Re: Favorite CV Libraries
- From: "RoboRealm" <contact@xxxxxxxxxxxxx>
- Date: 30 Oct 2005 10:10:25 -0800
There appears to be several factors that contribute to the popularity
of any particular package. From what is currently available on the web,
access to source code, speed and specific functionalities seem to be
the major components of popularity at this point.
Several of the libraries are written in C/C++ for performance reasons
and perhaps also for a limited degree of cross-platform compiling. Its
unclear to us which OS would be the most favored for vision based
projects. Many of the tools compile under Linux or other Unix flavored
platforms but due to this can cause issues for novice users to get a
working compiled binary. (See the many questions raised in the OpenCV
discussion forum on compiling, linking, etc issues). Nevertheless, it
is clear that for high performance vision based apps C/C++ is used to
some degree with higher level scripting or GUI components used for
'clue' or abstract processing.
Each package seems to cover some basic aspect of vision processing.
Many include the basic edge, convolution filters, etc but are
traditionally popular due to a specific feature like face detection,
object recognition or stereo alignment. This is to be expected as the
range of desired features for image/video processing is very large and
very specific to the task at hand. If someone were to start compiling a
list of how vision is used in the industry you'd be including items
from farming to industrial quality assurance. How can one package cover
all that? The solution is to offer something that can be widely
extended. Thus the desire for access to source code or some extension
API becomes paramount to the success of the package even if the package
only offers the most advanced image processing capabilities.
The bulk of the actual code in the packages surround the ability to
acquire images or get access to images. Processing of those images is
naturally included but due to the non-standard nature of cross platform
image acquisition a great deal of code needs to be written just to load
or import images. Thanks to packages like FreeImage, ImageMagic, etc.
images of various formats can be loaded into processing toolkits. When
it comes to direct importing of images from cameras the solutions seem
to be quite numerous with most of the standardization falling towards
Windows with VFW or DirectShow (not that video acquisition for Windows
is easy!!). The Unix variants appear to require more specific drivers
to access cameras and it is not clear what standard is developing.
We suspect that most programmers are eager to start developing their
own vision based libraries in much the way that the previous respondent
to this question did. Its also likely to conclude that the people that
are getting the most out of these libraries are the authors themselves
in essence looking for interesting projects to use their libraries on.
The most interesting and best feature of each library is always the
next one to be created ...
We have begun the same process at www.roborealm.com We do maintain a
list of links to other such systems in order to keep track of this
growing area of interest and to understand the more global trends of
vision software. You're welcome to use this list as your own starting
point to find the right package for you.
http://www.roborealm.com/links.php
If you are the author of a package that is not included let us know!
We hope this helps!
STeven.
.
- References:
- Favorite CV Libraries
- From: RyanSturmer
- Favorite CV Libraries
- Prev by Date: Re: PGM file
- Next by Date: Re: PGM file
- Previous by thread: Re: Favorite CV Libraries
- Next by thread: Placido Ring Detection
- Index(es):
Relevant Pages
|
