Slide 16 of 27
Once we have calibrated the cameras, we can capture individual events very easily. The basic idea is shown here.
For each camera, we can automatically detect the part of the image corresponding to a moving object. Imagine that we then project the silhouette of that moving object through the image and out into a representation of the 3D world in front of the camera. If we do this for several cameras, then the intersection of those projections will be a good approximation to the object. Of course the detailed structure will not be correct, and there are methods to deal with this, but the general structure can be captured in real time.