Object-of-interest selection for model-based 3D pose tracking with background clutter

Abstract

Extensive research on model-based 3D object tracking has yielded a number of effective methodologies. However, work reported in the literature on initiating tracking has been limited. This paper addresses this issue via a novel framework that can automatically find an Object-of-Interest (OI) in a dynamic scene and initiate tracking. Since OI definition is, typically, application dependent, the proposed framework is modular and customizable. It combines a realtime motion segmentor with a set of customizable interest filters to separate, highlight, and select the OIs. Furthermore, our earlier model-based object tracker is extended in this paper to utilize OI-selection data and track objects in the presence of background clutter. Thus, the overall computer-vision system presented in this paper can automatically select, model, and track the six degree-of-freedom position and orientation of an OI, whose model is not known a-priori. Proposed algorithms were verified via extensive simulations and experiments, some of which are presented herein. © Springer Science+Business Media B.V. 2008.