International Journal of Information Systems for Logistics and Management Vol. 3, No. 2 (2008) 115-120
http://www.knu.edu.tw/ijislm
Implementation of the Wireless Activity Monitoring System Using Accelerometer and Fuzzy Classifier Do-Un Jeong*, Kyeong-Hoon Do, Wan-Young Chung Division of Computer and Information Engineering, Dongseo University Sasang-ku, Busan, 617-716, Republic of Korea Received 25-27 February 2008; Select for review from ILM 2008; Accepted 16 May 2008
ABSTRACT The rapid development of information communication technologies has changed people’s perception on public health and medicine and this is, in turn, increasing demand for improved medical services. The real-time monitoring about the activity of the human provides useful information about the activity quantity and ability. The present study implemented a small-size and low-power acceleration monitoring system for convenient monitoring of activity volume and recognition of emergent situations such as falling during daily life. For the wireless transmission of acceleration sensor signal, we developed a wireless transmission system based on a sensor network. In addition, we developed a program for storing and monitoring wirelessly transmitted signals on PC in real-time. The performance of the implemented system was evaluated by assessing the output characteristic of the system according to the change of posture, and parameters and a context recognition algorithm were developed in order to monitor activity volume during daily life and to recognize emergent situations such as falling. In particular, recognition error in the sudden change of acceleration was minimized by the application of a falling correction algorithm. Keywords: activity, accelerometer, classifier, fuzzy, sensor network.
1. INTRODUCTION In response to the demand, new paradigms of public health and medicine are being proposed in combination with information communication technologies. Such paradigms are changing medical services from treatmentoriented to prevention-oriented and from disease control to health management (Park & Jayaraman, 2003). Another important factor of the change in medical paradigm is the expansion of the aged population who are most closely related to the development of relevant technologies. Korea has already become an aged society, in which the percentage of elders aged over 65 was 7.2% in 2000 and increased to 8.4% in 2004. The percentage *Corresponding author:
[email protected]
is expected to reach 14.4% in 2019 and 20.0% in 2026, turning Korea into a super-aged society. With the aging of the population, demand for medical services and medical expenses are growing and as a consequence it is required to enhance the productivity and efficiency of the current medical service provision system. Recently with the rapid advance of IT and the miniaturization and integration of diagnosis and health management systems, remote home-based medical systems are being developed actively through the combination of portable home medical equipment and communication networks. Furthermore, attention is focused on ubiquitous healthcare technology, with which people can monitor their health information and perform health management at any time
International Journal of the Information Systems for Logistics and Management (IJISLM), Vol. 3, No. 2 (2008)
2. IMPLEMENTATION AND METHODS 2.1 Sensor System In this study, acceleration of axis X, Y and Z according to physical activity was measured in order to monitor posture change and activity volume during daily life. For this, we used a MMA7260Q (Freescale Co. Ltd., USA) acceleration sensor that can measure 3axis acceleration on a single chip. The acceleration sensor operates at a low voltage and low power. As its current consumption is 3 µA under slip-mode operation, the sensor is applicable to various types of sensor networks where low-power operation is essential. What is more, the acceleration sensor used in this study can be applied extensively because the output sensitivity can
Sensor Board
Y
and in any place. Aged people are prone to functional disorders of the body, diseases and injuries. For example, old people can have syncope when they rise from their bed due to the attenuation of the blood pressure control function. For the aged with diseases such as hypertension, myocardial infarction and cerebral apoplexy, the possibility of falling is very high because their body functions may become suddenly uncontrollable. In such cases, if a proper measure is not taken immediately, their life may fall in great danger. Researches have been made so far on how to distinguish activities in daily life using various sensors and discrimination algorithms (Bouten et al., 1997, Mathie et al., 2004, Aminian et al., 1999, Winters et al., 2004, Dean et al., 2006). They use acceleration measuring sensors to monitor falling or activity volume of elders who spend most of their time indoors, and apply signal processing techniques such as wavelet conversion, fuzzy clustering and neural network processing in order to extract parameters useful in the classification of data obtained from sensors (Lee et al., 2004, Mathie et al., 2004, Najafi et al., 2003, Mathie et al., 2004). The present study implemented a small-size and super-low-power acceleration monitoring system for convenient monitoring of activity volume and recognition of emergent situations such as falling during daily life. For the wireless transmission of acceleration sensor signal, we built a wireless transmission system based on a sensor network. In addition, we developed a program for storing and monitoring wirelessly transmitted signals on PC in real-time. Using the implemented system, we measured change in acceleration signal according to the change of posture and activity pattern. In addition, from measured signal were extracted parameters for the classification of posture changes and activity patterns, and using the parameters was developed a classification system that can distinguish posture changes, activity patterns and situations such as falling.
