Feature Trajectory Retrieval with Application to
Accurate Structure and Motion Recovery

7th International Symposium on Visual Computing (ISVC 2011)

Short Abstract

Two Extensions to common Structure and Motion Recovery approach:

1. Feature Trajectory Retrieval (FTR)
increases reliability

2. Gaussian Gradient Approximation in SIFT subpel feature localization
increases accuracy

Results

the video shows a visual comparison of the estimated camera paths with integrated virtual objects [1]

1. input sequence
2. reference SIFT features and frame to frame correspondences only
3. reference SIFT features and Feature Trajectory Retrieval (FTR)
4. Gaussian Gradient Approximation (Gaussian SIFT) and FTR

for an objective evaluation see paper / presentation slides

the combined method - Gaussian SIFT and FTR - provides best results with minimal drift

Paper

[1] Kai Cordes, Oliver Müller, Bodo Rosenhahn, and Jörn Ostermann: Feature Trajectory Retrieval with Application to Accurate Structure and Motion Recovery, Advances in Visual Computing, 7th International Symposium (ISVC), Springer, LNCS 6938, pp. 156--167, 2011

ISVC11-Cordes_FTR-SAM.pdf

Presentation

ISVC2011-pres.pdf

High Quality Video (720x576, 19 MB)

ISVC2011.avi

Models

Snoopy (.obj)

Woodstock (.obj)