Enabling Multiple BSN Applications Using the ... - Semantic Scholar

6 downloads 138929 Views 4MB Size Report
basically, two radios, one 2.4 GHz 802.15.4 and one LF for the. HR, coexist on the .... an Android implementation of the SPINE coordinator to create an open and ...
Enabling Multiple BSN Applications Using the SPINE Framework Raffaele Gravina†‡ , Andreoli Alessandro§ , Alessia Salmeri∗ , Luigi Buondonno∗ , Nikhil Raveendranathank , Vitali Loseuk , Roberta Giannantonio∗ , Edmund Seto¶ and Giancarlo Fortino‡ ∗ Telecom Italia, Turin, Italy Wireless Sensor Networks Lab, Telecom Italia, Berkeley, CA 94720 ‡ Department of Electronics, Informatics, and Systems, University of Calabria, Rende, Italy § Department of Electronics, Telecommunication, and Biomedic Engineering, Universita’ Politecnica delle Marche, Ancona, Italy ¶ School of Public Health, University of California, Berkeley, CA 94720 k Embedded Systems and Signal Processing Lab, University of Texas at Dallas, Dallas, TX 75080 †

Abstract—Employment of BSN-based technologies in real world scenarios requires a flexible infrastructure at both hardware and software level. In this paper, we emphasize how the use of SPINE (Signal Processing In-Node Environment), a software framework for BSN, supports the development of heterogeneous health-care applications based on reusable subsystems. One of the main goal of SPINE is to provide a flexible architecture that can support variety of practical applications without the need for costly redeployment of the code running on sensor nodes. We also present a SPINE sensor node emulator that supports the first phase of the algorithm design, when the actual hardware devices may not be available. This approach can guide the choice of the required hardware (e.g. the sensors) to meet the application requirements based on the results obtained in the emulated environment. Such tool can simplify the research collaboration during the specification stage of a project, due to availability of a common (virtual) architecture. Index Terms—Body Sensor Networks, Domain specific frameworks, SPINE, Health care application.

I. I NTRODUCTION The design of health care systems based on Wireless Sensor Networks (WSNs) is complex, particularly due to the challenge of implementing intensive signal processing algorithms for data interpretation on highly resource limited wireless nodes, and have to meet hard wearability and battery duration requirements [1]. Furthermore, debugging software on sensor nodes is very difficult and time consuming due to the lack of support architecture in embedded operating systems; redeploying the debugged code on the actual sensing devices takes significant amount of time as well. SPINE (Signal Processing In-Node Environment) [2] is an open source domain-specific framework designed to support flexible and distributed signalprocessing for wireless body sensor network (WBSN) systems. The main goal of SPINE is to provide WBSN developers with support for rapid prototyping of signal-processing applications. In SPINE, sensors and common processing blocks, such as math aggregators and threshold-based alarms, can be configured independently and connected together arbitrarily at run-time based on external controls. Such an approach

allows heterogeneous applications to be built atop the same basic software components, enhancing code reusability and, more importantly, removes the need for redeploying the nodeside code based on a particular application. For instance, in section V we show that a heart rate monitor and a physical activity recognition system can all be implemented without intervention on the code running on the nodes. Obviously, to support different applications, the wearable sensing node(s) must be equipped with all the required sensors. When testing signal processing for BSNs it is important to verify its accuracy with data collected from real subjects that application is trying to target. It is a simple task for the off-line signal processing, since all that is required is a data collection tool that can record sensor readings for future processing. The task is complicated when on-line in-node processing is involved. It requires the software to function accurately since it might be difficult to repeat the experiment with real subjects. Additionally, it may be difficult to evaluate usefulness of the hardware before physically integrating it into the system. To address these problems, SPINE has been extended with an emulator tool which virtualizes SPINE-enabled sensor nodes. The tool allows emulation of a set of nodes forming a WBSN and requires a data set for each node. The data set can be built using a provided data collector tool which records data from real sensor nodes. Hence, a particular emulated node is virtually equipped with sensors determined by the given data set. There are several advantages of using a SPINE emulator. For instance, processing functionalities can be implemented in the emulated environment first to simplify the debug process. Furthermore, data set from real sensors can be used to validate and compare objectively different processing algorithms or hardware set-ups. Finally, the emulator with a standard data set can be used by interested developers to investigate the potential of the SPINE framework itself, even if they do not have suitable physical sensor nodes. The remainder of the paper is organized as follows. Section II summarizes the related work. Section III introduces the functionalities of the latest SPINE release. Section IV

