1. Introduction

Wearable sensors are continuously emerging over last decade and they are presenting important innovation tool for services associated with healthcare. There are several typical applications of wearable devices such as fall detection ^{[1, 2]}, distance walked and indoor positioning ^{[3, 4]} or physiological parameters, e.g. blood pressure or ECG ^[5]. Ageing population ^[6] in developed countries opens an important topic, how to enable suitable and economically feasible health and social care using new unobtrusive technology for new type of social/health care services.

In our laboratory, we are trying different technology, preferably a wireless technology to assist in services for elderly and other vulnerable persons too. The key parameters for long-term monitoring of user’s states based on measurements provided by wearable sensors were proposed to innovate existing laboratory system equipped with stationary sensors and actuators. For selected parameters, the framework for processing and evaluation of measurements was developed and its usability was examined in living lab – lab that simulates household’s conditions. Emphasis was on the robustness of the solution as well as on its applicability in various conditions. The attention was paid to complementary data procession from several sensors with intention of minimization of divert distortive effects or with intention of undesirable sensors’ properties.

Our research approach follows the rules for an effective work aiming in fast implementation of the applied research into real life. Therefore, we concentrated on existing technology on the market and related services ^{[6, 7]}.

Current situation on the market with embedded technology and services offers several partial solutions based on the implementation of new functions into smart watches, mobile phones or tablets. There are quite complex solutions between them like PAMSys, PLUXbiosignal, Simmer, Sensoplex, etc. (Karchnak 2015) ^[8].

However, there is still a space for the development of more complex services with affordable prices for the group of vulnerable persons.

2. MEMS Sensor Development and Trials

Our research group developed a sensor based on MEMS technology originally intended for tracking of a remote controlled car model. Later, we have been looking for other applications like assessment of a person’s mobility and her/his range of motion ^{[6, 7, 8]}. As we had already used optical methods for human motion analysis in our laboratory, we decided to use MEMS sensor for a persons’ motion abilities evaluation because we expected more simple measurement and data evaluation too. The further step after measurements of biomechanical and kinesiological parameters was to include such sensor into technology for monitoring of a person’s physical activities and states.

A small circuit board was developed aiming in minimizing of sensor size (PCB). Our functional sample was double-sided board PCB, and after adding other units our prototype got a cover too as shown in Figure 1.

Download as

• PPT
  PowerPoint Slide
• PNG
  Larger image(png format)

Veiw figureFigures index

•  NEW
  View larger figure in new window

View next figure

Figure 1. IMU prototype

As sensor of acceleration, we used 3-axial accelerometer MMA7431 from FreeScale and gyroscope ITG3200 from InvenSense for angular velocity sensing. As control unit, we used microcontroller ATMEGA32L.

Communication of prototype with PC via USB is provided by FT232RL chip. Device communicates with PC via USB that is serving also as power supply. For implementation of wireless technology, UART pins RX and TX are ready. ^[9]

Implemented accelerometer MMA7431 is analog device, so MCU has to convert data to digital form utilizing its 10bit ADC. Gyroscope ITG3200 has 16-bit ADC on-chip, so the digital output can be send to MCU using I2C. As mentioned above, UART is employed for communication – FT232RL for USB or XBee, etc. Data is coded into a communication chain that is designed according to the following form:

$, x, y, z, T, 255, XX, YY, ZZ, 13, 10

Where:

$ - is the beginning of the chain,

x, y, z – are incoming raw acceleration data

T - is output of temperature sensor on gyroscope

255 – is separator character

XX, YY, ZZ – is gyroscope data

13, 10 – is end of the chain

Each of incoming sensor data are encoded by two bytes – high and low. These data need to be compiled and converted into one number and processed after that. Testing of device was performed using USB for communication as well as using wireless technology.

A very important issue for a successful monitoring is an optimal selection of a scenario. Variety of wearable sensors and their functions enable different innovative solutions adjusted to specific type of persons and their environments. Monitoring of daily activities is now commonly integrated with functionalities oriented on long time monitoring of physiological parameters served for the health care aims. Clients then may stay at home and in the same time their basic functions are under supervision. In such way, wearable sensors offer important information about health conditions, activities of clients, and their behavior.

