Template Matching for Head Pose Estimation

Social signals are observable behaviours which are displayed during an interaction. By observing these signals, it is possible to make predictions about the intent of that individual and other individuals within the vicinity. Measuring social signals has often proved challenging as they are often characterised by subtle movements which are difficult to detect. Head pose is one such social signal used to indicate where an individual’s attention is focused. This paper will discuss the problem of head pose estimation by defining the problem in terms of two fields of view, pan and tilt. The presented approach uses template matching and tested using a well-known dataset. The template matching approach is shown to be able to determine head pan to within one class ~51% of the time and head tilt to within one class ~71% of the time.