Region-based pose tracking

Abstract

In this article, we present a technique for region-based pose tracking that uses a surface model of the object to be tracked and at least one calibrated camera view. The proposed algorithm can follow the movement of rigid objects or kinematic chains without explicitly computing contours in the image. The aim to match the model surface to the contour of the object seen in the image is achieved by optimizing an energy with respect to the pose parameters. We explain how kinematic chains can be included, and how using knowledge from the previous frame can help to track objects that move with a high speed or accelerate strongly. We present experimental results for complex scenes observed with one, two, or four cameras.