As the golden standard in robust estimation, the classic RANSAC approach has undergone extensive research that contributed to further enhancements in run-time performance, robustness, and multi-structure support to name a few. Yet, the accelerating growth of multi-modal co-registered datasets requires a new adaptation of the RANSAC algorithm. In this paper, we propose a multi-modal fault-tolerant extension to RANSAC, termed FT-RANSAC, with a model-independent tolerance to degenerate configurations. Besides building on stateof- the-art RANSAC variants, such as PROSAC, our approach introduces a Hough inspired dimensionality reduction and consistency voting processes, to enable robust estimation in the presence of non-homogenous multi-modal correspondence sets. Through experimental evaluation using homography estimation of RGB-D data, we demonstrate that our approach outperforms the classic single-modality RANSAC in robustness and tolerance to degenerate configurations. Finally, the proposed approach lends itself to parallel multi-core implementations, and could be adapted to specialized RANSAC extensions found in the literature.
A. Barclay, H. Kaufmann: "FT-RANSAC: Towards robust multi-modal homography estimation"; Talk: 8th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS), Stockholm; 08-24-2014; in: "8th IAPR Workshop on Pattern Recognition in Remote Sensing (PRRS), 2014", IEEE, (2014), 1 - 4.
Click into the text area and press Ctrl+A/Ctrl+C or ⌘+A/⌘+C to copy the BibTeX into your clipboard… or download the BibTeX.