cudacam: experiments in cuda, opencv, and kinect under Linux
This is a playground. Things will probably work if you install enough dependencies.
To use all the examples here, install these dependencies. Most can be
- numpy - http://numpy.scipy.org/
- freenect - https://github.com/OpenKinect/libfreenect
- cv2 - http://opencv.willowgarage.com/
- mdp - http://mdp-toolkit.sourceforge.net/
- scikits.audiolab (requires libsndfile)
- CUDA Toolkit 4.0 - http://developer.nvidia.com/cuda-toolkit-40
- pypy when numpy support is more complete
Performance will vary wildly depending on your setup. Compile opencv yourself to get maximum performance (do not trust pre-packaged versions):
mkdir build cmake -DWITH_OPENNI=yes -DBUILD_EXAMPLES=yes -DWITH_TBB=yes -DWITH_IPP=yes -DIPP_H_PATH=/opt/intel/ipp/include/ ..
- OPENNI can come from https://github.com/OpenNI/OpenNI
- IPP can come from http://software.intel.com/en-us/articles/non-commercial-software-download/ (tip: the registration forms accepts fake credentials).
- TBB can come from your distro or from http://threadingbuildingblocks.org/file.php?fid=77
After I enabled TBB (from distro) and IPP (downloaded and installed manually), the haar face detector doubled in speed (=> video frame rate doubled).