This paper provides a novel technique of efficiently and reliably tracking
features in a sequence of images. The method we provide for tracking featur
es is based on the Bayesian multiple hypothesis tracking (MHT) technique co
upled with a multiple model filtering (MMF) algorithm. We show the results
of our work comparing it with some of the existing single-model-based track
ers using a variety of video sequences. Initially, we demonstrate the abili
ty of the MHT-MMF tracker, and later in the paper we extend the MMF-based t
racker to the interacting multiple model(IMM) tracker and show the superior
ity of the latter in handling motion transition of features efficiently. Th
e primary purpose of this paper is to show how the IMM algorithm combined w
ith an extension of the classical MHT framework can be used in a visual tra
cking scenario. The study shows that the method proposed can distinguish be
tween different motions depicted in an image sequence with good tracking re
sults. (C) 2001 Pattern Recognition Society. Published by Elsevier Science
Ltd. All rights reserved.