filters.pclblock¶

The PCL Block filter allows users to specify a block of Point Cloud Library (PCL) operations on a PDAL PointView, applying the necessary conversions between PDAL and PCL point cloud representations.

This filter is under active development. The current implementation serves as a proof of concept for linking PCL into PDAL and converting data. The PCL Block filter creates a PCL Pipeline object and passes it a single argument, the JSON file containing the PCL block definition. After filtering, the resulting indices can be retrieved and used to create a new PDAL PointView containing only those points that passed the filtering stages.

At this stage in its development, the PCL Pipeline does not allow complex operations that may change the point type (e.g., PointXYZ to PointNormal) or alter points. We will continue to look into use cases that are of value and feasible, but for now are limited primarily to PCL functions that filter or segment the point cloud, returning a list of indices of the filtered points (e.g., ground or object, noise or signal). The main reason for this design decision is that we want to avoid converting all PointView dimensions to the PCL PointCloud. In the case of an LAS reader, we may very well not want to operate on fields such as return number, but we do not want to lose this information post PCL filtering. The easy solution is to simply retain the index between the PointView and PointCloud objects and update as necessary.

See Filtering data with PCL for more on using the PCL Block including examples.

See 1   Draft PCL JSON Specification for complete details on the PCL Block JSON syntax and the filters available.

Dynamic Plugin

This stage requires a dynamic plugin to operate

Options¶

filename
Path to external PCL JSON file describing the pipeline
methods
Raw PCL JSON array describing the pipeline

PCL Block Schema¶

The PCL Block json object describes the filter chain to be constructed within PCL. Here is an example:

[
{
"name": "FilterOne",
"setFooParameter": "value"
},
{
"name": "FilterTwo",
"setBarParameter": false,
"setBounds":
{
"upper": 42,
"lower": 17
}
}
]


Implemented Filters¶

The list of PCL filters that are accessible through the PCL Block depends on PCL itself. PDAL is rather dumb in this respect, merely converting the PDAL PointView to a PCL PointCloud object and passing the JSON filename. The parsing of the JSON file and implementation of the PCL filters is entirely embedded within the PCL Pipeline.

A summary of the currently available filters is listed below. For full details of the filters and their parameters, see the 1   Draft PCL JSON Specification.

ApproximateProgressiveMorphologicalFilter
faster (and potentially less accurate) version of the ProgressiveMorphologicalFilter
GridMinimum
assembles a local 2D grid over a given PointCloud, then downsamples the data
PassThrough
allows the user to set min/max bounds on one dimension of the data
ProgressiveMorphologicalFilter
removes nonground points to produce a bare-earth point cloud
removes outliers if the number of neighbors in a certain search radius is smaller than a given K
StatisticalOutlierRemoval
uses point neighborhood statistics to filter outlier data
VoxelGrid
assembles a local 3D grid over a given PointCloud, then downsamples and filters the data

1. Add the filter function declaration of the form applyMyFilter to PCLPipeline.h.
2. Add the implementation of applyMyFilter to PCLPipeline.hpp.
3. Add a one-line description of the shiny new filter to this file, filters.pclblock.rst.
5. Add a test to PCLBlockFilterTest.cpp. Make sure each parameter is independently verified.