Implement treshholding for new individuals

[rohansaw]

We require a mechanism to detect whether a prediction belongs to an individual in our database or whether a new individual is seen. We can distinguish a new individual by implementing a threshold that is used to determine whether the prediction is too far away from all known other individuals. Then this individual should be added as a new class (or even better do this in videos so that we have multiple images in the db for the new individual, so that k-nn works better)