SAMANTHA:
Structure-and-Motion Pipeline on a Hierarchical Cluster Tree

M. Farenzena, A. Fusiello and R. Gherardi Department of Computer Science
University of Verona, Verona - Italy

Overview

This work introduces a novel hierarchical scheme for computing Structure and Motion. The images are organized into a tree with agglomerative clustering, using a measure of overlap as the distance. The reconstruction then follows this tree from the leaves to the root. As a result, the problems is broken into smaller instances, which are then separately solved and combined. Compared to the standard sequential approach, this pipeline boost computational efficiency by one order of magnitude, it is independent from the initial pair of views, and copes better with drift problems.

Method

Keypoint matching
Views clusterning
Hierarchical structure-and-motion

Experiments

Images	Reconstruction	Video
Pozzoveggiani 54 JPG images / 4.5 Mb file	11094 points [OFF] 48 cameras Points and cameras [MATLAB]
Piazza Dante 39 JPG images / 5.2 Mb file	2971 points [OFF] 39 cameras Points and cameras [MATLAB]
Piazza Erbe 259 JPG images / 23.3 Mb file	39961 points [OFF] 198 cameras Points and cameras [MATLAB]
Piazza Bra 380 PNG images / 784.4 Mb file	104047 points [OFF] 322 cameras Points and cameras [MATLAB]

Reference Paper

A. M.Farenzena, A.Fusiello, R. Gherardi. Structure-and-Motion Pipeline on a Hierarchical Cluster Tree. Proceedings of the IEEE International Workshop on 3-D Digital Imaging and Modeling, Kyoto, October 2009. (PDF)
Riccardo Gherardi, Michela Farenzena, and Andrea Fusiello. Improving the efficiency of hierarchical structure-and-motion. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2010), pages 1594 – 1600, San Francisco, CA, 13-18 June 2010. (PDF)

SAMANTHA: Structure-and-Motion Pipeline on a Hierarchical Cluster Tree

Overview

Method

Experiments

Reference Paper

SAMANTHA:
Structure-and-Motion Pipeline on a Hierarchical Cluster Tree