Abstract
Visually-based person recognition is studied most commonly using static images of faces. We studied how recognition accuracy and confidence evolves as a person approaches from a distance in a natural viewing environment. We familiarized participants with 30 people, using multiple video clips showing head rotation, smiling, talking, and walking. To test recognition, participants viewed 8s videos that began with a full-body view from approximately 13.6 meters and continued as the person approached the camera, veering off at the end of the clip. This allowed for a close view of the face only at the end of the video. Specifically, faces subtended approximately .36 x .54 degrees VA at the start point of the video and 2.55 x 3.31 degrees VA at the closest point to the camera. On half of the trials, participants responded (old/new) as soon as they were confident of their judgment (free response trials). On the remaining trials, participants made recognition judgments at three equally spaced time points in the video (multiple response trials), using a 5-point scale (1: sure old to 5: sure new). For the free response trials, known people were recognized accurately throughout the video. RTs were distributed normally around a mean RT of 5054 ms, when the person was approximately 3 meters from the camera. Across all time points, correct recognition of known people occurred faster (i.e., when they were farther away from the camera) than correct rejections of unknown people (t(118) = 2.94, p = .004). The multiple response trials indicated a linear increase in recognition accuracy and confidence over time (p <.01). Finally, to examine the recognition process qualitatively across videos, a cluster analysis was conducted on measures derived from pooled participant responses. This revealed diverse response patterns, indicating that in naturalistic settings, recognition evolves in a variety of ways.
Meeting abstract presented at VSS 2013