Analysis of the long time collected data enables identification and even prediction of a critical health situation. For example, changes in gait speed, and hip extension may indicate a negative changes in health conditions related to a person’s mobility. Such data may play an important role in the health and social care systems, especially for vulnerable persons, elderly including.

The main goal for implementing of wearable sensors is to enable identification of dangerous behavior, or health status of vulnerable persons without creating an obstructive home environment. In the same time such service will improve safety, self-confidence, and well- being of users. Such complex goal may be reached by the integration of wearable sensors into the intelligent environments, where stationary sensors network exists, and control system provides supervisory tasks and actions.

The general structure of a complex supervising system consists of:

1. Stationary sensors of environment.

2. Wearable sensors for motion activities and physiological parameters.

3. Communication interfaces.

4. Techniques for data processing and evaluation.

5. Models of user’s behavior and actual behavioral analysis to discover significant changes.

Monitoring of users living in a supported environment is a subject of many research projects. Many of them are focusing on the design of complex supervising system. However, many of them stress a specific role of such system like saving energy ^{[10, 11, 12]}, recognition of individual users, and monitoring of users health status ^{[13, 14, 15, 16, 17]}. For example, the system may identify possible Parkinson’s disease in behavioral changes of a person like slowing medium walking speed, and changes in hip extension ^[18]. Above mentioned researchers apply different methods for data analysis. Basically, the system may work in the following structure shown in Figure 2.

Download as

• PPT
  PowerPoint Slide
• PNG
  Larger image(png format)

Veiw figureFigures index

•  NEW
  View larger figure in new window
• PREV
  View previous figure

View next figure

Figure 2. System architecture for detection of changes in user’s behavior (adapted from [15])

Stationary sensors play an important role in monitoring of the environmental data like temperature, lighting, energy consumption, but also for identification of daily activities of users. Embedded wearable sensors add more personal information dealing with health status or critical situations during physical activities – fall, etc. Therefore both kinds of sensors are important for identification of critical situations, and behavioral changes too.

Wearable sensors usually offer the following data:

•  Numbers of steps per day n(i).

•  Medium walk speed v(i).

•  Medium step lengths l(i)

•  Energy consumption e(i)

•  Percentage of time spent in static and dynamic activities a(i).

•  Total time of dynamic activities t(i).

•  Extension in hip joint r(i).

•  Numbers of activations of the panic button p(i).

•  Part of the room space where user spends majority of time during day q(i).

Wearable sensors must be warned all the day. In the case of a critical accidental situation, a signal is generated and sent to supervising system (social centre with 24 hours service, or a member of family, etc.). Potential dangerous situations and necessary actions are described in the next chapter.

Definition of dangerous and potentially dangerous situations depends on many factors. Analysis of human motion activities offers prediction of many dangerous health critical states. In Table 1, we summarized the most frequent acute statuses and raised actions:

Table 1. Classification of dangerous, potentially dangerous situations

Download as

PowerPoint Slide

Larger image(png format)

• Tables index

Veiw figure

View current table in a new window

There is a relation between less walking intensity - lowering number of steps, length of walk, and walk speed-and fall risk; variation in step length and in the same time increasing of hip extension may indicate the Parkinson disease. Increasing time spent in the bed may indicate a depression, or great mobility problems, etc. These are reasons to monitor above mentioned parameters, evaluate them in long time period, and provide actions when necessary.

3. Experimental Testing

Fall detection plays an important role in vulnerable person’s life. Fall is the main reason of many injuries, even deaths, especially in elderly population. Fast reaction and contacting of caregivers or family can save life of injured person. There are many different ways how to recognize the fall, however utilizing MEMS sensors has following advantages:

•  device is small and reliable,

•  device can work remotely,

•  capable of employing the smart algorithms,

•  low cost device.

