Authors：Che-Chang Yang, Yeh-Liang Hsu
(2006-08-22); Recommend: Yeh-Liang Hsu
Note: This paper is presented at the International Computer Symposium
2006, Dec. 4-6, 2006, Taipei,
Development of a Portable System
for Physical Activity Assessment in a Home Environment
assessment of daily physical activity in a home environment provides
significant information in evaluation of health and the quality of life of
subjects with limited mobility and chronic diseases. This study developed a
system for human physical activity assessment in ambulatory monitoring using
only one portable sensing device combining a tri-axial accelerometer and its
distributed data processing platform. This real-time system is able to identify
several postures, posture transitions and movements with the embedded
algorithm. In addition, this system also features fall detection capability.
The results of the test for evaluating the performance of the system show high
identification accuracy for both still postures and dynamic movements. A
long-term ambulatory test was also conducted, and the recorded data shows
sufficient information of the subject’s activities of daily living at home.
Some inherent limitations concerning real-time identification were discussed.
Despite those limitations, this system is technically viable for ambulatory
application to provide sufficient information in evaluating a person’s activity
of daily living (ADL) and his status of physical mobility. Potential wok of
this system in the future is also discussed.
activity assessment, tri-axial accelerometer, fall detection, ambulatory
monitoring, postural transition
activity can be regarded as any movement or posture that is produced by
skeletal muscles and results in energy expenditure [Capspersen et al., 1985]. Various health conditions
such as heart disease, senile dementia, degeneration in mobility will directly
affect one’s physical activity level. The quantitative assessment of daily
physical activity at home is a determinant in the evaluation of health and the
quality of life of subjects with limited mobility and chronic diseases, such as
of physical activity is difficult due to the subtle and complex nature of body
movement which requires precise and reliable measuring techniques. Standard
human motion capturing techniques based on optical, magnetic and ultrasonic
systems allow a complete kinematic analysis but require a dedicated laboratory
[Dickstein et al., 1996; Kemp et al.,
1998; Aminian et al., 2004]. From a
technological point of view, these techniques mentioned above can precisely
capture human motions and have been used in the applications of computer
animation and virtual reality. However, for the assessment of daily physical
activity in a home environment the cost of such techniques is unacceptable for
common use. Moreover, the subjects must be restrained inside a laboratory-like
space, which is entirely different from a free-living home environment.
Therefore, acquisition of body movement using portable measuring devices or
body-fixed motion sensors is an appropriate alternative for physical activity
assessment in ambulation and home environments.
In the past,
mechanical motion sensors have been used for various physical activity
assessments. Saris et al.  used
pedometers and actometers to study daily physical activity. Energy expenditure
related to various physical activities has also been estimated using this
technology [Verschuur et al., 1980].
Meanwhile, the use of accelerometers to measure body movement began as early as
the 1970s. Morris  developed an accelerometry-based technique to measure
human body movement. Gerwin et al. 
proposed a method to assess physical activity with motion sensors and
accelerometers. They also developed a small data acquisition unit with a solid
state memory in place of a large tape recorder which discouraged subjects from
wearing it. Veltink et al. 
investigated the feasibility of distinguishing several static and dynamic
activities in a domestic environment using a small set of two or three uniaxial
accelerometers. Bouten et al. 
also presented a tri-axial accelerometer which consists of three separate
orthogonal uniaxial accelerometers to measure body movement. Najafi et al. developed a portable data
processing unit for the off-line acquisition, processing, and storage of the
acceleration data. An ambulatory system was presented for daily physical activity
monitoring of the elderly using a kinematic sensor which consists of a
miniature gyroscope and two dual-axial accelerometers. In addition, fall risk
evaluation was also proposed [Najafi et
The purpose of
this research is to develop a wearable system for physical activity assessment
based upon the Portable Tele-homecare Monitoring System (PTMS) infrastructure,
which is a decentralized tele-healthcare application system proprietarily
developed by Gerontechnology Research Center (GRC) at Yuan Ze University. As
for the body-worn device, a miniature tri-axial accelerometer is used for
motion detection of body movement. A dedicated body movement algorithm embedded
in the microcontroller in that device is developed to actively recognize three
still postures (sitting, standing, and lying), four postural transitions
(sit-stand transitions and lie-sit transitions) and locomotion (walking) in a
home environment. This system also features the fall detection capability to
recognize any possible fall event. Long-term monitoring in a subject’s
residence was also demonstrated to register the majority of daily activities.