describes in detail the proposed SPINE-node emulator. Section V provides an overview of some applications developed with SPINE. Finally, the main contribution of the paper are discussed in Section VI. II. R ELATED W ORK Several academics and industry research projects have recently focused on the use of BSNs including sensors such as accelerometers, gyroscopes, ElectroCardioGram (ECG) sensors, ElectroMyoGraphy (EMG) sensors, and pulse oxymeters [1]. Typically, BSN applications are built on top of a network architecture composed of a coordinator node and several sensor nodes connected in a star topology. Sensor nodes transmit raw readings or processed data to the coordinator node, which in turn may run algorithms for off-line data analysis or realtime monitoring. In the last years, research efforts have been dedicated to the introduction of middleware and frameworks to address the important issue of simplifying the development of BSN-based applications. In [3], a signal-processing-oriented framework, called Titan (Tiny Task Network), is proposed. The goal of Titan is to provide a mechanism to deploy and dynamically configure a network of data processing tasks on a WSN. In particular, Titan, which is developed atop TinyOS, allows the definition of tasks, implementing signal processing functions, linked through connections, which transport data from one task to another to form a task network, which describes the application to be run on the WSN. Titan has been applied to activity recognition based on SensorButton devices equipped with accelerometers [4]. In [5], the CodeBlue framework for developing BSN-based health care monitoring applications is described. CodeBlue, which is implemented in nesC/TinyOS at the sensor-side and in Java at the base station-side, provides protocols for device discovery; publish/subscribe multihop routing, and a simple query interface allowing caregivers to request data from groups of patients. Other existing systems include HealthGear [6], MobiCare [7], WWBAN [8], ALARM-NET [9], Participatory Sensing [10], and Intel MSP [11]. These systems instrument the human body as an active mobile platform, and because they are noninvasive they can support persistent monitoring in people’s normal living environments. Several further projects have developed prototypes for various human activity monitoring applications [12], [13]. Most of the above applications have limitations that make them less practical. For example, the implementation of most of these systems is based on centralized architectures where raw data is transmitted from the sensor nodes to the coordinator that performs most of the data processing and classification tasks. This often results in application specific node code. Instead, the SPINE architecture is based on a model where a general node side code is used for all of the applications. To make the model flexible, the architecture also defines a set of controls that allows the server node to customize the node behavior for each specific application. This functionality is achieved via a communication system that relates Java code of the server node to the nesC code of a sensor node.

