12-09-2012, 04:10 PM
Biometric Gait Recognition
[attachment=33632]
Abstract.
Psychological studies indicate that people have a small but
statistically significant ability to recognize the gaits of individuals that
they know. Recently, there has been much interest in machine vision
systems that can duplicate and improve upon this human ability for
application to biometric identification. While gait has several attractive
properties as a biometric (it is unobtrusive and can be done with simple
instrumentation), there are several confounding factors such as variations
due to footwear, terrain, fatigue, injury, and passage of time. This paper
gives an overview of the factors that affect both human and machine
recognition of gaits, data used in gait and motion analysis, evaluation
methods, existing gait and quasi gait recognition systems, and uses of
gait analysis beyond biometric identification. We compare the reported
recognition rates as a function of sample size for several published gait
recognition systems.
Introduction
People often feel that they can identify a familiar person from afar simply by
recognizing the way the person walks. This common experience, combined with
recent interest biometrics, has lead to the development of gait recognition as a
from of biometric identification.
As a biometric, gait has several attractive properties. Acquisition of images
portraying an individual’s gait can be done easily in public areas, with simple
instrumentation, and does not require the cooperation or even awareness of the
individual under observation. In fact, it seems that it is the possibility that a
subject may not be aware of the surveillance and identification that raises public
concerns about gait biometrics [1].
Gait and Gait Recognition
We define gait to be the coordinated, cyclic combination of movements that result
in human locomotion. The movements are coordinated in the sense that they
must occur with a specific temporal pattern for the gait to occur. The movements
in a gait repeat as a walker cycles between steps with alternating feet. It is
both the coordinated and cyclic nature of the motion that makes gait a unique
phenomenon.
Examples of motion that are gaits include walking, running, jogging, and
climbing stairs. Sitting down, picking up an object, and throwing and object are
all coordinated motions, but they are not cyclic. Jumping jacks are coordinated
and cyclic, but do not result in locomotion.
Therefore, we define gait recognition to be the recognition of some salient
property, e.g., identity, style of walk, or pathology, based on the coordinated,
cyclic motions that result in human locomotion. In the case of biometric gait
recognition, the salient property is identity.We make the distinction between gait
recognition and what we call quasi gait recognition in which a salient property
is recognized based on features acquired while a subject is walking, but the
features are not inherently part of the gait. For example, skeletal dimensions may
be measured during gait and used to recognize an individual. However, skeletal
dimensions may be measured other ways, and are therefore not a property of
the gait.
Human Perception of Gait
The ability of humans to recognize gaits has long been of interest to psychologists.
Johansson [2, 3] showed that humans can quickly (in less than one second)
identify that a pattern of moving lights, called a moving light display (MLD),
corresponds to a walking human. However, when presented with a static image
from the MLD, humans are unable to recognize any structure at all. For example,
without knowing that the dots in a single frame of the sequence shown in
Fig. 1 are on the joints of a walking figure, it is difficult to recognize them as
such. What we cannot show in a print medium is, that within a fraction of a
second after the dots move, one can recognize them as being from a human gait.
Johansson’s contributions are important because they provide an experimental
method that allows one to view motion extracted from other contextual information.
With the context removed, the importance of motion becomes obvious.
Johansson also suggests a set of gestalt rules that humans use to connect the
moving dots and infer structure.
Optimistic Viewpoint
Bhanu and Han [6] present an optimistic view of the potential for biometric gait
recognition. Their analysis is built upon a gait recognition system that measures
a subject’s skeletal dimensions as he walks. Therefore, it is possible to estimate
an upper bound on the performance of the system from known distributions of
skeletal dimensions in a human population. They compute their estimate using
a Monte Carlo simulation seeded with the population statistics and a set of
assumptions about the accuracy of the skeletal dimension measurements. Plots
showing the bounds they compute are in Fig. 8.
Since theirs is a quasi gait recognition system, it is reasonable to ask whether
or not the bound might reasonably apply to gait recognition too. Do skeletal
dimensions sufficiently constrain a gait for the purposes of recognition? The
answer is unknown, but work in mechanical engineering can shed some light.
McGeer [7, 8], and later Coleman and Ruina [9], Garcia et al. [10], and Collins
et al. [11] have demonstrated passive mechanical walkers. These are mechanical
machines that oscillate without external force to produce a gait as the machine
falls down an incline. This implies that gait is a natural bi-product of the structure
of the human body, and the mass and skeletal dimensions of the body are
what determine the oscillations that produce the gait. Thus, to a large extent,
Bhanu and Han are right to equate skeletal dimensions with gait. However, mass
and other factors contribute to a human gait.
Human Performance
People often have the impression that they can recognize friends by their gaits.
Although this ability has been confirmed by experiments using MLDs, human
ability to recognize people from motion is limited.
For example, Barclay et al. [12], and Kozlowski and Cutting [13] showed
that humans can recognize the gender of a walker from an MLD. However, for
short exposures to the MLD (two seconds or less), humans were no better than
random. It required longer exposures, on the order of four seconds, for humans
to perform better than random. Even at that, the recognition rate was 66% when
random was 50%.
