Slide 11 of 27
Now given that we have input from a large number of cameras, we need to coordinate their views. This means we need a way of inserting cameras into the View Cube and having the system automatically calibrate the position of that camera with respect to all the other ones.
To do this, we have been using a very robust and very efficient real-time tracking system. This system runs on each camera unit, and uses statistical information about what it has seen recently to keep track of the background. Anything that does not fit the background is detected as a moving object. This statistical description adapts over time, so the system is not fooled by waving tree branches, flags, changes in cloud cover or other illumination effects.