Change default min distance to sensor from 3m to 1.5m (1m causes trouble with some datasets) because a lot of datasets (particularly indoor ones) needs this parameter setting.
The PCA relative params are modified :
The first idea was to replace the planarity/linearity thresholds that are not really intuitive by the state of the art reference ones, respectively (l1 - l0)/l2 and (l2 - l1)/l2.
However, this lead to not so intuitive values because of the low number of points that is used to fit the models (edge/planes).
Moreover, in the plane case, the smallest eigen value is already checked when thresholding the MSE value and the same is done in the edge case with the two smallest eigen values.
Therefore, the only check that is kept is to ensure the plane points do not lay on a line which would lead to a false normal computation. To do so, we compare the two highest eigen values, we chose l1/l2 to keep the value between 0 and 1. A 1/5 ratio is chosen as default limit. WARNING : the eigen values ratio is equivalent to the neighborhood main directions expansion ratio SQUARED. Therefore, if one wants to filter only planar patches that have a ratio of 1/5 between its 2 main directions, he has to put a value of 1/25 between the patch main eigen values.
Example of improvements on the trajectory in an indoor context (red = master, white = this MR result)