Evaluating pcl/registration

This is a collection of ideas on how to build an evaluation framework of pcl/registration.

Data generation

• synthetic data
• real word data (how to get ground truth?)
  • Kinect
  • PR2 laser scanner
  • SICK laser data
  • small range 3D scanner
  • mid range 3D scanner (Faro)
  • high end 3D scanner (Riegl, Velodyne)
• Point Types
  • 2D(?)
  • 3D
  • RGB
• dynamics
  • static scans
  • scanning while driving (e.g. robots)
• size
  • room
  • building
  • outdoor (street)

Architecture

• some lib for polygonal data
• modeling different sensors
• modeling noise
• add a trajectory file
• output a pile of .pcd files
• integrate command line tools from PCL grandfather

Evaluating different algorithms

ICP ^^^

• how does the algorithm cope with outliers

• how are the point pairs evaluated:

  • does it use normal or RGB information

  • does it weight the pairs differently

  • which kind of point pairs are used:

    • one-to-one
    • one-to-many
    • many-to-many

Similar Projects

• GICP <http://stanford.edu/~avsegal/resources/papers/Generalized_ICP.pdf>_
• Gazebo
• slam benchmarking <http://kaspar.informatik.uni-freiburg.de/~slamEvaluation/index.php>_
• Automated SLAM Evaluation <http://slameval.willowgarage.com/workshop/>_