I just ported a small OpenCV example to Maemo. It builds on http://qt-apps.org/content/show.php/...?content=89995 and is available here: http://www.zshare.net/download/771049800b033e35/
I have three short questions about it:
1) Klen and others recommended using OpenCV for an AR application for obvious reasons, but in this example the performance is quite poor. How could it be improved? Is it e.g. better to read the image via GStreamer and then run OpenCV methods over it? This is very important to me, since I'd love to use this library extensively.
2) QtCreator behaves strangely when I try to build this project: It throws 13 errors if I open the augmentedvision.pro (doesn't find opengl/cv.h & opengl/highgui.h) and use "Build All", whereas it compiles perfectly if done manually in a scratchbox shell (qmake augmentedvision.pro; make - build can be found in ./src). Any ideas how to avoid this?
3) If I now want to build a fancy AR app - would you use Qt + GStreamer + OpenCV + OpenGL + SDL or would you stick to a subset? My two starting points are this code, some corner detectors and Klen's OpenGLGst example. What are the best parts of both approaches, how can they be combined optimally and what should I add / change? Any last recommendations before I start programming a serious application?
Thanks in advance.
Sorry for the last two threads - they were intended to be simple posts instead, but obviously I shouldn't have changed the subject. They can be deleted as far am I'm concerned. Thanks for the helpful answers, though.