III. T HE SPINE 1.3 F RAMEWORK : AN OVERVIEW SPINE (Signal Processing In-Node Environment) [2] is an open source domain-specific framework designed to support flexible and distributed signal-processing for WBSNs. The main goal of SPINE is to provide WBSN developers with support for rapid prototyping of signal-processing applications. Other design principles include code reusability and efficiency, application interoperability and specific support for sensing operations. SPINE acts at application-level and is independent of the specific network topology and underlying low-level communication protocol. However, the current implementation focuses on star-topology networks with multiple sensor nodes and a single coordinator node; this network architecture is particularly suitable for BSNs that often move responsibilities from resource constrained nodes to a more powerful portable device or a computer. A. Architecture SPINE 1.3 release [14] focuses on optimization and simplification of the existing framework, while the core architecture defined in the previous release is unchanged. For example, tailoring mechanism of the node-side framework deployment has been significantly simplified. Multiple changes have been made to achieve that. The required sensor drivers are automatically included in the compilation process depending on the selected platform and sensor-board. Available processing units can be plugged and unplugged, before compiling and flashing the nodes, through simple flags located into the framework makefile. Another significant contribution is the integration of a security service to the framework communication, which uses the hardware encryption of the CC2420 radio chip, which is the most commonly used transceiver in WSN hardware platforms. It is optionally enabled at compile time, as it requires extra memory resources. B. Available functionalities Along with the feature computation (common mathematical aggregator functions on sensor reading windows) and the user tunable threshold-based alarms, we have integrated a new service into SPINE for transmitting over the air chunks of sensor readings in single messages. This improves the maximum sampling rate and, at the same time, reduces the radio duty cycle, therefore reducing bandwidth usage and battery consumption. However, the disadvantage is caused in the event of message dropped over-the-air, because multiple readings contained in a single message would be lost at once. Recently, a simplified version of the Hidden Markov Model (HMM) classifier presented in [15] has also been ported into SPINE. A simple algorithm for counting walk steps using accelerometer data has also been included in the 1.3 distribution. It can be activated on-demand through the coordinator side APIs. Furthermore, we implemented an algorithm, which accounts for gravity compensation to compute vertical and horizontal activity counts from three-axis accelerometer data. This

Fig. 1.

Telosb Cardio Shield

functionality is a standard feature in SPINE and has been used for energy expenditure application, which is discussed in subsection V-D. To date, SPINE supports Telosb/tmote sky, Micaz and Shimmer sensor platforms. In particular, supported sensors are: the optional kit (humidity, environmental temperature, total solar radiation and photosynthetically active radiation sensors) for the Telosb/tmote sky platform, the three-axis accelerometer sensor on the Shimmer and the two-axis accelerometer on the MTS300 sensor-board of the Micaz platform. Furthermore, two open sensor-boards for the Telosb/tmote sky platform were developed. The first is a motion sensor-board equipped with a three-axis accelerometer and a two dimensional gyroscope, which is presented in [16]. The second board is shown in Fig.1; it features a three-axis accelerometer and a wireless heart beat monitor. An additional sensor-board uses four electrodes connected to the surface of the ribcage [17], also including a three-axis accelerometer. The electrodes are used for both heart (ECG signal) and breathing measurements. Breathing is measured using electrical impedance pneumography (EIP). IV. SPINE 1.3 N ODE E MULATOR Designers of algorithms and application developers that want to test an implementation on data collected from the sensors sometimes have to cope with the lack of the specific hardware. To avoid this issue we developed the SPINE node emulator application that virtualizes the real hardware. In this way users with the hardware can collect, tag, store, and share data with other users. Moreover, the same sensor node might be used several times for different data acquisitions. Stored data can be used for testing different algorithms as if the nodes were real. This approach will easily allow comparing different solutions to the same problem (e.g. different algorithms solving the same classification problem), since the same data can be used several times. Furthermore, this will speed up the development process in a real application, since the application developed for virtual sensors will perfectly run with real sensors without the need of writing new code. In the following we first describe the application for data gathering (data collector) and then briefly explain the principles behind the node emulator module. A. Data Collector The main goal of the SPINE Data Collector application is to collect sensor data and store them in a common format. Therefore, Data Collector main functionalities are nodes configuration, sensor data collection and storage. Data file formats

currently used for data storage are: Attribute-Relation File Format (ARFF) [18], Comma Separated Values (CSV) and TeXT file (TXT). Sensor datasets basically consist of two-dimensional arrays composed of five columns and one row for each sensor reading. In detail, dataset columns are described below: • ClassLabel: the logical label that tags sensor data (e.g. outdoor walking); • FeatureDataId: a numeric attribute reporting the packet data sequence number; • FeatureId: the index of feature in the packet data; • SensorCodeFeatureCodeChNum: a string reporting three codes, the sensors, the features and the channels; • FeatureValue: the numeric value sent by the sensor. An ARFF file is an ASCII text file that describes a list of instances sharing a set of attributes. ARFF files were developed by the Machine Learning Project at the Department of Computer Science of The University of Waikato for use with theWeka machine learning software. ARFF files have two distinct sections. The first section is the Header information, which is followed the Data information. The Header of the ARFF file contains the name of the relation, a list of the attributes (the columns in the data), and their types. The Data of the ARFF file contains the instances. Each instance is represented on a single line and attribute values for each instance are delimited by commas. B. Spine Node Emulator SPINE Node Emulator is a Java stand-alone application that uses collected data (Data Collector output) to emulate SPINE sensor nodes. Each “Node Emulator” instance is a “virtual sensor node”, multiple Node Emulator instances might be launched to emulate a network. A “virtual sensor node” has one or more “virtual sensors”, responds to SPINE commands (i.e. Discovery, Reset, SensorSetup, FunctionSetup, ActivateFunction and Start) and provides collected sensor data or features. Data is read from datasets. As a consequence, a virtual node will not be able to provide data that has not previously collected. Node Emulator might be used in two configurations, depending if the feature computation is disabled or not. When feature computation is disabled, the emulator provides only the data that is already present in the datastore. If the feature computation is enabled, the emulator computes requested features on the stored raw data samples. V. SPINE-BASED BSN A PPLICATIONS One of the key advantages of SPINE is the ability to satisfy diverse application needs at run-time, avoiding, in most situations, the costly redeployment of the code running on the remote sensing devices. In fact, processing functionalities that are very common in the WBSN domain, such as timedomain feature computation, threshold-based alarms, but also simple raw data transmission, are enabled by default during the first deployment of the node-side software. Furthermore,

available sensors can be set-up, activated and disabled dynamically and their output can be arbitrarily connected to the on-line processing, adding a further level of flexibility. As aforementioned, removing the need to reprogram the nodes while switching or changing applications, is very desirable especially in real world scenarios. For instance, a doctor could use SPINE nodes and a suitable coordinator device, such as a smartphone or a personal digital assistant (PDA), to monitor weekly energy expenditure of a patient. The same nodes could be used later with another patient in a rehabilitation scenario, as long as the proper application software is available on the coordinator node. In the following paragraphs, we show four test-case applications built on top of SPINE. These applications configure the SPINE network differently to meet their requirements; however the same hardware devices and node-side software (i.e. SPINE) has been used by all the three applications. A. Activity monitoring Detection of physical activity might enable new types of health assessment and intervention tools that help people maintain their energy balance and stay physically fit and healthy. Recent research [12], [13] has shown that wearable accelerometers can be used to reliably detect physical activity. Past work in recognizing activities from accelerometer data has used computationally intensive supervised classification algorithms that typically require off-line analysis. In this application we utilize a k-nearest neighbors (k-NN) centroid based algorithm that runs directly on SPINE sensor nodes. This allows the transmission of only the label representing the activity state as it happens. Since the energy cost of the wireless transmission is defined in terms of the amount of data to be transmitted, replacing data or even feature transmission with a small label drastically decreases the energy expenditure of each classification. We also study the heart rate (HR) data for discriminating the intensity of activities. HR may be useful since it correlates with energy expenditure for aerobic exercise; however, if used alone it provides little information about the activity type, in fact it is influenced by other factors for example by emotional states, environment temperature, and fitness level. The overall system overview is shown in Fig.2. It consists of two SPINE nodes, both equipped with a three-axis accelerometer. One of them is also featured with a HR wireless sensor; basically, two radios, one 2.4 GHz 802.15.4 and one LF for the HR, coexist on the same board. The wireless solution for the HR monitoring is less invasive then the typical wired electrode based HR monitor. The application recognizes the following activities: standing, walking, lying and sitting [2]. B. Physical rehabilitation Wearable wireless sensors might be also used for physical rehabilitation purposes. It is quite common to require repetitive physical exercises for instance to recover from a muscle strain or a surgery. Having a real-time feedback about the exercise performance quality would allow users to independently

Fig. 2.

Placement of the sensors on the body.

exercise properly without the need of a continuous professional assistance. Motion sensors are the most appropriate for this kind of applications: they can be worn on the part of the body that needs to be exercised and report precise data about how the movement is being performed. Different motion sensors (accelerometers, gyroscopes, magnetometers, etc) would provide different information about the position of the body in space. In our physical rehabilitation application we used SPINE nodes fitted with accelerometer sensors to monitor arms and legs movements. The application consists of monitoring legs and arms bending movements in realtime and comparing them with the ones recorded during set up phase. Main requirements of this application are realtime processing and high precision in movement detection, therefore we did not use SPINE distributed signal processing capabilities but we used the raw data feature functionality of the SPINE framework. This way we can be sure all sampled data is provided to the SPINE coordinator as input to the algorithm. Despite not using the SPINE on-node signal processing functionalities in this application, the usage of the SPINE framework did speed up the development process. The standard on-node firmware application (SPINE node side) removed the need to develop any additional nesC code. Additionally, the existent communication framework between the node and server SPINE applications allowed developers to focus on Java implementation of the classification algorithm on the coordinator computer. The application scenario consists of two steps, namely set up and exercise phases. During the set-up phase, the user wears a couple of sensors on either leg or arm that needs to be exercised and performs the correct exercise under the guidance of rehabilitation professional. Meanwhile the system records the data and stores it as reference exercise. Set up phase might be optional (default references are provided) but it is strongly suggested under supervision of the rehabilitation personnel. Then, during the exercise phase, the user repeats the bending movement and is provided with a realtime feedback about how the movement is done with respect to the stored reference. The application can also provide or store a final score indicating how good the exercise has been performed. Such score can be an indicator for a therapist to assess effectiveness and decide on future course of patient’s exercises.

Fig. 3.

Hidden Markov Event Model training procedure.

C. Gait analysis Gait analysis [19] is the study of human walking. Recently, it has been augmented by instrumentation for measuring body movements, body mechanics and the activity of the muscles. Gait analysis is used to assess, plan and treat individuals with conditions affecting their ability to walk. It is also commonly used in sports such as athletics to help athletes run more efficiently and to identify posture-related or movement-related problems in people with injuries. Pathological gait may reflect compensations for underlying pathologies, or be responsible for causation of symptoms in itself. The study of gait allows these diagnoses to be made, as well as permits future developments in rehabilitation engineering. Minor variations in gait style can be used as a biometric identifier to distinguish individual people. A novel method of gait analysis using a HMM-based technique to extract temporal parameters from gait has demonstrated good results and potentialities to be partially implemented on the hardware constrained sensing nodes [15]. The system is initially trained based on the initial walking data sample. The overall training process is shown in Fig.3 To validate the system, the sensor node performed annotation and broadcast the raw samples so results between the embedded implementation and a Matlab and a pure C version could be compared. As shown in Fig.4, the versions produce similar results. While the state transition are not completely synchronized, the total times for a complete cycle are almost identical (on average, the difference is less then 10ms). The dotted box highlights the components also involved during the classification phase. The HMM classifier has been implemented in SPINE to run real-time on the nodes, while an off-line processing is still required to train the algorithm. Given the trained HMM model, SPINE nodes can classify the four walk phases (initial swing, mid-swing, terminal swing/initial stance, mid-stance) and report back to the coordinator node the classification results. In particular, in our test case, we extracted heel-down and heel-lift events from a walking subject, using a single wearable node at the waist equipped with a three-axis accelerometer sensor. The sensor node performed annotation as well as broadcast the raw samples so results

Fig. 4.

