Currently the cost function can cause ransac to find a local minimum with a low number lines (but low fit error). We should perhaps change back to just using the number of lines matched and also look at adding a cost term to encourage the fit to be more linear (e.g. an L1 norm of the fit coefficients).
Currently the cost function can cause ransac to find a local minimum with a low number lines (but low fit error). We should perhaps change back to just using the number of lines matched and also look at adding a cost term to encourage the fit to be more linear (e.g. an L1 norm of the fit coefficients).