• Implement tools to analyse the data received from the network cameras.
• find timespans in which it is mostly common the detect a certain target

E.g. train a Neural Network to predict how many targets should be entering the building during the next hour and warn the employees if it significantly exceeds this value.

• use the confidence values to measure the security level of a certain camera

Maybe placing the camera in another corner would increase the confidence by detecting human targets?
Is there a camera which did detect a lot of targets, but due to internal restructures in the organisation, the camera now is falling below the predicted value quite often? Well, then maybe also restructure your surveillance network.