X
Z
116
3-Axis Accelerometer
X Y Z
Filter & Buffer
Fig. 1. Schematic diagram of the implemented sensor system
Table 1. Design specifications of the sensor board 35 × 34 mm 1.5 g/2 g/4 g/6 g (800 mV, 1.5 g) Active 500 µA Current Consumption Sleep 3 µA Operation Voltage 2.2 V ~ 3.6 V Board Size Sensitivity
be adjusted to 4 levels, and it contains a filter circuit inside for convenient sensor interface. In order to detect and process acceleration sensor signal, we designed and implemented an acceleration measuring sensor board. First, 3V power, which is supplied from the battery of a wireless sensor node, was used to operate the acceleration sensor, and an external terminal was built separately for adjusting the sensitivity of sensor output. In addition, we added a filter circuit to each output terminal of the sensor to remove noises caused by filter drive frequency inside the sensor, which may be contained in output signal. Moreover, a buffer circuit was designed using OP-Amp in order to prevent impedance mismatch that may happen in the interface of output signal. Fig. 1 and Table 1, respectively, show the schematic diagram of the implemented sensor system and its design specifications. 2.2 Sensor Node and Monitoring For the wireless transmission of measured data from the acceleration sensor, we used TIP710CM (Maxfor Co. Ltd., Korea), a Zigbee-compatible wireless sensor node based on IEEE802.15.4 sensor network technology. The sensor node was designed based on Telos Platform of Moteiv, and is controlled by MSP430F1611, a low-power microprocessor manufactured by TI. MSP430F1611 processor is a microprocessor of 16 bit RISC structure operable at a low voltage of around 1.8 V, containing 48 KB program memory and 10 KB static memory inside. In addition, it can be used in various applications using the external 1 MB flash memory. In this study, analog signal from the acceleration sensor was converted into digital signal through data sampling at a rate of 100 times/
117
D.U. Jeong et al.: Implementation of the Wireless Activity Monitoring System Using Accelerometer and Fuzzy Classifier
second at each channel using channel No. 0, 1 and 2 of an A/D converter, which is embedded inside MSP430F1611 and has 10-bit resolution. The TIP710CM sensor node used Zigbee-compatible 2.4G bandwidth for wireless communication, and CC2420 (Chipcon Co. Ltd., Norway) with a simple interface circuit around the chip. In addition, a ceramic chip antenna was applied for stable wireless transmission and reception. The sensor node adopted in this study can be used in various applications as it supports removable interface boards using USB or RS-232 communication for interface with PC. The OS of the wireless sensor node applied in this study is TinyOS, which is component-based embedded OS developed for the Smart Dust Project by UC Berkeley. The OS was designed to minimize code size for small memory consumption and low-power driving required in sensor network applications. In addition, the OS system was built with nesC of component structure that can be applied easily to new technologies. nesC declares and links several important components implemented, and one of them has ‘Provides’ and ‘Uses’ interface. These interfaces play the role of component designators and have bidirectional characteristic. In addition, nesC is a language optimized for embedded systems such as sensor networks, supporting TinyOS models such as structuring, naming and linking. In the present study, we designed the AxisSensorC component of TinyOS. The component is composed of embedded OscopeC and TimerC, and converts the analog signal of axis X, Y and Z into digital signal through sampling at a rate of 100 times/second. In addition, it was designed for wireless transmission through the GenericComm component provided. We implemented a monitoring program for receiving signal measured through the sensor board and displaying, storing and analyzing the signal on the PC. The data format for transmission between the sensor node and the monitoring program is 36 byte long, 10 bytes for head information and 26 bytes for message information. First, in head information, Length is the length of message information, Addr. is the address of the destination, and Group is the group ID. In addition, message information is composed of node ID, sample size, channel and 20 bytes’ data to be transmitted. Fig. 2 shows the format of data transmitted between the sensor node and the monitoring program. Data transmitted from the sensor node is received by the monitoring program. Information on the starting point of the data packet is extracted from the received data and data is formed with information in the packet unit. In addition, information on packet channel is extracted, and 20 bytes’ sensor data starting from the channel information is obtained using the string selection function. The obtained 20 bytes’ data include 10 samples, and lower bytes come first and then upper
Length FCF. 1 Byte
2 Byte
DSN Destpan.