Method and system design
Figure 1 shows
the structure of the system. A wearable data acquisition unit (DAU) primarily
consisting of a miniature tri-axial accelerometer (KXM52-1050, Kionix, Inc.), a PIC microcontroller
(PIC18F6722, Microchip) and a RF wireless transmitter
module (PT2262, Princeton Tech.;
TWS-CS-2, Wenshing Electronics Co., LTD)
is designed for acceleration measurement and real-time identification of human
postures and movements (Figure 2 (left)). As shown in Figure 3, the DAU is
designed to be carried at the waist level nearby close to the center of gravity
of the body by means of clips onto the pant belt for easier and convenient use.
From the empirical trial, the suitable position which lies within the range of
45 degrees from the frontal (antero-posterior) side to either of medio-lateral
sides is consistent with the statement in the related studies [Gerwin et al.,
1991; Bouten et al., 1997 ; Sekine et al., 2000; Mathie et al., 2003, 2004; Ohtaki et al., 2005;
Karantonis et al., 2006;]
This low-g (±2g) DC-responsive tri-axial accelerometer
measures the acceleration produced by human movement as well as the constant
gravity component. The accelerometer outputs are low-pass filtered at fc=50Hz and continuously
sampled at 60Hz via 10-bit A/D conversion of the PIC microcontroller.
Computation of a simple kinematic model is used to remove such a gravitational
component instead of using physical high-pass filters. Each real-time
identified activity event is transmitted cyclically via the 433.92MHz RF
wireless transmitter module then to the distributed data server (household DDS)
which features the capabilities of signal computing, I/O control, Ethernet
communication, wireless RF data reception, data storage (MMC) and the links
with external devices. Figure 2(right) shows the household DDS.
Figure 1. System structure
Figure 2. The wearable DAU (left) and household
Figure 3. The attachment of the wearable DAU
2.2 Algorithm design
algorithm embedded in the microcontroller of the DAU is developed to identify
three postures (lying still, sitting still, and standing still), four postural
transitions (sit-stand, stand-to-sit, sit-to-lie and lie-to-sit) and walking
movement. Detection for possible falls is also designed in this algorithm. Figure
4 describes the process flow of the algorithm which mainly includes five parts:
data sampling (Cx), pre-processing (Px), dynamic posture transition
identification (DBx), still posture identification (DAx) and possible fall
detection (DCx). All signals are processed in time-domain analysis in the
algorithm due to the limited computation capability of the PIC microcontroller
as well as the fact that batch data analysis in frequency-domain method yields
inaccurate results. All the identified results are stored in the variable STATE. In the case where there is no
definite result determined throughout the processes, the event will be recorded
as an “Uncertain movement” or an “Uncertain posture”.
sampling process consists of the primary stage (C1) and the secondary stage (C2)
in 0.5s and 2.0s, respectively. The use of the dual-stage data sampling strategy
ensures that the data of one event can be acquired within the same sampling
interval. Initially, Sections C1 and
D1 determine whether any sign of
dynamic movement exists. If no dynamic movement be detected (D1=No), the sampled 0.5s data is used
to identify one of the three possible still postures in the processes DAx. If
dynamic movement is detected (D1=Yes),
the secondary data sampling stage (C2)
is immediately activated to collect the subsequent 2.0s data. The 2.5s data
collected in both stages is combined and then median-filtered (window length n=3) and simplified (reduced to
one-third of the original data amount by averaging method) to represent an
“event” for the following step-by-step movement identification which is to
identify either one of the four postural transitions, walking or a fall event.
Figure 4. The flowchart of the algorithm
mapping” technique which registers whether there are apparent fluctuations or
changes in the data series is commonly applied in computation of the trunk tile
angles and the tri-axial acceleration data in the algorithm. Various thresholds
are given in most of the computation sections to yield the results of specific
determination. Figure 5 shows an example of a binary series (0 or 1 expressed
as the bar chart) which registers the apparent changes of a trunk tilt data
series (degrees in the form of curve) during a sit-to-stand postural transition
mapped by this technique.
