Training an existing classifier
I'm trying to detect the eyes and nose of a user, and it works reasonably well. But there are often frames in which the features are not detected. The application only has to track one user, that user never changes for the computer the software is running on.
Is it possible to take images of the eyes that the generic eye classifier finds and train the classifier with the user's specific eyes, as that should make it possible to detect the eyes in more frames. (The same goes for the nose as well).
alternatively to cascade classifiers, try landmarks, like faceX , or dlib's, or flandmark
Or do a google search on the newly released packtpub book "Opencv 3 Blueprints" that has been just released! It has a chapter that discusses the use of opencv and flandmark combined!