The linear algorithms described above can be outlined as follows:
Preprocessing of the data:
compute the inter-point distances
and
the cosines of the angular separations
using the image points and calibration matrix of the camera.
Solve for the depths of the reference points:
from the 4th degree polynomial equations of form ,
form the matrix ,
then use SVD to solve linearly
either in two steps for 4 points
or in one step for 5 or more points.
The square of the depth x is obtained from Equation (6),
then the depth is set to .
Absolute orientation:
the recovered depths of the reference points
are converted into a camera rotation and translation
using the absolute orientation method briefly described
in 2.
The experimental results of these linear algorithms
have been reported in [5].