A random projections model for object tracking under variable pose and multi-camera views

Embedded vision systems, such as smart cameras, provide a new framework for computer vision algorithms in resource constrained environments. In this paper, we present a new object tracking methodology based on random projections, which offers the benefits of fast, low-complexity transformation of the input data into accurate and computationally attractive representations. Random projections are used for the generation of a template library that describes the object's appearance and achieves robustness under pose variations. Furthermore, the random projections model is used for reliable handoff between different cameras with partially overlapping fields of view. The proposed object tracking algorithm is tailored to the limited processing capabilities of smart cameras by requiring reduced network bandwidth during camera handoff and low memory requirements for the template library maintenance. Experimental results indicate that the proposed algorithm can maintain robust tracking under varying object pose and across camera views while using limited resources, a key benefit for embedded vision systems.