Spontaneous Perception of Whole Persons According the Distance
Samuel Demarchi
Laboratoire Parisien de
Psychologie Sociale (EA 4386)
Université Paris 8, France.
samuel.demarchi@univ-paris8.fr
Anaïs Taddeï
Laboratoire Cognition
Humaine et Artificielle
Université Paris 8, France.
anais.taddei@gmail.com
Charlie Frowd
School of Psychology
University of Central
Lancashire, UK.
cfrowd@uclan.ac.uk
Abstract
In the field of automated facial identification, relevant
facial information are not always available (disguises,
poor spatial orientation, distance, etc.). As a
consequence, human perception performance decreases.
However, the "whole person" can provide useful
information. In the present study, we determine the way
by which a person is visually explored, in order to identify
a potential chronology in visual exploration. We also
investigate the consequence of increasing the distance
between the observer and the target person, since in this
situation identification becomes more ‘global’ due to less
processing possible for the individual features of the face.
Results showed that the spontaneous perceptual strategy
evolves with distance: the more distance increases, the
more the interest in the person decreases, which is
observed both for the body and face. Our data provide
knowledge that will help to guide conception of effective
automated systems.
1. Introduction
The use of video for automated identification of
criminals is a booming industry. Current surveillance
systems can detect the presence of a person sought by the
police in a public place, analyzing person characteristics
such as face or gait [1,2,3,4]. However, no system is yet
fully autonomous. Current performance depends on the
ability of humans to integrate information. Computer
systems analyze and are able to give a warning (presence
of a wanted person, motion detection, crossings,
misinterpretation, poaching, etc.), but an operator is still
required to make a decision in this process.
We postulate that the evolution of human beings
allows us to select—by the processes of selection,
competition and trial-and-error—the most efficient
perceptual processing strategy. In fact, the majority of
analysis systems of automatic perception and
identification try to emulate natural human functions. For
example, automatic facial recognition is more efficient
when algorithms analyze faces holistically, as evidenced
by how we naturally recognize familiar faces [5,6]. In
cases where facial information is unavailable (due to
disguise, misdirection of the camera relative to the face,
or where the face is simply occluded by an object), we
have to use other methods to estimate identity, such as the
“silhouette” technique described below [7,8,9,10].
However, there is not, to our knowledge, studies that
specifically focus on the visual exploration of the whole
person or body shape. Therefore, it is difficult to propose
effective techniques.
A fundamental characteristic of identification is the
distance between the stimulus (target) and the video
acquisition device. Currently, cameras already installed
may not provide sufficient accuracy when distance
increases, and high-definition cameras are expensive, thus
hindering installation by states, communities and
businesses. The lack of precision due to remoteness is
therefore of interest. A fairly-recent study [11] has shown
that we do not process faces in the same way with
changing distance, due to variation in spatial frequency.
Up close, high spatial frequencies (HSF) are not filtered
[12,13]. By contrast, when distance increases, the
environment naturally filters these high frequencies to
present only the lower, less accurate frequencies. In other
words, detailed information conveyed by high spatial
frequencies are not visible from far away, and we see only
an imprecise shape due to the predominant presence of
low spatial frequencies (LSF). Thus, the greater the
distance, the more the identification tasks will be based on
treatment of LSF when perceiving the whole person, thus
corresponding to a “silhouette”.
So, if the exploration of a person depends on the
spatial frequency, we should be able observe differences
between conditions about the duration of fixation and
areas of interest (AOI). Specifically, it is expected that,
for a distant person, a global, holistic treatment will occur
because it is the only possible processing due to natural
low spatial frequencies filtering. When a person is seen
close up, due to the presence of high spatial frequencies,
2012 Third International Conference on Emerging Security Technologies
978-0-7695-4791-6/12 $26.00 © 2012 IEEE
DOI 10.1109/EST.2012.28
33
2012 Third International Conference on Emerging Security Technologies
978-0-7695-4791-6/12 $26.00 © 2012 IEEE
DOI 10.1109/EST.2012.28
33