1 Byte
2 Byte
Type Addr.
Group
2 Byte 1 Byte 1 Byte
Head 10 Byte Source Mote ID Last Sample Number Channel 2 Byte 2 Byte 2 Byte
Data
20 Byte
Message 26 Byte Fig. 2. Data format for transmission between the sensor node and the monitoring program
bytes. Thus, the complete dataset is formed by switching the position of upper bytes and lower bytes, and then is displayed and analyzed. 3. CLASSIFICATION ALGORITHM In order to classify postures and activities from measured 3-axisacceleration information, SVM (signal vector magnitude) was calculated, and then DSVM (differential signal vector magnitude) was calculated by Equation (2) through differentiating SVM and calculating the mean of the absolute values (Mathie et al., 2004).
SVM =
x 2i + y 2i + z 2i
DSVM =
1 t
t
Σ0
SVM ′ dt
(1)
(2)
Using 3-axis acceleration signal and parameters, we developed an algorithm that can classify the state of activity automatically. First, if DSVM was higher than the base level (2.5) the subject was considered to be in activity, and if it was lower he was considered to be at a pause. In case the subject was in an active state activities such as walking and running were distinguished by DSVM and fuzzy classifier, and in case the subjects was in a static state the change of posture was discriminated by calculating the means of axis X, Y and Z for 0.5 second. Fig. 3 shows the flow chart of the implemented recognition algorithm. 4. EXPERIMENT AND RESULTS 4.1 Experiment Set The present study implemented a system for monitoring activity volume and recognizing emergent situations by sensing acceleration during activities in daily life and distinguishing postures and activities. In order to
118
International Journal of the Information Systems for Logistics and Management (IJISLM), Vol. 3, No. 2 (2008)
Ghz
Signal
Wirel ess 2. 4
Calculate SVM Calculate DSVM DSVM > 2.5
Sensor + Node
N (Rest) Y (Activity)
Calculate X, Y, Z-axis data
Base Station Xaxis data < 0.8 Y
Fig. 5. Implemented experiment set
N Standing & Siting
Compare Y, Z-axis data
Yaxis data > 0.5
Zaxis data > 0.5 N
N
Y Lying
Monitoring Program
Y Lying Font
Lying Right
Lying Left
Fig. 3. Flow chart of the recognition algorithm
Fig. 6. The result of posture change classification
4.2 Posture and Activity Classification Fig. 4. Implemented sensor board
detect acceleration signal, we designed and implemented an acceleration sensor measuring board embedded with a 3-axis acceleration sensor, a filter and a buffer circuit. Fig. 4 is the PCB photograph of the implemented sensor board fixed on the wireless sensor node. For wireless monitoring of acceleration information measured by the sensor board, we built up a base station using the sensor node with the sensor board fixed on it and another sensor node, and prepared an experiment set for wireless monitoring from a laptop computer. Fig. 5 shows the structure of the experiment set.
In order to evaluate the possibility of monitoring not only static postures but also activities during daily life, we developed an algorithm that assesses the output characteristic of 3-axis acceleration signal according to various activity patterns and classifies the state of activity based on the characteristic. First, we artificially made static states such as standing (St), sitting (Si), lying (Ly), lying-left (LyL), lying-front (LyF) and lying-right (LyR) and measured the change of acceleration in each situation. Fig. 6 shows the result of posture change classification in static stage. Second, we made dynamic states such as slow walking (SW), fast working (FW), slow running (SR) and fast
D.U. Jeong et al.: Implementation of the Wireless Activity Monitoring System Using Accelerometer and Fuzzy Classifier
Table 2. Statistic value of the DSVM for membership function design S.W
F.W
S.R
F.R
DSVM (xyz)
Average Std.
2.6 0.5
5.1 1.7
15.4 1.7
26.9 3.3
DSVM (y)
Average Std.
2.7 0.5
4.8 2.2
17.4 1.9
29.5 3.2
A1
A2
119
A5
A3
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
B1
B2
B3
B4
Table 3. Result of the activity classification
M #1 M #2 M #3
S.W
F.W
S.R
F.R
Avg.
