catree / mlpnp-python

Python implementation of ML-PnP

Python implementation of ML-PnP

Credits

This code is based on the work from Steffen Urban et al. : MLPnP - A Real-Time Maximum Likelihood Solution to the Perspective-n-Point Problem.

Also, this implementation is partly based on the author's Matlab implementation.

If you use MLPnP, please cite the original paper :

@INPROCEEDINGS {mlpnp2016,
  title={MLPNP - A REAL-TIME MAXIMUM LIKELIHOOD SOLUTION TO THE PERSPECTIVE-N-POINT PROBLEM},
  author={Urban, Steffen and Leitloff, Jens and Hinz, Stefan},
  booktitle={ISPRS Annals of Photogrammetry, Remote Sensing \& Spatial Information Sciences},
  pages={131-138},
  year={2016},
  volume={3}
}

Documentation

The mlpnp.py file containts all the functions to execute ML-PnP.

The function executing it is mlpnp.

The main function is provided for testing purposes :

• the algorithm randomly generates a 3D transformation from camera to world
• it samples random points in camera frame
• the transform is used to generate world coordinates (ensuring that they are in the camera FOV)
• then this set of simulated points and measurements are used in mlpnp to recover the ground truth 3D-transformation