Efficient Appearance-Based Tracking

One of the major challenges that visual tracking algorithms face nowadays is being able to cope with changes in the appearance of the target during tracking. Linear subspace models have been extensively studied recently and are possibly the most popular way of modeling target appearance. Unfortunately, efficiency is one of the limitations of present linear subspace models, and this is a key feature for a good tracker. In this paper we present an efficient procedure for tracking based on a linear subspace model of target appearance (grey levels). A set of motion templates is built from the subspace base, which is used to efficiently compute target motion and appearance parameters. It differs from previous works in that we impose no restrictions on the subspace used for modeling appearance. In the experiments conducted we have built a modular PCA-based face tracker which shows that video-rate tracking performance can be achieved with a non optimized implementation of our algorithm