1 | initial version |
Oh dear ... cvHaarDetectObjects is the most ancient interface for face detection around here. Instead move to the more advanced C++ interface called traincascade which does the same but in a more optimized way. Some references:
Using the LPB features gives a speed increase but at the cost of some accuracy. If you do not need to retrain your own model, I would still use the HAAR wavelets.
Eye direction, like in the case of viewing is a much harder task to do, and will not be easy with current openCV functionality.