GAIT
IDENTIFICATION AND RECOGNITION SENSOR=
HInstitute for Automatics,
Swiss Federal Institute of Technology
HH
SUMMARY
One of the major obstacles in developing reliable
walking neural prostheses is poor performance of the sensors which are used for
gait phases identification. Improper functioning of these sensors causes wrong
stimulation pattern selection and wrong stimulation sequencing of the walking
neural prostheses. These malfunctions often cause unstable walking patterns in
patients that are using the prostheses. Sensors that are commonly used for gait
phase identification are: foot switches, force sensitive resistors (FSRs),
accelerometers, pendulum resistors and goniometers. Since none of these sensors
is capable of identifying gait phases with accuracy greater than 95 %, a
decision was made to develop more reliable gait identification sensor.
A new gate identification sensor, which consisted of
three FSRs, an inclinometer and a rule-based observer, has been proposed. Every
50 ms from the FSRs and the inclinometer readings the proposed sensor
identified one of the following gait phases: heel off, swing phase, heel strike
and mid stance. The experiments conducted with able-bodied and disable subjects
showed that the proposed sensory system detected the above gait phases with
reliability greater than 99 %. This foot sensor was capable of distinguishing
walking sequences from weight shifting during standing, and it did not give
false gait annunciation when the instrumented foot was sliding during standing.
Our future research is aimed at further improving
the foot sensor packaging, and sensor’s robustness to different environmental
conditions and shocks.
STATE OF
THE ART
In order to design a walking neural prosthesis,
which can automatically detect gate phases and accordingly select stimulation
sequences, one has to have a reliable gate recognition sensor. One of the first
foot sensors proposed was a heel switch [1] which was used to detect the heel strike during
normal gate. Advanced walking neural prostheses require information about other
gate phases in addition to heel strike. Hence, the heel switch is not an
appropriate sensor for this application. Second approach suggests that at least
3 FSRs, placed in a shoe sole, can be used to detect the most important gate
phases [2, 3]. Experiments conducted in our laboratory clearly
showed that the FSRs alone cannot reliably detect gate phases. This system has
a number of problems, out of which identifying weight shifting during standing
as walking pattern is probably the most severe one. In order to resolve this
problem some researchers proposed using goniometers, in conjunction with FSRs,
which are attached to hip, knee or ankle joints [4, 5]. Experiments conducted in our laboratory demonstrated
that goniometers and FSRs together do not provide reliable gate identification.
In particular, this sensor configuration generated wrong gait identification
when subjects were making short brakes or rests during walking. Since none of
the existing foot sensors is capable of identifying gait phases with sufficient
accuracy, a decision was made to develop more reliable foot sensor.
|
|
Figure 1: Position of the
FSR’s in the shoe sole
MATERIALS
AND METHODS
The proposed foot sensor consisted of three FSRs, an
inclinometer and a rule-based observer. One 174NN and two 152NS FSRs,
manufactured by Interlink Electronics Inc. [6], were used to measure forces generated by subject’s
heel and metatarsal bones during walking. The FSRs were placed in the shoe sole
as indicated in Figure 1. Time response of the FSRs was 2 msec.
|
|
Figure 2: Position of the
inclinometer
In house developed inclinometer was used to measure
relative position of the heel with respect to the walking surface. The
inclinometer consisted of a gyro sensor ENC-05A, manufactured by Murata [7] (see Figure 2), and an integrator which calculated
the position of the heel from the raw gyro data. The gyro was attached to the
shoe heel and its sensory axis was parallel to the walking surface (see Figure
2). The time response of the inclinometer (combined time response of the gyro
and the integrator) was 30 msec.
The rule-based observer was designed to identify mid
stance, heel off, swing phase and heal strike gate phases, and was implemented
using Hitatchi SH7032 evaluation board. The proposed observer functioned as
follows. Once the sensor was turned on, a subject instrumented with the sensor
had to stand still for one second before it made the first step. During this
period of time the observer automatically reset FSRs’ and inclinometer readings
and set them to initial values (FSRs = ON and inclinometer angle = 0 deg).
After the reset, the rule-based observer shifted into gate recondition mode
described in Figure 3. Note, except for mid stance, all other gate phases could
be identified only if the previous gate phase was successfully identified. This
feature was introduced in order to prevent false gate phase identification. In
addition the observer’s algorithm was enhanced with an adaptive routine which
compensated for FSRs’ drifts. Time constant of this algorithm was 11 sec.
|
|
Figure 3: Observer’s gate
recognition algorithm (INCL. represents the inclinometer angle in [deg] and
P.S. represents the previous state of the sensor)
RESULTS
Preliminary experiments performed with 10
able-bodied subjects and 10 disabled subjects showed that the proposed foot
sensor can identify mid stance, heel off, swing phase and heal strike with
reliability greater than 99 %. The proposed foot sensor was capable of
distinguishing walking sequences from weight shifting during standing, and it
did not give false gait annunciation when the instrumented foot was sliding
during standing. It is important to mention that subjects that were trained to
use the sensor achieved better results then the subjects which used the sensor
for the first time. Representative experimental results obtained with the
proposed sensor are given in Figure 4.
Figure 4: Gait pattern
recognition - intermittent walking of an able body subject
REFERENCES
[1] “Microfes, Unifes, Decus Personal, Decus Hospital, Nervobol
Personal, Nervobol Hospital, ALT-2, Measuring Crutches with a Biological
Feedback, and Stimulator Scolifes,”: Institut 'Jozef Stefan'.
[2] T. L. Lawrence and R. N. Schmidt, “Wireless In-Shoe Force
System,” presented at 19th International Conference of the
Engineering in Medicine and Biology Society/IEEE, Chicago, USA, 1997.
[3] M. M. Skelly and H. J. Chizeck, “Real Time Gait Event
Detection During FES Paraplegic Walking,” presented at 19th
International Conference of the Engineering in Medicine and Biology
Society/IEEE, Chicago, USA, 1997.
[4] A. Kostov, R. B. Stein, D. Popovic, and W. W. Armstrong,
“Improved Methods for Control of FES for Locomotion,” presented at Proc. IFAC
Symposium on Biomedical Modeling, Yaluestone, TX, USA, 1994.
[5] A. Kostov, B. J. Andrews, D. B. Popovic, R. B. Stein, and W.
W. Armstrong, “Machine Learning in Control of Functional Electrical Stimulation
Systems for Locomotion,” IEEE Tr. on
Biomedical Engineering, vol. 42, pp. 541-551, 1995.
[6] FSR Integration Guide
& Evaluation Parts Catalog. Camarillo, USA: Interlink Electronics,
1997.
[7] Gyrostar Family from Murata: Murata
Manufacturing Co. Ltd., 1997.
AUTHOR’S
ADDRESS
Dr. Milos R.
Popovic
Institute for Automatics, Swiss Federal Institute of
Technology Zürich
ETH Zürich /ETL K22.1, CH-8092 Zürich, Switzerland
tel: +41-1-632-3638, fax: +41-1-632-1211, E-mail:
popovic@aut.ee.ethz.ch