In this paper, we analyze and modify the Motion-Split-and-Merge (MSAM)
algorithm for the motion segmentation of correspondences between two
frames. Our goal is to make the algorithm suitable for practical use
which means realtime processing speed at very low error rates. We
compare our (robust realtime) RMSAM with J-Linkage and Graph-Based
Segmentation and show that it is superior to both. Applying RMSAM in a
multi-frame motion segmentation context in the Hopkins 155 benchmark,
we show that compared to the original formulation, the runtime is
reduced by 72%, and the error by 68% to only 0.65%. This is in the
magnitude of the best results reported so far, but RMSAM additionally
allows realtime segmentation, and it can handle outliers and
unconstrained missing data.
Supplementary Videos
Implementation
A Linux binary of our implementation is available for download. It
allows the segmentation of arbitrary correspondences, varying all
parameters, choosing different motion models and thresholds. Also,
access to all internal parameters is possible. Just run the binary for a short
description.
For convenience, we provide m-files for access withing matlab. Just
addpath the rmsam directory and use it by calling
IDX=rsam_segment(correspondences).