Development of reliable and robust algorithm of fall detection is crucial. One of the possible ways of fall detection is comparison of absolute value of acceleration to the threshold and after passing this threshold follows computation of the tilt. For such sensing, one triaxial accelerometer is sufficient. Accelerometer senses gravitational vectors in each axis. Amplitude of measured gravity changes its value as sinus α between sensing axis and horizontal plane (Figure 3).

Download as

• PPT
  PowerPoint Slide
• PNG
  Larger image(png format)

Veiw figureFigures index

•  NEW
  View larger figure in new window
• PREV
  View previous figure

View next figure

Figure 3. Sensing axis and horizontal plane

Our device contains 3-axial accelerometer and gyroscope. Accelerometer provides information about higher acceleration values that can indicate fall. But similar information gives also a jump, or stair climbing. Gyroscope provides us with angular velocity and we can derive tilt value from that too. By combining these data – acceleration, angular velocity and tilt, we obtain complex motion information. Setting the appropriate thresholds for each parameter is of utmost priority.

According to the possibility of obtaining information about acceleration, angular velocity and tilt, we proposed simple fall algorithm (Figure 4). Absolute value of acceleration with frequency of 20Hz is monitored, and if it passes threshold |A_T| then the following 30 samples are monitored for passing two other thresholds – angular velocity threshold |ω_T| and tilt threshold |θ_T|. If all conditions are true, then a fall has occurred with a high probability.

Download as

• PPT
  PowerPoint Slide
• PNG
  Larger image(png format)

Veiw figureFigures index

•  NEW
  View larger figure in new window
• PREV
  View previous figure

View next figure

Figure 4. Fall detection algorithm

To verify proposed algorithm, we performed experiments. IMU was placed on the belt of the proband. During experiment, data was sent to PC with frequency of 20Hz. Person was walking on the floor, with little disturbing elements such as small stairs, and after more than 25 seconds asimulated fall occurred. Our threshold parameter settings were following: =2,5g, =300°/s, =45°. These thresholds were set by empirical testing. Graphical interpretations of experimental data are shown in Figure 5.

Download as

• PPT
  PowerPoint Slide
• PNG
  Larger image(png format)

Veiw figureFigures index

•  NEW
  View larger figure in new window
• PREV
  View previous figure

Figure 5. Visualized data from experiment

The device has to be reliable, it is essential to detect true alarm and neglect false ones. We decided to assess reliability of proposed algorithm as a ratio of false alarms to all alarms, as in (1):

(1)

where – reliability, – number of false alarms, – number of all alarms

Results were satisfactory – reliability was r=0,91. We assessed feasibility of acceleration as determinant of fall detection using (1), but results were not satisfying at all (r < 0,6). It means that acceleration itself does not provide enough information for fall detection. Consequently, using of appropriate sensor combinations and data fusion as angular velocity and tilt are essential. Acceleration data as a trigger for further assessment of other movement elements (e.g. angular velocity) seem to be acceptable and suitable solution.

Proposed algorithm was created in C++ and implemented into PC application for fall detection. In our future work we plan to apply various wireless technologies, such as XBee for communication of fall detector with control system. System will run on Raspberry Pi due to its low cost and appropriate properties.

Acknowledgement

This work has been supported by the Slovak Grant Agency VEGA contract Nb. 1/0911/14 “Implementation of wireless technologies into the design of new products and services to protect human health”.

