Not a dev, but there is a project to improve the whole face recognition process in digikam. However, I have not heard anything about recognizing people in videos, so I wouldn't be optimistic about it, at least in short-medium term.

Probably some of the developers will be able to give you a more specific answer.

A quick response from me :

Video face detection will be time consuming. You need to read frame by frame (in critical case) to check if a face exist in image. So the video need to be decompressed and analyzed.

Second point, we need a temporal property to add in database to localize a face in a specific frame from a video. This kind of value do not exist yet in database.

Anyway, detection and recognition algorithm do not needs to be adjusted for video as media is analyzed by frames (images).

Gilles Caulier

Real-time face detection and recognition on video are possible. However, if you wish to extract all the faces appear in the video, it is possible that too many faces are returned, depend on the number of frames. If you wish to return only distinct faces in the video, face clustering is needed. 

In both ways, video analyse like this, as Gilles said, is time-consuming. But if this feature is still in need, I could try to implement it

About time. I don't see problem with it. I can run it at night.

But I see another problem with shouldn't be done before: to see similar faces as one with some coefficient.

Maik, Michael,

I can imagine the amount of ressources from the computer to analyze a video contents frames by frames. At least an algorithm must determinate if one frame change really compared to previous ones and select only the best candidates to detect faces.

Gilles