[attachment=33632]
Abstract.
Psychological studies indicate that people have a small but
statistically significant ability to recognize the gaits of individuals that
they know. Recently, there has been much interest in machine vision
systems that can duplicate and improve upon this human ability for
application to biometric identification. While gait has several attractive
properties as a biometric (it is unobtrusive and can be done with simple
instrumentation), there are several confounding factors such as variations
due to footwear, terrain, fatigue, injury, and passage of time. This paper
gives an overview of the factors that affect both human and machine
recognition of gaits, data used in gait and motion analysis, evaluation
methods, existing gait and quasi gait recognition systems, and uses of
gait analysis beyond biometric identification. We compare the reported
recognition rates as a function of sample size for several published gait
recognition systems.
Introduction
People often feel that they can identify a familiar person from afar simply by
recognizing the way the person walks. This common experience, combined with
recent interest biometrics, has lead to the development of gait recognition as a
from of biometric identification.
As a biometric, gait has several attractive properties. Acquisition of images
portraying an individual’s gait can be done easily in public areas, with simple
instrumentation, and does not require the cooperation or even awareness of the
individual under observation. In fact, it seems that it is the possibility that a
subject may not be aware of the surveillance and identification that raises public
concerns about gait biometrics [1].
Gait and Gait Recognition
We define gait to be the coordinated, cyclic combination of movements that result
in human locomotion. The movements are coordinated in the sense that they
must occur with a specific temporal pattern for the gait to occur. The movements
in a gait repeat as a walker cycles between steps with alternating feet. It is
both the coordinated and cyclic nature of the motion that makes gait a unique
phenomenon.
Examples of motion that are gaits include walking, running, jogging, and
climbing stairs. Sitting down, picking up an object, and throwing and object are
all coordinated motions, but they are not cyclic. Jumping jacks are coordinated
and cyclic, but do not result in locomotion.
Therefore, we define gait recognition to be the recognition of some salient
property, e.g., identity, style of walk, or pathology, based on the coordinated,
cyclic motions that result in human locomotion. In the case of biometric gait
recognition, the salient property is identity.We make the distinction between gait
recognition and what we call quasi gait recognition in which a salient property
is recognized based on features acquired while a subject is walking, but the
features are not inherently part of the gait. For example, skeletal dimensions may
be measured during gait and used to recognize an individual. However, skeletal
dimensions may be measured other ways, and are therefore not a property of
the gait.
Human Perception of Gait
The ability of humans to recognize gaits has long been of interest to psychologists.
Johansson [2, 3] showed that humans can quickly (in less than one second)
identify that a pattern of moving lights, called a moving light display (MLD),
corresponds to a walking human. However, when presented with a static image
from the MLD, humans are unable to recognize any structure at all. For example,
without knowing that the dots in a single frame of the sequence shown in
Fig. 1 are on the joints of a walking figure, it is difficult to recognize them as
such. What we cannot show in a print medium is, that within a fraction of a
second after the dots move, one can recognize them as being from a human gait.
Johansson’s contributions are important because they provide an experimental
method that allows one to view motion extracted from other contextual information.
With the context removed, the importance of motion becomes obvious.
Johansson also suggests a set of gestalt rules that humans use to connect the
moving dots and infer structure.
Optimistic Viewpoint
Bhanu and Han [6] present an optimistic view of the potential for biometric gait
recognition. Their analysis is built upon a gait recognition system that measures
a subject’s skeletal dimensions as he walks. Therefore, it is possible to estimate
an upper bound on the performance of the system from known distributions of
skeletal dimensions in a human population. They compute their estimate using
a Monte Carlo simulation seeded with the population statistics and a set of
assumptions about the accuracy of the skeletal dimension measurements. Plots
showing the bounds they compute are in Fig. 8.
Since theirs is a quasi gait recognition system, it is reasonable to ask whether
or not the bound might reasonably apply to gait recognition too. Do skeletal
dimensions sufficiently constrain a gait for the purposes of recognition? The
answer is unknown, but work in mechanical engineering can shed some light.
McGeer [7, 8], and later Coleman and Ruina [9], Garcia et al. [10], and Collins
et al. [11] have demonstrated passive mechanical walkers. These are mechanical
machines that oscillate without external force to produce a gait as the machine
falls down an incline. This implies that gait is a natural bi-product of the structure
of the human body, and the mass and skeletal dimensions of the body are
what determine the oscillations that produce the gait. Thus, to a large extent,
Bhanu and Han are right to equate skeletal dimensions with gait. However, mass
and other factors contribute to a human gait.
Human Performance
People often have the impression that they can recognize friends by their gaits.
Although this ability has been confirmed by experiments using MLDs, human
ability to recognize people from motion is limited.
For example, Barclay et al. [12], and Kozlowski and Cutting [13] showed
that humans can recognize the gender of a walker from an MLD. However, for
short exposures to the MLD (two seconds or less), humans were no better than
random. It required longer exposures, on the order of four seconds, for humans
to perform better than random. Even at that, the recognition rate was 66% when
random was 50%.