Figure 5. Example of the effect of slope mapping
In order to
investigate the accelerometric characteristics of sit-stand transitions, a test
was performed on 15 ostensibly healthy subjects in various ages arranged into
three groups: Young (20-35yrs), middle-aged (35-50yrs) and elderly (50+yrs),
with 5 subjects in each group. From the observation of the test, the vertical
acceleration pattern of sit-stand postural transitions can be characterized by
three particular rules: (i) peak order, (ii) peak distance (time interval) and
(iii) peak values. Either of sit-stand postural transitions can only be
identified when the criterion of all the three rules are satisfied.
postural transitions can be identified and further distinguished from each
other by investigating both the trunk tilt variation and the final posture
orientation. Still posture identification requires the information of
previously known postural transitions or walking movement. A still posture can
be recognized as a lying still posture according to the posture orientation or
if there exists a previous sit-to-lie postural transition. Similarly, sitting
still or standing still postures can be identified by the existence of the
types of previous sit-stand transitions or walking movement.
In order to
evaluate the performance of the algorithm in still posture and dynamic activity
identification, 10 subjects were recruited for the laboratory-based test.
Sensitivity and specificity tests for posture (lying still) and posture
transitions (sit-stand transitions, lie-sit transitions) and walking movement
were conducted. Note that the evaluation did not include the sitting still or
standing still postures due to the fact that both still postures are associated
with the results of previously identified postural transitions or movement. In
addition, Falling was not included either because it was not easy for the
testers to simulate “standardized” falls. Table 1 shows the evaluation results
of sensitivity and specificity from 200 and 500 samples, respectively.
Table 1. Performance of the algorithm
in daily physical activity is important in the elderly care and rehabilitation.
Intuitively, a fall can be regarded as a movement accompanying by unusual high
acceleration peaks in a very short time interval. The measures of summation of
time integrals of the accelerometer outputs (IMA) or signal magnitude vector
(SVM) were proposed to evaluate the intensity of the physical activities [Bouten
et al., 1997; Karantonis et al., 2006]. According to the
definition given by Karantonis et al.
, falls are said to have occurred if at least two consecutive peaks in
the SVM above a defined threshold 1.8g
are recorded and followed by a 60s post-fall period of no activity. [Karantonis
et al., 2006] As for the algorithm in
this system, a “sign of fall” can be identified for a non-upright posture if
there are at lest two peaks at relatively higher magnitude of ±1.0g either in vertical or antero-posterior
acceleration component. A “possible fall” event can be further identified from
a previously registered “sign-of-fall” event followed by a 20s period of lying
still posture without any activity.
System integration and test
Ethernet communication capability of the household DDS, it is accessible via
the Internet. The monitoring of real-time activity status and the accumulation
sum of each monitored item can be displayed by using the Internet browser on
the client PC. This interface provides brief information of real-time acquired
event data for the users without any dedicated software.
management can be achieved by using dedicated PC-based VB-developed programs as
Figure 6 which features the following fundamental functions:
Remote data access capabilities: The
authorized users (e.g., the system administrator, care-giver or families) are
allowed to access to the DDS and retrieve the data stored in it through TCP
Real-time monitoring information: The
real-time monitoring information (still posture or dynamic activity) is
displayed on this interface.
Recorded data display: The daily
recorded data can be chronologically displayed to show the overall activity
distribution. As shown in the Figure 7 rests and activities can be
distinguished. They are also classified and given quantitative results, such as
the numbers, percentages of those events in the data. All the related
information can be saved to an Excel file (*.xls).
Event-driven function: The DDS can be
optionally equipped with a GSM module to provide event-driven capability. When
this function is enabled by the user, the DDS is able to actively send a cell
phone text message to a user-specified person when a possible fall has been
Figure 6. VB-developed program interface
Figure 7. Display of the statistical results of
Figure 7 also
shows an example of a subject’s long-term monitoring data in a day at home. In
this test, the system began to record data at about 1:00 and ended at about
16:00. In the home environment, the user (subject) is not expected to use this
system continuously throughout the monitoring period during some situations
such as taking a shower, going outside, etc. Therefore, the data recorded may
not be continuous, and the actual monitoring time in this test is about 405
minutes. That is, about 6.75 hours within the 15-hour period is recorded.
