Utilities

Before diving into the intricacies of face tracking, a number of book-keeping tasks and conventions common to all face-tracking methods must first be introduced. The rest of this section will deal with these issues. An interested reader may want to skip this section at the first reading and go straight to the section on geometrical constraints.

Object-oriented design

As with face detection and recognition, programmatically, face tracking consists of two components: data and algorithms. The algorithms typically perform some kind of operation on the incoming (that is, online) data by referencing prestored (that is, offline) data as a guide. As such, an object-oriented design that couples algorithms with the data they rely on is a convenient design choice.

In OpenCV v2.x, a convenient XML/YAML file storage class was introduced that greatly simplifies the task of organizing offline data for use in the algorithms. To leverage this feature, all classes described in this chapter will implement...