You'll need a copy of OpenCV 3.0.x to link against, and the Haar Cascade files from OpenCV 2.4.10 to do recognition, since the newer versions changed the format but the C API seems to not have been updated to be able to read them.
You can grab both versions from:
Build the 3.0 version (via cmake . && make
). Install it if you wish, otherwise follow below to just tell opencv-cffi
where you've downloaded it.
On OS X, you can also get OpenCV 3.0 by running:
$ brew edit homebrew/science/opencv
remove the , branch => '2.4'
, and then
$ brew install --HEAD --without-python homebrew/science/opencv
(We do not need the Python bindings, you're lookin' at Python bindings).
E.g.:
$ opencv=~/Development/opencv-3.0.0 \
LD_LIBRARY_PATH=$opencv/lib/ \
DYLD_LIBRARY_PATH=$LD_LIBRARY_PATH \
C_INCLUDE_PATH=$opencv/include/:$opencv/modules/core/include/:$opencv/modules/hal/include:$opencv/modules/imgproc/include:$opencv/modules/photo/include:$opencv/modules/video/include:$opencv/modules/objdetect/include:$opencv/modules/videoio/include/:$opencv/modules/highgui/include/:$opencv/modules/imgcodecs/include \
LDFLAGS=-L$opencv/lib\
pypy opencv_cffi/build.py