Object tracking

Abstract

Object tracking is an active research problem in computer vision. It is needed in several areas including video indexing, medical therapy, interactive games, and surveillance systems. Tracking and detection are very critical in monitoring systems as their accuracy greatly impacts the eventual success or failure of later scene analysis. This chapter reviews the main approaches previously proposed, top-down known as filtering and data association and bottom-up known as target representation and localization. It also presents a Bottom-up correspondence Matching scheme based on Non-Linear similarity Voting, named BuM-NLV. This scheme benefits from the lightweight requirements of bottom-up approaches while being robust to occlusions and segmentation errors. © 2014 Springer Science+Business Media, LLC.