According to the
statistical results shown in Figure 7, more than 90% of the recorded events are
still postures. Moreover, lying still postures occupy about 61% of total
recorded events. Figure 8 is the activity chronograph which chronologically
displays the recorded events represented as a series of event number 1 to 9. It
can be observed that the subject was mostly in the “lying still” (event number
1) posture from 4:00 to 9:00, which indicates that the subject was probably
sleeping during that period. In addition, the subject had large numbers of
“sitting still” (event number 2) postures from about 9:00 to 13:00. After
14:00, the subject performed many posture transitions and walking movements
(event number 4 to 8).
Figure 8. Activity chronograph of the recorded
activity assessment system which utilizes only one wearable sensing device has
been developed and demonstrated in ambulatory tests. This system also achieves
good performance in still posture and dynamic activity identification. However,
some inherent limitations are worth discussing here.
capacitance tri-axial accelerometer was used in this study to measure the
acceleration and trunk tilt of the human body. In fact, the most precise tilt
sensing can be maintained when the accelerometer is at static, or under
constant acceleration. Tilt sensing using accelerometers still has limited
accuracy of tilt sensing in changing acceleration magnitude. It was also
reported by Elber’s research on gravitational artifact in accelerometric
measurement [Elber, 2005]. However, In spite of this constraint, tilt sensing
and acceleration measurement using one tri-axial accelerometer is still valid
for physical activity because the resulting outputs still preserve apparent
characteristics for either trunk tilt and acceleration patterns.
The wearable DAU
has been designed to measure human body movement at waist level and clipped to
the belt for minimizing discomfort and inconvenience in use. However, carrying
the DAU might limit posture and movement when lying down and therefore further
influences the subject’s comfort. As for wireless data transmission, power
capacity has a significant influence on the effective distance and stability of
data delivery. To extend the time for use, a battery cartridge with (3×AA
alkaline batteries) can be used instead of the AAA batteries. The antenna
configuration also has a great influence on the performance of wireless data
delivery. However, the antenna design has not been further evaluated. In the
future, the onboard antenna and optimization must be taken into consideration.
in computation capability and memory capacity of the microcontroller used in
this study, coupled with the fact that human events must be identified
simultaneously to keep up with the next data acquisition process, limit the
identification performance. Most other off-line systems use powerful PC-based
computation software such as MATLAB to analyze the recorded data. Therefore,
identification accuracy of those systems is usually higher than that of the
real-time systems [Najafi et al.,
durations of posture transitions or movement are not the same each time, even
for the same person. People usually perform mixed and combined movements in
their normal activities of living. Due to the complex nature of human movement
and limitations in instrumentation, identification accuracy for such a
real-time system can be limited when applied in real ambulatory and home uses,
despite that fact that it achieves good performance for laboratory-set tests.
In this study, a
real-time system for human physical activity assessment using only one portable
sensing device was developed for real-time ambulatory monitoring in a home
environment. This system is able to distinguish rests from activities and
further identify several target posture transitions and movements. Although the
nature of actual human postures and activities of daily living are more complex
than what is considered and assumed in the algorithm, this algorithm still
exhibits acceptable performance in determining those target postures and
activities. Despite some limitations in the configuration for real-time data
processing, this system is technically viable to perform long-term ambulatory
monitoring in a home environment and to provide sufficient information in
evaluating a person’s activities of daily living (ADLs) and his status of
The results from
the ambulatory tests also show that this system can provide significant
information on the subject’s activities of daily living. In the future, the
application field of this system, system robustness and reliability and the
possibility for ubiquitous computing which integrates all the ADL-related data
altogether should be further considered.
Najafi, B., 2004. “Capturing human motion using body-fixed sensors: outdoor
measurement and clinical applications,” Comp.
Anim. Virtual Worlds, pp. 79-94, John Wiley & Sons.
Bouten, C. V.
