Outlier rejection by oriented tracks to aid pose estimation from video

This paper introduces a method for rejecting the false matches of points between successive views in a video sequence used to perform Pose from Motion for a mobile sensing platform. Typical methods for pose estimation require point correspondences to estimate the epipolar geometry between the two views. Algorithms for determining these correspondences invariably output false matches along with the good. We present an algorithm for identifying and removing these mismatches for scenes generated by a mobile scanning platform. The algorithm utilizes the motion characteristics of a rear-wheel drive sensing platform to identify correct point matches through their common motion trajectories. Our algorithm works in cases where the percentage of false matches may be as high as 80%, providing a set of correspondences whose correct/incorrect match ratio is higher than the mutual best match approach found in the literature. This algorithm is intended as a post-processing step for any point corres...

• Type: Journal
• Year: 2006
• Where: PRL
• Authors: Brad Grinstead, Andreas Koschan, Andrei V. Gribok, Mongi A. Abidi, David J. Gorsich
• Keywords: Algorithms, False Matches, Point Correspondence, PRL 2006