4.11 - 2D and 3D Motion Tracking in Digital Video

Abstract

This chapter describes the basic principles behind 2D and 3D object-tracking algorithms and provides a basic overview of the subject. 2D and 3D object-tracking algorithms are analyzed on the basis of rigid object tracking and articulated object tracking. Methods for 2D rigid object tracking can be classified in different categories according to the tools that are used in tracking: region-based methods, contour-based methods, feature-based methods, and template-based methods. 3D rigid object tracking can be defined as the estimation of the position and orientation of a rigid object in a 3D space from one or more video sequences. 3D articulated object-tracking approaches can be model free or model based. In the former case, no model of the articulated structure is used. Instead, a bottom-up approach is used to combine image information extracted locally to create coherent structures such as the limbs of the human body. In the model-based approaches, which are the most used in 3D-articulated object tracking, a model of the articulated object is used. The complexity of the model depends on the accuracy required in a specific application. 2D-articulated object tracking can be model free or model based. Model-free methods proceed by exploiting image information to create coherent structures that correspond to the rigid parts of the articulated structure. Model-based approaches use a 2D or 3D model of the articulated object, depending on the application and the precision required.