Estimation of Camera Locations in Highly Corrupted Scenarios: All About that Base, No Shape Trouble, CVPR 2018.

This project proposed a strategy for improving camera location estimation in structure from motion. Our strategy identifies severely corrupted pairwise directions by using a geometric consistency condition. It then selects a cleaner set of pairwise directions as a preprocessing step for common solvers. Please see https://arxiv.org/pdf/1804.02591.pdf for details.

What are included

4 matlab files:

The following function generates synthetic data using uniform corruption model of [1]. See detailed instructions in the header of the file.

UniformCorruptionModel.m

The following function computes Naive AAB statistic given a graph (adjacency matrix) and a set of pairwise directions. See detailed instructions in the header of the file.

NaiveAAB.m

The following function computes IR-AAB statistic given a graph (adjacency matrix) and a set of pairwise directions. See detailed instructions in the header of the file.

IRAAB.m

The following file demonstrates the performance of AAB statistics given synthetic data generated from uniform corruption model. The code for drawing scatter plots and ROC curves are included.

demo.m

License

© Regents of the University of Minnesota. All rights reserved.

Reference

[1] Yunpeng Shi and Gilad Lerman, Estimation of Camera Locations in Highly Corrupted Scenarios: All About that Base, No Shape Trouble, CVPR 2018.