- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I'm being asked to port an application using opencv(ms vc++ dot net) on windows to run on a network box (arm linux on ixp425).
Then sometime in the future part of the application
will be moved to blackfin or other dsp
both ixp425 and blackfin on pc104
Trying to work out the best way to proceed
(the only non pc developer around here).
Have to use gcc on arm linux.
Basically capturing images from ip cameras then processing them, can't really say to much more other using foreground/background estimation and other standard type image processing ops.
Thinking either
1. Port to linux(x86) using opencv. Then convert to ipp
then to arm linux on ixp425.
Problems
windows code isn't optimised.
Has very high memory usage.
Windows code is a moving target and I'm not developing it.
ixp425 board only has 64MB ram, same with gumstix boards, current windows code uses up to 512MB.
I'm fairly sure that the memory for the windows app could easily be brought down to under 256MB while still using opencv.
Some design changes would cut memory use by half , if I'm allowed to make them.
or
2. Convert windows version to use ipp directly.
Then to linux, then to arm linux.
Interfacing with the cameras isn't a problem as using ip cameras and its easy to capture from them.
Problem is the timeframe I've been given.
Lack of familiarity with arm linux , I usually
work on lower end embedded systems.
There was an option to use windows(x86) on the network box but we need very low power , small size and high reliability.
Using gumstix boards and ixp425 pc104 bboards for developemnt.
All suggestions and advice greatly appreciated.
Then sometime in the future part of the application
will be moved to blackfin or other dsp
both ixp425 and blackfin on pc104
Trying to work out the best way to proceed
(the only non pc developer around here).
Have to use gcc on arm linux.
Basically capturing images from ip cameras then processing them, can't really say to much more other using foreground/background estimation and other standard type image processing ops.
Thinking either
1. Port to linux(x86) using opencv. Then convert to ipp
then to arm linux on ixp425.
Problems
windows code isn't optimised.
Has very high memory usage.
Windows code is a moving target and I'm not developing it.
ixp425 board only has 64MB ram, same with gumstix boards, current windows code uses up to 512MB.
I'm fairly sure that the memory for the windows app could easily be brought down to under 256MB while still using opencv.
Some design changes would cut memory use by half , if I'm allowed to make them.
or
2. Convert windows version to use ipp directly.
Then to linux, then to arm linux.
Interfacing with the cameras isn't a problem as using ip cameras and its easy to capture from them.
Problem is the timeframe I've been given.
Lack of familiarity with arm linux , I usually
work on lower end embedded systems.
There was an option to use windows(x86) on the network box but we need very low power , small size and high reliability.
Using gumstix boards and ixp425 pc104 bboards for developemnt.
All suggestions and advice greatly appreciated.
Message Edited by alxx@alxx.net on 05-05-200606:40 PM
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi!
Now there is no official release of OpenCV for IXP or XScale platforms but some discussions of it can be found in Yahoo OpenCV forum.
OpenCV is not intended to provide the optimized code itself but has a method to call functions from optimized libraries. Now IPP and MKL are supported but a new optimized library can be easily added. IPP image processing functions are optimized for "heavy" platforms and Pentium and Itanium but not completely fo IXP that has now lower priority. Generally this optimization is directed to the calculation speed and then to the memory usage.You can promote IXP optimization by requests on premier.intel.com.
OpenCV is the open source library and if you can contribute smth valuable it will be added to the public release.
Thanks,
Alexander

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page