Dear Colleagues,

We are excited to release the code of the paper “TEASER: Fast and Certifiable Point Cloud Registration”.

The code solves challenging 3D registration problems (also known as point cloud alignment or scan matching) where many (e.g., 99%) point correspondences (e.g., from traditional or deep-learned descriptors) are incorrect.

Our algorithm (TEASER++)

• runs in milliseconds and outperforms popular alternatives (e.g., RANSAC,  Fast Global Registration, GORE)
• is extremely robust (>99% outlier correspondences). TEASER++ can even solve correspondence-free registration (with up to few hundred points) using all-to-all correspondences.
• is certifiable: it either confirms it was able to find the best solution given the data, or declares failure (it is the first certifiable algorithm for outlier-corrupted problems).
• It provides formal performance guarantees (first of their kind for robust registration).

Paper: https://arxiv.org/abs/2001.07715 (conference version: https://arxiv.org/abs/1903.08588)

Code: https://github.com/MIT-SPARK/TEASER-plusplus

Video: https://youtu.be/xib1RSUoeeQ

TEASER++ can be used in many applications from object detection, scan matching, 3D reconstruction, panorama stitching, and medical imaging.

The certification routine is currently given in Matlab and will be released in C++ shortly.

We will keep maintaining TEASER++ so feel free to ping us with github issues if you have comments or questions.

We hope you find it useful!

All the best,

Luca, Hank, Jingnan

——————————————————————————————

Luca Carlone

Charles Stark Draper Assistant Professor

Department of Aeronautics and Astronautics

Laboratory for Information and Decision Systems (LIDS)

Massachusetts Institute of Technology (MIT)

office: 77 Massachusetts Ave, Cambridge, MA 02139, Room: 31-243

website: https://lucacarlone.mit.edu/

Spark Lab: http://web.mit.edu/sparklab/   

————