References

[1]	Huang, C.N., Chan, C.T.,A ZigBee-Based Location-Aware Fall Detection System for Improving Elderly Telecare, International Journal of Environmental Research and Public Health, vol. 11, 2014, 4233-4248.
	In article		View Article  PubMed
[2]	Bianchi, V., Grossi, F., De Munari, I., Ciampolini, P., MuSA: a multisensor wearable device for AAL. Proceedings of the Federated Conference on Computer Science and Information Systems, 2011, 375-380.
	In article
[3]	Bishop, E., Li, Q., Walking speed estimation using shank-mounted accelerometers. IEEE International Conference on Robotics and Automation. 2010, 5096-5101.
	In article		View Article
[4]	Lee, Ch. Y., Lee, J.J., Estimation of Walking Behavior Using Accelerometers in Gait Rehabilitation. International Journal of Human-friendly Welfare Robotic Systems, vol. 3, no. 2, 2002, 32-35.
	In article
[5]	Patel, S. et al., A review of wearable sensors and systems with application in rehabilitation. Journal of NeuroEngineering and Rehabilitation, 9, 21, 2012, 1-17.
	In article
[6]	Biros, O., Karchnak, J., Simsik, D., Multiagent approach in home care based on open architecture. 2013. In: American Journal of Mechanical Engineering. Vol. 1, no.7, 2013, p. 217-220.
	In article
[7]	Karchnak, J., Simsik, D., Jobbagy, B., Galajdova, A., Onofrejova, D., MEMS sensors in evaluation of human biomechanical parameters. In: Procedia Engineering: Modelling of Mechanical and Mechatronic Systems MMaMS 2014: 25th-27th November 2014, High Tatras, Slovakia. Netherland: Elsevier, 2014 Vol. 96, 2014, p. 209-214. Available: https://www.sciencedirect.com/science/article/pii/S1877705814032044#...[Accessed 27 oct 2015].
	In article
[8]	Karchnak, J., Simsik, D., Jobbagy, B., Onofrejova, D., Feasibility Evaluation of Wearable Sensors for Homecare Systems. In: Acta Mechanica Slovaca. Nb. 2, 2015, 1-8. Available: https://www.actamechanica.sk/e-publications/2015/volume-19-issue-no-2...[Accessed 21sept 2015].
	In article
[9]	Simsik, D., Karchnak, J., Jobbagy, B.,Galajdova, A.,Design of inertial module for rehabilitation device.In: SAMI 2013: IEEE 11th International Symposium on Applied Machine Intelligence and Informatics proceedings: January 31 - February 2, 2013, Herľany, Slovakia. Print: Budapest: IEEE, 2013, 33-36.
	In article		View Article
[10]	Vavilov, D., Melezhik, A., Platonov, I., Smart Home User’s Behavior Prediction. In: IEEE Third International Conference on Consumer Electronics (ICCE). 2013, s. 261-264.
	In article		View Article
[11]	Bao, K., Allerding, F., Schmeck, H., User Behavior Prediction for Energy Management in Smart Homes. In: Eighth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD). 2011, s. 1335-1339.
	In article		View Article
[12]	Chen, C., Cook, D.J., Behavior-based Home Energy Prediction. In: Proceedings of the 2012 Eighth International Conference on Intelligent Environments. IEEE, s. 57-63.
	In article		View Article
[13]	Fatima, I. et all., A Unified Framework for Activity Recognition-Based Behavior Analysis and Action Prediction in Smart Homes. Sensors (Basel, Switzerland) 13.2 .2013, 2682-2699.
	In article
[14]	Abdelsalam, H., Cook, D.J., Schmalz, M., Smart Home-Based Health Platform for Behavioral Monitoring and Alteration of Diabetes Patients. Journal of diabetes science and technology (Online), 2009, 141-148.
	In article
[15]	Aztiria, A., Farhadi, G., Aghajan, H., User Behavior Shift Detection in Ambient Assisted Living Environments. Ed. Antonio Jara et al. JMIR mHealth and uHealth, 2013.
	In article		View Article
[16]	Aztiria, A., Augusto, J.C., Izaguirre, A., Autonomous Learning of User’s Preferences improved through User Feedback. In: Proceedings of BMI. 2008, 72-86.
	In article
[17]	Kadouche, R., Chikhaoui, B., Abdulrazak, B., User’s behavior study for smart houses occupant prediction. In: Annales des Télécommunications, vol. 65, 2010, 539-543.
	In article		View Article
[18]	Whittle, M.V., Gait Analysis: An Introduction. Butterworth-Heinemann, 3. edition, p. 232.
	In article