approach to face detection in video (variable sizes and 3/4 turns)?

asked 2015-10-17 20:19:36 -0500

peterl gravatar image

Hi All,

I'm wondering if there are any available cascades trained for video? I'm looking for the way to blur multiple faces on a video and struggling with making it work with cascades that come with OpenCV 3.0. The video has variable size faces (closer and farther from the camera) and my first naive attempt was to run the classifier with smallest minimum size and no maximum size to detect all of the faces. That failed meaning there were too many false positives. The second attempt was to do a multi pass with min and max sizes set to detect a certain group of faces on each pass. This is not ideal since it's hard to predict the bounds and obviously takes much more time. Also there is a problem that I couldn't solve with faces that are 3/4 turned, those are not detected by either frontal or profile cascades. I was wondering if there are any specially trained cascades for more accurate face detection within frames that contain faces of variable sizes and with 3/4 turns. If I'm completely off the point here and there is a good way to solve this with cascades available in OpenCV 3.0 please point me in to the right direction.

Thank you

edit retag flag offensive close merge delete