It is useful to understand how changes to a complex interface such as air traffic control will affect operator and system behavior. For example, does a new automated decision aid cause the controller to spend more time looking at the display, or less? Does the amount of time spent looking out the window increase, or decrease? In an operational environment, it is not feasible to have working controllers wear a head-mounted eye tracking device, or remain in one position as required by a stationary off-head eye tracking device. We have therefore explored the recovery of gaze from estimation of head pose in low-resolution video images (left image above). The chart above right shows the computed distribution of fixation targets for a short segment of video, illustrating the type of data which may be obtained from the analysis.

We rely upon the fact that during normal behavior, gaze is not shifted by moving the eyes alone, but rather through a coordinated sequence of both eye and head movements. The eyes typically lead the head, but rarely deviate by more than a few degrees from primary position ("straight ahead" with respect to the head) once the complete movement has been completed.

The process is composed of a number of steps:

Creation of a 3D model of the environment
Construction of a textured 3D model of the head of the controller
Estimation of head pose (position and orientation)
Determination of gaze target by combining head pose estimate with environment model

Additional technical details may be found in our paper:
Model-based head pose estimation for air-traffic controllers
Xavier L.C. Brolly, Constantinos Stratelos and Jeffrey B. Mulligan
(to be presented at ICIP 2003)