The celebrated minimum inertia line problem is reconsidered: a line is to b
e fitted to a planar cloud of points so that the sum of squared distances o
f all points to the line becomes minimal. The classical algebraic solution
based on the tensor of inertia is complemented by a closed form trigonometr
ic solution allowing various generalizations including the fit of elastic p
olygons. Proper polygons will be fitted numerically with non-closed partial
solutions being reduced to the lowest dimension possible. This is compleme
nted by segmentation heuristics for the measurement cloud.
The approach allows to solve the robot localization problem with high accur
acy for position and orientation to be inferred from distance measurements
in a known polygonal environment. The essential feature of the current appr
oach is to fit the polygonal geometry as a whole.