CVPR17 icon indicating copy to clipboard operation
CVPR17 copied to clipboard

(Convex) dual approach for global 3D registration between points, lines and planes

Code for Convex Global 3D Registration with Lagrangian Duality (CVPR 17)

This repository contains the implementation of the methods compared in

@inproceedings{briales17CVPR, title = {{Convex Global 3D Registration with Lagrangian Duality}}, author = {Briales, Jesus and Gonzalez-Jimenez, Javier}, booktitle = {International Conference on Computer Vision and Pattern Recognition}, month = {jul}, year = {2017} }

In this work we proposed a novel convex relaxation for registration of points to points, lines and planes. Empirically, the relaxation results always tight and certifies global optimality. Indeed, the framework is able to deal with any optimization problem that has a quadratic objective on rotation matrix elements.

Getting started

Clone the repository

This repository include some dependencies as submodules, so clone it with the --recursive option:

  git clone --recursive https://github.com/jbriales/CVPR17.git

If you already cloned it, you can still set the submodules with

  git submodule update --init --recursive

Install the library

To use the provided code and methods, just run the setup.m script. Note you should have installed CVX (available here) in the path:

  • Download CVX for your platform
  • Install CVX: Run cvx_setup.m from Matlab

Run the examples

For a working example, see example.m.