C., Koekkoek, K. T. M., Verduin, M., Kodde, R., Janssen, J. D., 1997. “A
Triaxial Accelerometer and Portable Data Processing Unit for the Assessment of
Daily Physical Activity,” IEEE Trans.
Biomed. Eng., vol. 44, pp.136-147.
J., Powell, K. E., Christenson, G. M., 1985. “Physical activity, exercise and
physical fitness: Definitions and distinctions for health related research,” Public Health Rep., vol. 110, pp.
Abulaffio, N., Gelernter, I., Pillar, T.,
1996. “An ultrasonic-operated kinematic measurement system for assessment of
stance balance in the clinic,” Clinical
Biomechanics, vol. 11, pp. 173-175.
Rodger, 2005. “Gravitational artifact in accelerometric measurements of
tremour,” Clinical Neurophysiology,
vol. 116, pp. 1638-1643.
Gerwin A. L.
Meijer, Klaas R. Westerterp, Francois M. H. Verhoeven, Hans B. M. Koper, and
Foppe ten Hoor, 1991, “Methods to Assess Physical Activity with Special
Reference to Motion Sensors and Accelerometers,” IEEE Trans. On Biomed. Eng., vol. 38, No. 3, pp. 221-229.
M., Narayanan, Michael R., Mathie, Merryn., Lovell, Nigel H., Celler, Branko
G., 2006. “Implementation of a Real-Time Human Movement Classifier Using a
Triaxial Accelerometer for Ambulatory Monitoring,” IEEE Trans. Info. Tech Biomed., vol. 10, No. 1, pp. 156-167.
Janssen, A., Kamp. B. V. D., 1998. “Body position can be monitored in 3D using
miniature accelerometers and earth-magnetic field sensors,” Electroencephalography and clinical
Neurophysiology, vol. 109, pp. 484-488.
Mathie, M. J.,
Coster, A. C. F., Lovell, N. H. Celler, B. G., 2003. “Detection of daily
physical activities using a triaxial accelerometer,” Medical & Biological Engineering & Computing, vol. 41, pp.
Mathie, M. J., Celler, B. G., Coster, A. C. F., Lovell, N. H., 2004. “Classification of basic daily movements using a triaxial
accelerometer,” Medical & Biological
Engineering & Computing, vol. 42, pp. 679-687.
Moris, J. R.,
1973. “Accelerometry-a technique for the measurement of human body movements,” J. Biomech., vol. 6, pp. 729-736.
Aminian, K., 2002. “Measurement of Stand-Sit and Sit-Stand Transitions Using a
Miniature Gyroscope and Its Application in Fall Risk Evaluation in the
Elderly,” IEEE Trans. Biomed. Eng., vol. 49, pp. 843-851.
Susumago, M., Suzuki, A., Sagawa, K., Nagatomi, R., and Inooka, H., 2005.
“Automatic classification of ambulatory movements and evaluation of energy
consumptions utilizing accelerometers and a barometer,” Microsystem Tech., vol. 11 pp. 1034-40.
Saris, W. H. M.
and Binkhorst, R. A., 1977. “The use of pedometer and actometer in studying daily
physical activity in man. Part I: reliability of pedometer and actometer,” Eur. J. Appl. Physiol., vol. 37, pp.
Saris, W. H. M.,
Binkhorst, R. A., 1977. “The use of pedometer and actometer in studying daily
physical activity in man. Part II: validity measuring the daily physical
activity,” Eur. J. Appl. Physiol.,
vol. 37, pp. 229-235.
Tamura T., Togawa T., Fukui Y., 2000. “Classification of waist-acceleration
signals in a continuous walking record,” Medical
Eng. and Phy. Vol. 22, pp. 285-291.
Verschuur, R., Kemper, H. C. G., 1980. “Adjustment
of pedometers to make them more valid in assessing running,” Int. J. Sports. Med., vol. 1, pp. 95-97.
Veltink, P. H., Bussmann, H. B. J., Vries, W. D., Martens, W. L. J.,
Lummel, R. C. V., 1996. “Detection of static and
dynamic activities using uniaxial accelerometers,” IEEE Trans. Rehab. Eng., vol. 4, pp. 375-385