Abstract:

The goal of this work is to classify the focus of attention of a subject who is switching his or her attention between a number of surrounding objects. The specific application is to classify the focus of attention of a car driver as straight-ahead, towards the rear-view mirror, towards the dashboard etc. An explicit quantitative approach to this problem requires {bf (a)} {rm a priori} information about the interior geometry of the car and the calibration of the camera, and {bf (b)} accurate computation of the subject's location and gaze direction. This paper describes a more qualitative approach. The subject is observed over an extended period of time, and a "pose-space histogram" is used to record the frequency with which particular head poses occur. For observation of a car driver, peaks will appear in the histogram according to the frequently viewed directions of straight-ahead, toward the dashboard, and the mirrors. Each peak is labelled, and the focus of attention of the driver in all subsequent images is then classified by use of the histogram.