Efficient Algorithms for Maximum Consensus Robust Fitting in Computer Vision

Code for the paper: "F. Wen, R. Ying, Z. Gong, and P. Liu, Efficient Algorithms for Maximum Consensus Robust Fitting, IEEE Transactions on Robotics, 2019".

To use the code, firstly unzip 'src.zip'.

This code is modified from the code of Huu Le at https://www.researchgate.net/publication/320707327demo_pami, details please see the paper "H. Le, T. J. Chin, A. Eriksson, and D. Suter, “Deterministic approx-imate methods for maximum consensus robust fitting,” arXiv preprint, arXiv:1710.10003, 2017."

Note that, some codes of Huu Le are directly coped here from https://www.researchgate.net/publication/320707327demo_pami to facilitate the ease of use for interested readers who want to reproducing the results in our paper. The coped codes (contained in the '/src' folder) include the EP method, RANSAC method and their dependency. We copy them here only for academic use purpose to illustrate the comparison results of the algorithms in our paper.

SeDuMi is need in solving the LP problems, which is available at http://sedumi.ie.lehigh.edu/, we used SeDuMi 1.3.

Application in SLAM

The “C++ for Application in ORB-SLAM2” folder contains the C++ implementation of the ADMM algorithm for homography and fundamental matrix estimation, which can be directly used in the popular open source ORB-SLAM2 system (https://github.com/raulmur/ORB_SLAM2).

It has been tested in a monocular example of the KITTI dadaset.