85% 100% 100%
100% 100% 100%
92.5% 95.0% 92.5%
100% 80% 100%
94.4% 93.7% 97.5%
running (FR), and measured the change of acceleration using running machine in each situation. And we calculated the standard deviation and average value of the DSVM for the design of the membership function. Table 2 shows the statistic value of the DSVM. We used a three kind of a membership function in this research. The membership function which the first method is based on the statistics of the DSVM and the second method use median value of the DSVM and the third method is hybrid of a first and second method. Fig. 7 shows the output of various membership functions which applies in this research. And Table 3 shows the results of the activity classification using respectively fuzzy membership function. 5. CONCLUSION The present study implemented a system and a classification algorithm for recognizing the state of activity and emergent situations such as falling during daily life. For this, we built up a small-size, low-power wireless acceleration measuring system distinguished from previous researches, and implemented a program that can monitor signal in real-time on the PC. The performance of the implemented system was evaluated by assessing the output characteristic of the system according to the change of posture, and parameters and a context recognition algorithm were developed in order to monitor activity volume during daily life and to recognize emergent situations such as falling. In particular, recognition error in the sudden change of acceleration was minimized by the application of a falling correction algorithm. Based on the results of this study, we plan to develop algorithms for more accurate recognition of posture changes and activity patterns and to design an application system that links emergent situation recognition to an emergency center so that emergency measures can be taken promptly.
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
(a) Membership function output using statistic method (M #1)
A1
A2
A3
A5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
B1
B2
B3
B4
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
(b) Membership function output using median value (M #2)
A1
A2
A3
A5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
B1
B2
B3
B4
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
(c) Membership function output using hybrid method (M #3) Fig. 7. Output of the various membership functions
120
International Journal of the Information Systems for Logistics and Management (IJISLM), Vol. 3, No. 2 (2008)
ACKNOWLEDGMENTS This research was financially supported by the Ministry of Commerce, Industry and Energy (MOCIE) and Korea Industrial Technology Foundation (KOTEF) through the Human Resource Training Project for Regional Innovation REFERENCES Aminian, K., Robert, P., Buchser, E. E., Rutschmann, B., Hayoz, D. and Depairon, M. (1999) Physical activity monitoring based on Accelerometry: Validation and comparison with video observation. Med. Biol. Eng. Comput., 37(1), 304-308. Bouten, C.V., Koekkoek, K. T., Verduin, M., Kodde, R. and Janssen, J. D. (1997) A triaxial accelerometer and portable data processing unit for the assessment of daily physical activity. IEEE Trans. Biomed. Eng., 44(3) 136147. Karantonis, Dean M., Narayanan, Michael R., Mathie, Merryn, Lovell, Nigel H. and Celler, Branko G. (2006) Implementation of a real-time human movement classifier using a triaxial accelerometer for ambulatory monitoring. IEEE Trans. Information Technology in Biomedicine,
10(1), 156-167. Lee, S. H., Lim, Y. T. and Lee, K. J. (2004) A disign of algorithm for analysis acitive using 3-axis accelerometer. J. of KIEE, 53(5), 361-367. Mathie, M. J., Coster, A. C. F., Celler, B. G. and Lovell, N. H. (2004) Classification of basic daily movements using a triaxial accelerometer. Med. Biol. Eng. Comput., 42(5), 670-687. Mathie, M. J., Coster, A. C. F., Lovell, N. H. and Celler, B. G. (2004) A pilot study of long term monitoring of human movements in the home using accelerometry. J. Telemed. Telecare, 10(3), 144-151. Mathie, M. J., Lovell, N. H., Coster, A. C. F. and Celler, B. G. (2002) Determining activity using a triaxial accelerometer. in Proc 2nd joint EMBS-BMES Conf., Houston TX. Najafi, B., Aminian, K., Paraschiv-Ionescu, A., Loew, F., Büla, C. J. and Robert, P. (2003) Ambulatory system for human motion analysis using a kinematic sensor: monitoring of daily physical activity in the elderly. IEEE Trans. Biomed. Eng., 50(6), 711-723. Park, S. and Jayaraman, S. (2003) Enhancing the quality of life through wearable technology. IEEE Eng. Med. Biol. Mag., 22(3), 41-48. Winters, J. M., Wang Y. and Winters, J. M. (2003) Wearable sensors and telerehabilitation. IEEE Eng. Med. Biol. Mag., 22(3), 56-65.
