The Sample Filter performs Poisson sampling of the input
practice of performing Poisson sampling via “Dart Throwing” was introduced
in the mid-1980’s by [Cook1986] and [Dippe1985], and has been applied to
point clouds in other software [Mesh2009].
Our implementation of Poisson sampling is made streamable by voxelizing the
space and only adding points to the output
PointView if they do not violate
the minimum distance criterion (as specified by
radius). The voxelization
allows several optimizations, first by checking for existing points within the
same voxel as the point under consideration, which are mostly likely to
violate the minimum distance criterion. Furthermore, we can easily visit
neighboring voxels (limiting the search to those that are populated) without
the need to create a KD-tree from the entire input
PointView first and
performing costly spatial searches.
filters.decimation, filters.fps, filters.relaxationdartthrowing, filters.voxelcenternearestneighbor, filters.voxelcentroidnearestneighbor, and filters.voxeldownsize also perform decimation.
Starting with PDAL v2.3, the
filters.sample now supports streaming
mode. As a result, there is no longer an option to
shuffle points (or
to provide a
seed for the shuffle).
Starting with PDAL v2.3, a
cell option has been added that works with
radius. The user must provide one or the other, but not
both. The provided option will be used to automatically compute the other.
The relationship between
radius is such that the
radius defines the radius of a sphere that circumscribes a voxel with
edge length defined by
Care must be taken with selection of the
Although the filter can now operate in streaming mode, if the extents of
the point cloud are large (or conversely, if the cell size is small) the
voxel occupancy map which grows as a function of these variables can still
require a large memory footprint.
To operate in streaming mode, the filter will typically retain the first point to occupy a voxel (subject to the minimum distance criterion set forth earlier). This means that point ordering matters, and in fact, it is quite possible that points in the incoming stream can be ordered in such a way as to introduce undesirable artifacts (e.g., related to previous tiling of the data). In our experience, processing data that is still in scan order (ordered by GpsTime, if available) does produce reliable results, although to require this sort either internally or by inserting filters.sort prior to sampling would break our ability to stream the data.
This stage supports streaming operations
Voxel cell size. If
cellis automatically computed such that the cell is circumscribed by the sphere defined by
Minimum distance between samples. If
radiusis automatically computed to defined a sphere that circumscribes the voxel cell. Whether specified or derived,
radiusdefines the minimum allowable distance between points.
An expression that limits points passed to a filter. Points that don’t pass the expression skip the stage but are available to subsequent stages in a pipeline. [Default: no filtering]
A strategy for merging points skipped by a ‘where’ option when running in standard mode. If
true, the skipped points are added to the first point view returned by the skipped filter. If
false, skipped points are placed in their own point view. If
auto, skipped points are merged into the returned point view provided that only one point view is returned and it has the same point count as it did when the filter was run. [Default: