Adaptive Template Based Object Tracking with Affine Model

Abstract

Traditional model-based tracking methods apply template of changeless size and shape to represent the object, which are only suitable for target translation. Whereas, in the course of tracking there are various transformations for the object, including flip, rotation, scaling and shear. This paper presents a new method that can perform well in coping with object's affine transformation and make the template follow the object's state duly and precisely. In the related algorithm, Newton-Downhill method is utilized to calculate the six parameters of affine transform model. The template is updated based on the final matching result, modulating its size and shaping to follow the temporal evolution of the object.