A Context Based Tracking for Similar and Deformable Objects

Abstract

Object detection and tracking is the most basic and important process to understand video content. Due to its numerous applications, it has attracted many researchers to the field. Lots of work has been done in tracking; however, tracking of a deformable object is still in its initial stage. Most of the start-of-art methods fail to handle rapid variations in an object's shape and size. The task becomes even more difficult when there is multi-target tracking. In this article, the authors have proposed a novel tracking method based on context (spatio-temporal) formed from the corner points of object regions. The correlation between the targets and background based on context is used for tracking. The proposed method is successful in tracking objects that are changing their appearance (shape and size) frequently and are having near similar appearance.