# Estimating exterior camera parameters by matching lines using least square estimations (Lowe, 1991)

I'm new to computer vision and graphics. I want to estimate exterior camera parameters using non-linear least square estimation as suggested by "Lowe, David G. "Fitting parameterized three-dimensional models to images." IEEE transactions on pattern analysis and machine intelligence 13.5 (1991): 441-450.". However, I'm stuck at the implementation of the same. So, my question is, is there a MATLAB implementation of the same?

Note: I've also posted it in computer science stack exchange

I am not aware of an open implementation of this paper. However, there are other methods to do what you like:

1. One of them is SoftPosIT is one of them. It has many open implementations available: (1), (2), (3). The downside is that one needs to give some sufficiently close initialization.

2. If you can live with planar models, then you could use cross ratios for identification. Line-sweep gives one such nice algorithm.

3. Birdal et. al. also give a simple method for extrinsic camera pose estimation from 2d points not assuming that correspondences are known, but for the moment the open implementation is not available.

4. Line3D++ gives ability to perform line reconstruction from multi-view images: https://github.com/manhofer/Line3Dpp . This doesn't only solve the extrinsic estimation problem, but camera poses should be an intermediate stage in this work. Maybe one could benefit from such reconstruction and can then calibrate the cameras automatically using the available 3D geometry.

There are also many other 2d-3d pose estimation methods (especially for textureless objects), but for your scenario, where simple geometric primitives suffice, I would go for simpler ones, as aforementioned.

• Thank you so much for your time and help. I'm going through the links and will let you know what worked for me. – Debaditya May 18 '17 at 8:14
• Thank you very much, I've tried have tried the link and it works as desired. – Debaditya May 18 '17 at 8:54
• I noticed that (1) estimates the 3D object's pose. So I can know the translation of the object and its rotation. But, my aim is to calculate the pose of the camera. Is there a way to get the pose of the camera from that information? – Debaditya May 18 '17 at 11:45
• If you like to operate in the object space, then the camera pose is just the inverse of the object pose. – Tolga Birdal May 18 '17 at 12:32