University Homepage School Homepage School Contacts School Search

 

Michael Isard

Robotics Research Group, Department of Engineering Science, University of Oxford.

Visual motion analysis using the Condensation algorithm

Abstract

Over the last decade, visual motion analysis has emerged as one of the principal areas of research within the computer vision community. The increasing interest is due in part to the falling cost of computing power and, perhaps as important, the storage necessary to process extended image sequences.

A sequence of video images typically does not change radically from frame to frame, and this redundancy of information can be extremely helpful in disambiguating the visual input. The problem of taking full advantage of the redundancy in an image sequence is challenging.

Our approach is to build probabilistic models to describe the likely motion and appearance of an object of interest. Together with a sequence of input images, these models define a probability density function which encodes all the available information about the positions and velocities of the object in the sequence.

The Condensation algorithm is one of a family of random sampling algorithms called particle filters which are attracting a lot of attention in the nonlinear filtering and statistics communities at the moment. It is particularly well suited to tracking objects in dense background clutter and this talk will give an overview of the algorithm and show examples of tracked motion sequences.