bigbrotherBENL

Face recognition on the Big Brother episodes in Belgium and the Netherlands. Keeping statistics of whom are most visible and recognisable in the series and wether or not it has an impact on who wins.

How does it work?

A video is nothing more than a sequence of images.
Each image in these videos (also named frames) are analysed one by one to see if there are any visible faces.
If there are, it will try to match the faces with any of the known faces of the participants and hosts.
If it wasn’t able to do that, it will get the label unknown.

At the end of analysing each video, there will be a counter for each individual of how many times they were visible in the episode.

The very first ‘episode’ was a small intro video by the hosts Peter and Geraldine.
The video was about 7000 frames long and by analysing it frame by frame, we got the following data.

Host Geraldine: 2108
Host Peter: 1516 
Unknown: 31 
Grace_35: 12
Kitty_58: 9
Tobias_26: 8
Vera_26: 1

Reliability

As you can see above for the intro video, things aren’t always 100% accurate.
It recognised people that weren’t even in the video.

However, this amount is marginal and neglectable.

A bigger issue would be someone facing sideways, so still getting airtime but not being recognised in these statistics.
As well as facing away or even out of frame but still hearible by what the person says.

Potential optimisations

Connect multiple frames to smoothen out the potential errors.
For example: host Geraldine spins around 360°.
So the frames would have counted something like this: Geraldine - Geraldine - Kitty - unknown - Kitty - Geraldine - Geraldine all in less than a second.
Assuming the tv-producers don’t want to give us an headeach, all the frames should’ve probably recognised the person as Geraldine.

Also, doing the same for voice recognition and then overlapping both data would probably give an even better idea of airtime per individual.