of footstep data. This floor system
may be used to transparently identify users in their
everyday living and working environments. We have
created user footstep models based on footstep profile
features and have been able to achieve a recognition rate of
93% using this feature-based approach. We have also
shown that the effect of footwear is negligible on
recognition accuracy.
Keywords
Interaction technology, ubiquitous computing, tactile I/O,
user
identification,
biometrics,
indoor
positioning,
intelligent home environments, intelligent systems.
INTRODUCTION
Transparent user identification has in recent years become
a goal of computer science researchers. With the advent of
ubiquitous computing [17] and smart environments,
transparent user identification has become an even more
pressing goal than before the rise of these paradigms. If a
computer or environment could transparently identify the
user, it could customize its interface and behavior to match
the preferences, history, and context of that particular user.
Computers could become easier to use, and could
themselves become more transparent overall. But the
system must provide a service or services of enough value
that the user will tolerate the additional technology, no
matter how transparent it may be.
For example, many users may appreciate a system that
tracks certain objects around their living spaces for them.
Frequently lost objects, such as keys, wallets, and glasses,
could be tagged with small radio frequency ID tags, and
their location could be correlated with the location of
people in the house [8]. As an example, if Mary were to
walk out of the house with a set of keys and Joe needed to
locate the keys some time later, the system could inform
Joe that The keys were last seen 30 minutes ago at the
front door with Mary. Joe could then deduce that the
keys were with Mary and coordinate with her. In this case,
the identity and location of Mary is an important piece of
information. Furthermore, if the system cannot
transparently track and identify users, they will be much
less likely to use the system, and the services the system
offers will be much less likely to be successful. On the
other hand, if the system is transparent enough, the ability
to track frequently lost objects may be a compelling
enough service that users are willing to be tracked and
identified, and willing to have that information made
available to a small group of people within the house.
However, most identification methods to date have not
been especially transparent. The Active Badge system [16]
is one of the most widely used systems and illustrates some
of the problems with many identification schemes. (Radio
frequency identification systems, or RFID, have many of
the same features and problems.) First and foremost, the
user must carry a badge or tag in order to be identified.
While this can be a feature at times--if the user desires
privacy, all she does is remove the badge--it can be an
impediment to use and also narrowly defines the
environments in which the system can be used. For
example, the Active Badge system is not particularly
amenable to use in a smart house: users will not wear the
badge while sleeping (in order, for example, that the house
can identify them when they arise to use the facilities in
the middle of the night), or while doing work in the yard.
They must remember to put it on when they arise or come
back into the house. In addition, adding new users, such
as frequent visitors, requires another physical badge or tag.
Finally, badge systems only provide gross positioning.
The best badge-based indoor positioning system to date
[18] only has a resolution of 6 feet. In many cases, we will
want to know the position of a user to a finer resolution.
There has also been much work recently that has focused
on more passive forms of user recognition, such as face
recognition using video and voice recognition using audio.
These types of recognition do not require that the user
carry any tag or badge; they utilize only biometric data
from the user. Video and audio can also both be used to
track the location of users in a space, and to a much finer
resolution than badge and tag systems. However, these technologies have problems, too. Video recognition is
stymied
by
occlusions,
shadows,
and
lighting
inconsistencies, and wont work at all in the dark. Audio
recognition suffers from problems of background noise and
requires the user to speak in order to function (not very
useful in the middle of the night when one is trying to
tiptoe quietly without waking the other occupants of the
house).
An Alternative Biometric Approach
Passive biometric approaches have the advantage that the
user does not need to carry anything or remember
anything. The badge and tag systems require the user to
carry a badge or tag, but they work just fine in noisy
environments and occluded or dark rooms. We have
designed a system that has many of the advantages and
few of the drawbacks of both classes of systems. Our
Smart Floor system measures the force exerted on a floor
tile and is able to recognize the user based on their footstep
force profile as they walk over the tile.
The floor has a number of characteristics that make it an
obvious choice for instrumentation: users always walk over
it; it is always there (even in the dark); and it can sense
information not only about users but also about objects.
With the Smart Floor, the user does not need to carry
anything (like a badge) or remember anything (like a
password); she simply walks over the floor tile and the
system utilizes the users biometric data for recognition.
The Smart Floor also works fine when the room is dark or
noisy, and it does not care if a view of the user is occluded.
In addition, by its very nature the floor gives accurate
position information. Lastly, the algorithms for
identification and tracking are fairly simple and not
computationally intensive.
Purpose of the Project
The purpose of the Smart Floor project [7] has been to
create and validate biometric user identification based on
users footsteps. As mentioned above, we have outfitted a
floor tile with force measuring instruments and are using
the data gathered as users walk over the tile to identify
them. We rely on the fact that footstep profiles are unique
enough within a small enough group of people to provide
recognition capabilities matching or exceeding the
capabilities of other biometric technologies. (We will
address this claim further below.) Specifically, we have
been able to achieve a 93% overall user recognition rate
with our system, and have been able to show that footwear
is not a significant factor in identifying users.
Furthermore, we have created a system that can
transparently identify users and now allows us to prototype
useful services for users.
We had a number of research goals for the Smart Floor
system at the start of this project:
Create an accurate system for recognizing a
users identity from their footsteps;
Investigate the similarity of users footsteps
and show that for a small group of users (up
to about 15), different users footstep profiles
are dissimilar enough for our system to work
satisfactorily;
Create a system that can track a user over an
area larger than just a single floor tile;
Use the system in a real environment with
real users and real applications.
In this paper, we will describe the progress we have made
towards the first three of these four goals, and our plans
for the fourth goal.
Technology Tradeoffs
It is important to note at the outset that we do not intend
the Smart Floor to be a single technology replacement for
the other types of identification technologies. If that were
our goal, we would have aimed to design a system that
gives perfect recognition and is transparent to use (i.e., the
user need not expend any additional effort for the
technology to do its job). Rather, we intend that the Smart
Floor will work in conjunction with other technologies.
For example, the floor system may provide a set of
weighted identities to a voting system that has inputs from
other systems such as voice and face recognition systems.
Further, a video recognition system may overcome some of
the shortcomings of the floor system (at least in its current
form), such as the inability to distinguish between people
who have very similar walking profiles but who may be of
widely different heights.
No identification system is perfect. A technology may give
very accurate (or even perfect) results, but these
technologies are not usually transparent. Transparent
systems usually do not give 100% accurate results. Our
floor system falls into this