---

-- Note --

A newer, better tested, python only implementation of this code, with better documentation, is available at https://github.com/memorydump85/undistort.

---

local-homography

Paper

The paper decsribing this method is available at
http://april.eecs.umich.edu/papers/details.php?name=ranganathan2014iros

Datasets used in the research publication are checked into the repository as datasets.tar.gz

An online demo of this software is available at
http://rpradeep.webhop.net/local-homography/

Code organization

This code is for locally weighted homography estimation and non-parameteric lens distortion correction.

The code is organized as a bunch of python/jython scripts invoking java code to perform distortion correction on a tag-mosaic image. The evaluation scripts operate on a tag-mosaic image directory to compare the classic camera calibration method with the augmented-method using the non-parametric lens distortion.

This set of scripts has a dependency on the april robotics toolkit java library:
http://april.eecs.umich.edu/wiki/.

Most of the useful code lives in jython/estim with adhoc evaluation scripts in jython/eval. Each script performs some calculation and spills its guts into a JSON output file. Subsequent scripts read from these JSON files and do further processing.

Script Pipeline

This is the rough sequence for the pipeline of estimation scripts:

(A) Start with straightness evaluation on undistorted image

1. Obtain tag training correspondences > (corrs0.json)
2. Obtain tag testing correspondences > (corrs1.json)
3. Estimate local homography at center > (homographies0.json)
4. Estimate GP/poly undistortion model > (undistortion_model.json)
5. Estimate Gaussian process undistortion table > (gptable.json)
6. Evaluate straightness > (straightness.json)

(B) Classic vs. Augmented calibration comparison

1. Obtain tag training correspondences > (corrs1.json)
2. Obtain tag testing correspondences > (corrst.json)
3. Initial intrinsics estimate > (intrinsics_d.json)
4. Classic camera calibration > (calib_classic.json)
5. Non-parametric undistort correspondences > (ucorrs1.json)
6. Initial intrinsics estimate (after undistortion) > (intrinsics.json)
7. Non-parametric undistort test correspondences > (ucorrst.json)
8. Monte-carlo sample from undistortion model > (mc_urcorrs1.json)
9. Augmented camera calibration > (calib_augmented.json)
10. Calibration convergence comparison > (calib_convergence.json)