This is an implementation by Daniel Fontijne of a motion capture camera calibration algorithm by Joan Lasenby and Adam Stevenson. The algorithm first makes an initial guess about the relative position and orientation of the cameras, and then iteratively refines the solution until convergence is archieved. This computation can be done in a couple of seconds for about 10 cameras looking at 100 markers. The algorithm uses a euclidean 3d geometric algebra.
Here is an animation of the algorithm in (simulated) action. 10 cameras, randomly positioned, looking at 100 randomly positioned markers, are calibrated. Noise is superimposed over the marker positions at each camera to make the simulation more realistic.
Todo: little m2v or avi file...