Date of Award

2009

Degree Type

Thesis

Degree Name

Master of Science

Program

Computer Science

Supervisor

Yuri Boykov

Abstract

RANSAC [15, 38, 1] is a reliable method for fitting parametric models to sparse data with many outliers. Originally designed for extracting a single model, RANSAC also has variants for fitting multiple models when supported by data. Our main insight is that, in practice, inliers for each model are often spatially coherent — all previous RANSAC-based methods ignore this. Our new method fits an unspecified number of models to data by combining ideas of random sampling and spatial regularization. As in basic RANSAC, we randomly sample data points to generate a set of proposed models (labels). We formulate model selection and inlier classification as a single problem — labeling of triangulated data points. Geometric fit errors and spatial coherence are combined in one MRF-based energy. In contrast to basic RANSAC, inlier classification does not depend on a fixed threshold. Moreover, our optimization framework allows iterative re-estimation of models/inliers with a clear stopping criteria and convergence guarantees. We show that our new method, SCO- RANSAC, can significantly improve results on synthetic and real data supporting multiple linear, affine, and homographic models.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.