Our team was hired by a company which provides professional broadcasting systems to implement and integrate automatic and intelligent face tracking into an existing robotic camera control system.
In addition to a robust and high accuracy face detection and tracking we have implemented many custom algorithms for handling different scenarios for natural and smooth tracking, making it close to manual camera movements.
We achieved a real time performance with 4 cameras using GPU optimized mechanisms for detection and tracking.
The software allows to track not only a single face but also two faces - keeping them in a constant position on frame by automatically applying required zoom, pan and tilt.
It requires minimal setup and user input. In some cases, it can even initiate tracking without any user intervention.
It provides fully configurable camera movements including maximum speed, acceleration/deceleration, response delay.