <le...@wi...>]

The company I work for, Willow Garage, is developing an open-source robotics
software platform and as such we want to provide good support for firewire
cameras.  We've started making use of libdc1394 and have largely been very
satisfied.

However, there have been a couple problems that have popped up that I have
been unable to resolve and I was wondering if someone could point me in the
right direction.

The first is an apparent bug in the deinterlacing library.  I submitted a
patch to the libdc1394 patch tracker on sourceforge some months ago fixing
the bug, but, it does not appear that the patch tracker is being actively
monitored or used.  What is the correct way to get a patch incorporated into
the libdc1394 library?

The second problem is more involved and relates to our choice of the Videre
Design STOC cameras for doing stereo vision.  The bottom line is that the
cameras do not actually adhere to the IIDC specs.  The camera does not
respond to reading from a register while streaming video, which indirectly
means any of the dc1394_feature_set_value calls hang since they always read
from the register first.  At the moment, I'm keeping a local patch against
the libdc1394 respository which adds a set of functions:
dc1394_feature_set_*_blind, that basically mirror the normal calls, but
without performing a read from the register first.  This bothers me greatly,
but is still preferable to having to stop the video every time I want to
adjust whitebalance or exposure.

Furthermore, the camera does not seem to set all of the bits correctly at
the 0x40x offset causing many of the check feature calls to fail.  And, the
camera supports a number of non-standard features which involve setting
special values at the control register offset 0xFF000.  At the moment I'm
dealing with this through my own library built on top of libdc1394 which
deals with these things just through making low-level calls.  This works for
me, but certainly wouldn't work for someone without access of the workings
of the Videre or the want to deal with the low-level calls.  It also clearly
doesn't scale with an arbitrarily large number of different cameras.

It raises some issues with regard to the right way to handle these kinds of
situations, and how frequently these kinds of situations actually arise.  Is
Videre alone in their disregard for following the specs to the letter, or
are there a number of pathological cameras out there that are just not
compatible with the library?  My initial thought is that without too much
work the relevant parts of the library could be refactored for the purposes
of allowing someone to specify known IIDC spec exceptions on a per-camera
basis.  Libdc1394, after determining the make/model/version of the camera,
should be able to apply these exceptions "safely," and provide the end-user
with a more positive experience for these borderline cameras.  When someone
finds a camera that doesn't work, it changes the fix from a software
development problem to a characterization problem.  Formulating this as a
characterization problem means support for these cameras can be added
without risk of introducing bugs for other cameras that are already working,
and moreover can likely be done by someone with little to no software
experience.

Is this something that people would be interested in, or does it clash with
the overall philosophy of the dc1394 project?

Jeremy Leibs
Systems Engineer
Willow Garage, Inc.