Comments on: Compiling OpenCV for ARM9 platform

By: Yuri Vashchenko

Yuri Vashchenko — Sun, 08 Aug 2010 23:12:24 +0000

Hi JPN,

We did not use any tools to change doubles to fixed point operation.
We rewrote the big part of opencv manually.

Unfortunately, it had not given significant improvement in performance on our target ARM platform.

So, if you need fixed point operations for your DSP of whatever your target platform is, you will probably have to rewrite a part of opencv code yourself.

Start with running a profiler on your target platform to understand which code takes most of the execution time.
Then, using opencv source code, try to understand how it works and start replacing double precision variables and operations with fixed point ones. In our experiments, we even had been able to rewrite some code to use 32 bit integers only. It affected precision in some cases, but until overall quality is acceptable, you can do such trade-offs.

Hope this helps,

Best regards,
Yuri

By: JPN

JPN — Wed, 04 Aug 2010 03:40:43 +0000

I am working on optimizing and compiling components of OpenCV to ARM Cortex A8. It is very portable in that it will run on many platforms, but it is only set up to run *well* on Intel x86. ARMs and most other DSPs do not like double precision numbers, for one thing, but with OpenCV we are stuck with double. The other problem is that OpenCV is difficult to analyze with call graph tools, so it is difficult to extract just the components you need. I have had no luck with such tools.

By: Yuri Vashchenko

Yuri Vashchenko — Sun, 20 Jun 2010 21:48:46 +0000

Hello, Santosh,
We did not compile OpenCV library for Palm. As for Mobile devices platform, we ported OpenCV on Android and Windows Mobile.

To compile OpenCV for Palm, you need:
1. Cross-compiler that can genrate binaries for your target Palm OS.
2. Standard libraries (like libc) compiled for your target platform.
3. Libraries openCV needs to work ( depending on your application may be GTK, FFMpeg, Libjpeg, etc).

Hope this helps,

Best Regards,
Yuri

By: Santosh

Santosh — Thu, 17 Jun 2010 09:55:17 +0000

is it possible to compile the OpenCV code for Palm mobile device which uses ARM?

By: Bryce

Bryce — Sun, 27 Dec 2009 01:22:37 +0000

I’m making a robot that will recognize objects like a ball, cup, ect. I have the basic application down, but I need to actually need to test it on the interface to see if it works right. The basic application is it gets image from the camera, loads it into its RAM, analyzing it to find those objects, sends signals to the Arduino for where the servo’s to move.

I have the basic “seeing” application source now.

By: Yuri Vashchenko

Yuri Vashchenko — Fri, 25 Dec 2009 08:40:06 +0000

Hi Bryce,

Unfortunately, I cannot give you any source code files (including makefiles) as our management restricts it. If you really need any source code please contact our sales/marketing team (consult contacts page for more info).

Anyway, I can probably help you with an advice if you give more details on your project and problems you are facing. How many source code files do you have? Do all your source files compile or you are getting some compiler errors?

By: Bryce

Bryce — Fri, 25 Dec 2009 00:38:06 +0000

Any chance you will just post up the files needed already followed through instructions. I can’t get past, “Create/Modify Makefiles”. I really need this for my project.

By: Yuri Vashchenko

Yuri Vashchenko — Mon, 23 Nov 2009 23:04:36 +0000

Hentati,

Personally, I have not tried to compile and use the OpenCV library for MicroBlaze or Nios II. Nevertheless, OpenCV library is written to be cross-platform and highly portable. So, if you have a cross compiler for your platform with corresponding platform-specific libraries chances are that OpenCV and your algorithm will compile and work. You may face some problems with Highgui part of the OpenCV library (graphical output, jpeg decoding, etc.), but this is another question.
If your are running Linux on your target platform, I recommend you to use a gcc compiler (such as one included into the CodeSorcery G++ Toolchain) it is available.

As for a specific FPGA – I am not sure I understand your question. Do you need to choose a target board for your application or you have a specific FPGA and you want to know if it will help you running OpenCV algorithms? Choosing a FPGA mostly depends on what algorithm are you going to run on it. Depending on your application, you may want to run a FPGA for a fast fourier transform or convolution on FPGA to reduce the load of your CPU and improve the overall performance.

By: hentati

hentati — Fri, 20 Nov 2009 10:35:29 +0000

hi!
please can some one answer me this question
can I use opencv library wilth a microblaze or niosII.?
the algorithm run in windowsXP
if yes which fpga should I have?
best regards