The 27th IPP Symposium
BraMBLe: A Bayesian Multiple-Blob Tracker
John MacCormick, Compaq
Blob trackers have become increasingly powerful in recent years, largely due to the adoption of statistical appearance models that allow effective background subtraction and robust tracking of deforming foreground objects. It has been standard, however, to treat background and foreground modeling as separate processes-background subtraction is followed by blob detection and tracking-which prevents a principled computation of image likelihoods. This talk presents two theoretical advances that address this limitation and lead to a robust multiple-person tracking system suitable for single-camera real-time surveillance applications.
The first innovation is a multi-blob likelihood function that assigns directly comparable likelihoods to hypotheses containing different numbers of objects. This likelihood function has a rigorous mathematical basis: it is adapted from the theory of Bayesian correlation, but uses the assumption of a static camera to create a more specific background model while retaining a unified approach to background and foreground modeling. Second, I will describe a Bayesian filter for tracking multiple objects when the number of objects is unknown and varies over time. A particle filter can be used to perform joint inference on both the number of objects present and their configurations, and the resulting system runs comfortably in real time on a modest workstation when the number of objects is small. This is joint work with Michael Isard.
