Problem using opencv on Maemo for ARM (Nokia N810)

[Andrew Daviel-3]

I built libdecodeqr in scratchbox for my Nokia N810 (Maemo Linux on ARM
processor), It uses many opencv routines such as cvFindContours.
When I run it, it can't decode any QR codes, but the same version with
the same libraries built on an x86 works OK (decodes many images
successfully)

The entire opencv package did not build cleanly - many of the Python
bindings did not get made. But libcv.so etc. were successfully built, and
I did not get any errors building libdecodeqr. It runs OK - does not
segfault - but prints a status saying it could not find the finder
pattern in the test images.

The only change I made after running "configure" was to substitute -O2
for -O3 to get one object to build.

I tried building and running some of the tests (cxcoretest,cvtest) and
they seem OK though quite slow on the tablet - I did not run them all.

There are obviously differences between my AMD desktop and my ARM tablet
(different CPU, speed, display depth, etc.) but I can't see anything
blindingly obvious that would cause opencv to perform differently. I
don't have missing libraries else it would not run.

Any suggestions ?

--
Andrew Daviel, TRIUMF, Canada

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

[Nils Hasler-2]

Hi Andrew,

>  When I run it, it can't decode any QR codes, but the same version with
>  the same libraries built on an x86 works OK (decodes many images
>  successfully)

I am not familiar with libqrdec. It seems to be a library for image loading.

>  The entire opencv package did not build cleanly - many of the Python
>  bindings did not get made.

You are probably just missing python libraries. That should not be the
problem since libdecodeqr does not seem to be using python.

>  But libcv.so etc. were successfully built, and
>  I did not get any errors building libdecodeqr. It runs OK - does not
>  segfault - but prints a status saying it could not find the finder
>  pattern in the test images.

Are you actually sure that it is OpenCV's fault and not a problem of
libdecodeqr?

>  There are obviously differences between my AMD desktop and my ARM tablet
>  (different CPU, speed, display depth, etc.) but I can't see anything
>  blindingly obvious that would cause opencv to perform differently. I
>  don't have missing libraries else it would not run.

ARM is not officially supported by OpenCV but I don't see a general
problem with OpenCV running on ARM either. It is still possible that
some function breaks.

>  Any suggestions ?

* Try to narrow the problem down.
* Run all the tests of OpenCV.
* Look at libdecodeqr in detail and try to find out whether OpenCV is
actually the problem.
* Which functions of OpenCV are actually used by the library?
* If you can find a limited number of functions that fail, we can try
to help you fix it. We'd certainly appreciate any resulting patches.
:-)

For more interactive help check out our IRC channel:
#opencv on freenode

cheers,
nils

--
Max-Planck-Institut Informatik
Campus E1 4, D-66123 Saarbrücken
Phone: [phone]
www.mpi-inf.mpg.de/~hasler

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace
_______________________________________________
Opencvlibrary-devel mailing list
Opencvlibrary-devel@...
https://lists.sourceforge.net/lists/listinfo/opencvlibrary-devel

[karnare ()]

Any suggestions ?

(1) modify "char " to " signed char " if there is comparation
(2) swap high word and low word for  cvRound()
(3) use the function cvAdaptiveThreshold() instead of apaptive_white_leveling() in function _create_posterized_image()

I have decoded the 125.bmp successfully in S3c2410(arm-linux-gcc 2.95.3), but it is not successfully for other pictures and the speed is slow.