Add face matching between cameras

[integraledelebesgue]

Rationale

We need to match all the detections accross the cameras to be able to precisely describe observations.
Currently, we use simple heuristics that don't generalise over multi-people environments.
They can fail even in the default mother-child setup.
We need some sort of facial recognition to match correspondning people views between particular cameras.

Task

Find and introduce some efficient model able to recognise people seen from different perspectives