Abstract
How accurately can you determine what another person is looking at? This ability is essential for joint attention, a core, early-developing form of social interaction. While it has been shown that people can accurately determine whether another person is looking directly at them versus away, little is known about our ability to determine where another person is looking when their gaze is directed away from us. We measured subject's accuracy in 'reading" another person's gaze in a study that met three criteria: (1) subjects viewed another person's natural looking behavior live while that person (the "Actor") gazed at experimenter-indicated objects (2) without any constraint on head and eye movements and (3) in a crowded display containing many objects. We used a Kinect sensor to capture Actors' head movements for modeling the target's gaze behavior. In our "what-are-you-looking-at" task, the Actor gazes at one of 52 objects arrayed on a table in an arc around the Actor. In each trial, the Actor gazes at a different object for 9s. Our participants seated on the opposite side of the table to decide which target the Actor is looking at. Objects are spaced 10° apart in the Actor's visual angle. Our results indicate that: (1) Participant's accuracy of choosing the exactly correct object is about 40%, suggesting a limit on the acuity of joint attention; (2) Observers' incorrect responses are not random, but are systemically biased toward certain regions of the actor's visual field; (3) Individual observers differ reliably from each other in their accuracy, which ranges from 20%~60%; (4) The difference in readability of each Actor's gaze can be largely explained by the variance of the head movements recorded by the Kinect sensor. These results collectively reveal the acuity of joint attention by showing its limit, bias and variance.
Meeting abstract presented at VSS 2014