Skip to content

A Collection of Algorithms for Relative Pose Estimation of a Calibrated Camera

License

Notifications You must be signed in to change notification settings

prclibo/relative_pose

Repository files navigation

Relative Pose Estimation of Calibrated Cameras with Known SE(3) Invariants

A Collection of Algorithms for Relative Pose Estimation of a Calibrated Camera

BSD-3-Clause ECCV 2020

Introduction

This repository contains the following relative pose estimation solvers, in C++ and Matlab API:

  • The conventional 5-point algorithm (5P), wrapped from Hartley's well-known implementation.
  • 4-point algorithm with a known rotation angle (4P-RA)
  • 4-point algorithm under planar motion without knowing the plane direction (4P-ST0)
  • 3-point algorithm with a known rotation angle and under planar motion without knowing the plane direction (3P-RA-ST0).

Compared to other relative pose estimation algorithms, 4P-RA, 4P-ST0 and 3P-RA-ST0 leverage extra sensor/motion constraits without requiring extrinsics calibration. This is due to the interesting property of SE(3) invariants.

Reference

This repository is the source code for the following paper:

@article{li2020relative,
  title={Relative Pose Estimation of Calibrated Cameras with Known SE(3) Invariants},
  author={Li, Bo and Martyushev, Evgeniy and Lee, Gim Hee},
  journal={ECCV},
  year={2020}
}

Contents

Compilation

The default CMakeLists.txt requires Matlab to be installed to compile the Matlab API.

mkdir build && cd build
PATH=/usr/local/MATLAB/<version>/bin/:$PATH cmake ..
make

About

A Collection of Algorithms for Relative Pose Estimation of a Calibrated Camera

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published