In this section we apply our autocalibration algorithms to synthetic data in order to analyse the effect of different kind of motions on the computation of autocalibration. 3-D points were generated and projected onto the cameras of a virtual stereo rig performing different kind of motions. Gaussian noise of 1-pixel standard deviation was added to the data.
For simplicity, we show results only for the calibration of the left camera of the rig. The actual intrinsic parameters are :
The aim of this experiment is not to obtain accurate computation of intrinsic parameters, but to show that if the constraint is used, there is only one critical kind of motions for the affine-to-Euclidean calibration (instead of three) : motions whose rotation axes are orthogonal to the image plane.