The selection of points (step 2) can heavily affect the final alignment accuracy. Therefore, globalICP offers two advanced selection strategies, which are especially useful for point clouds where one normal direction is predominating, but the data still includes some valuable features for the alignment. First, points are always selected with the uniform sampling strategy described above, where the mean sampling distance is defined by the parameter UniformSamplingDistance
. To select a subset of this selection, one of the two following strategies can be applied:
NormalSubsampling
to true. The percentage of selected points can then be defined with the parameter SubsamplingPercentPoi
MaxLeverageSubsampling
to true. The percentage of selected points can then be defined with the parameter SubsamplingPercentPoi
A comparison of the selection strategies offered by globalICP is given in Figure 1. For most ICP variants, this Airborne Laserscanning scene is rather difficult because only one feature - the ditch - can constrain the transformation at the finest level. Since normal space sampling and maximum leverage sampling consider the usefulness of points for the alignment process, the number of points can be dramatically reduced, without affecting the solubility of the adjustment. Further information can be found in Glira et al. (2015).
Glira, P., Pfeifer, N., Ressl, C., Briese, C. (2015): A correspondence framework for ALS strip adjustments based on variants of the ICP algorithm. In: Journal for Photogrammetry, Remote Sensing and Geoinformation Science (PFG) 2015(04), pp. 275-289. (DOI: 10.1127/pfg/2015/0270)