Learning a Local Feature Descriptor for 3D LiDAR Scans

An illustration of keypoint matching using our learned feature descriptor. Surface patches around the keypoints are passed through the feature learning network to estimate feature descriptors. These descriptors are then passed through the metric learning network to estimate a matching score. Red lines show the correspondences between the keypoints in the two 3D LiDAR scans.

Publication

• Ayush Dewan, Tim Caselitz, Wolfram Burgard
  Learning a Local Feature Descriptor for 3D LiDAR Scans
  In Proc. of the IEEE/RJS International Conference on Intelligent Robots and Systems (IROS), Madrid, Spain, 2018
  Download BibTeX

  @inproceedings{dewan2018iros,
    author = {Ayush Dewan and Tim Caselitz and Wolfram Burgard},
    title = {Learning a Local Feature Descriptor for 3D LiDAR Scans},
    booktitle = {Proc.~of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
    address = {Madrid, Spain},
    year = {2018},
    url = {http://ais.informatik.uni-freiburg.de/publications/papers/dewan18iros.pdf}
  }
  
  
  

Code and Dataset

We provide the scripts to train the model and a C++ library to interface the learned decriptor with PCL. We also provide the training and testing data used for learning the descriptor and example data to use the C++ API. For more details please refer to the GitHub page.