Point cloud

A point cloud is a set of data points in space.

Libraries

PCL - Point Cloud Library

Links:

• repo: https://github.com/PointCloudLibrary/pcl

Dependencies:

• FLANN - Fast Library for Approximate Nearest Neighbor
• Eigen
• VTK - manipulating and displaying scientific data

Cons:

• transitive dependencies can be hard to install

See an example Dockerfile.

Cilantro

“A lean C++ library for working with point cloud data "

Links:

• code: https://github.com/kzampog/cilantro
• technical report: (https://arxiv.org/abs/1807.00399)

Bundled dependencies:

• nanoflann for general dimension kd-trees
• tinyply for reading pointcloud data
• QHull for convex hull computation
• Spectra for spectral clustering

External dependencies:

• Eigen
• Pangolin for OpenGL display/interaction

Pros: easy to compile

Cons: no python bindings?

Open3d

“A modern library for 3D processing”

Links:

• site: http://www.open3d.org/
• repo: https://github.com/intel-isl/Open3D
• intro: http://www.open3d.org/docs/release/tutorial/Basic/pointcloud.html

Cons:

• does not seem to have an implementation of Generalized ICP.

Algorithms

Iterative Closest Point is one of the algorithms implemented by PCL Generalized_ICP.

A related approach is Real time correlative scan matching

Data formats

• pcd: http://pointclouds.org/documentation/tutorials/pcd_file_format.php
• ply: https://en.wikipedia.org/wiki/PLY_(file_format)
• stl: https://en.wikipedia.org/wiki/STL_(file_format)

Competitions datasets

Multiple Object Tracking Benchmark: https://motchallenge.net/