Best OpenCV solution for face detection?

asked 2016-07-13 01:51:46 -0600

I am trying to find and blur faces in many long video files that all have the following style: image description

Standard cascades yield a detection of around 1% with almost 100% of all detected faces being false positives. It has been suggested to me that perhaps I should use SVM with HOG though I do not understand how that works or how it is implemented. I am capable of using Cascade-XML files in regular OpenCV fashion to go through each frame, and that is where my expertise runs out I'm afraid.

Is there any concrete OpenCV method that is optimal for this particular task?

edit retag flag offensive close merge delete

add a comment

1 answer

Sort by » oldest newest most voted

answered 2016-07-13 08:17:02 -0600

MRDaniel
489 ●3 ●9 ●25

Did you try the LBP cascade?

The faces don't look ideal given the resolution.

Does a full person classifier score well? You could infer the face position from the body which might be more easily detectable.

edit flag offensive delete link

Comments

Which LBP cascade? Would you like to link me to one? Is an LBP cascade usable same as a Haar Cascade or is different code necessary?

I tried a full person haar cascade and it didn't work well at all.

I was referred to trainHOG which seems to be a training software for HOG style cascades, which might be perfect for this purpose, but I have no idea how to actually use it once it's generated. It's not an XML file. In fact, it's not even a cascade.