Comparison of HMM-based gait analysis implementations.

could be compared with a fully off-line MATLAB implementation. D. Energy expenditure Physical activity is essential to overall human health. The U.S. Surgeon General notes that regular physical activity can increase the chances of living longer, and reduce the risk of obesity, heart disease, and some cancers [20]. Adults should have at least 150 minutes of moderate-intensity physical activity per week, while children and teenagers should have 1 hour of physical activity per day that includes vigorous and bone-strengthening activities [21]. Accurate and objective measurement of daily activity is needed to determine whether these guidelines are met. Selfreported activity diaries are typically problematic because of difficulty recalling activities, and subjectivity of reporting intensity (how vigorous is the activity?). Pedometry and accelerometry measurements are a preferred measurement method. In a recent review of studies that compared direct assessment of childrens physical activity using accelerometry versus other indirect methods (e.g., surveys) [22], direct assessments were favored in that they avoid recall and response biases that affect self-report methods. These direct methods result in an estimation of energy expenditure. An open-source application estimating the energy consumed while performing daily basic activities has been implemented on top of SPINE. It uses a single sensor node equipped with a three-axis accelerometer, placed on the belt. Because accelerometer data are pre-filtered removing the gravity components [23], the mote can be arbitrarily oriented. Furthermore, experiments on 10 subjects walking, running, ascending/descending stairs, and sitting showed high correlation (Pearsons correlation coefficient = 0.97) to the results of a commercially available device [24], with good relative energy expenditures for different activities; sedentary activity (sitting) producing the lowest energy expenditure, with increasing order of expenditures for walking, stairs, and running. The application requires a SPINE 1.3 mote with the Kcal feature (integrated by default). As part of the algorithm runs on the node, in which the accelerometer is sampled at 33Hz and activity counts on the horizontal and

vertical axis are computed on 30-sample windows and sent to the SPINE coordinator every second. The SPINE coordinator collects 60 results and computes an estimation of the energy expenditure (expressed in Kcal or KJ) every minute. The algorithm can be tuned with the subject gender and weight. VI. C ONCLUSIONS In this paper we have shown the use of the SPINE framework to support heterogeneous health-care applications without redeployment of the code running on the nodes. We have demonstrated the flexibility of our framework describing four different applications for physical activity detection, physical rehabilitation, gait analysis and energy expenditure estimation, which all exploit the same sensor node hardware and software. We have also proposed a sensor node emulator that supports the first phase of the algorithm design, and guides the choice of actual hardware based on the performance results obtained during simulation. There are a number of advantages in using such tool. For instance, processing functionalities can be implemented in the emulated environment first to simplify the debug process; data set from real sensors can be used to validate and objectively compare different processing algorithms; the emulator is useful to investigate the potential of the SPINE framework itself, even if suitable physical sensor nodes are not initially available. We are currently working on an Android implementation of the SPINE coordinator to create an open and mobile environment for the development of the end-user applications. ACKNOWLEDGMENT The authors would like to thank Fabio L. Bellifemine, Stefano Galzarano, Antonio Guerrieri, Roozbeh Jafari, Kevin Klues, Philip Kuryloski, Ville-Pekka Sepp¨a, Marco Sgroi, Po Yan for their valuable contribution to the SPINE project. This work has been partially supported by CONET, the Cooperating Objects Network of Excellence, funded by the European Commission under FP7 with contract number FP7-2007-2224053. Partial support was also provided by CITRIS, the Center for Information Technology in the Interest of Society.

R EFERENCES [1] P. S. Pandian, K. P. Safeer, P. Gupta, D. T. I. Shakunthala, B. S. Sundersheshu, and V. C. Padaki, “Wireless sensor network for wearable physiological monitoring,” JNW, vol. 3, no. 5, pp. 21–29, 2008. [2] R. Gravina, A. Guerrieri, G. Fortino, F. Bellifemine, R. Giannantonio, and M. Sgroi, “Development of body sensor network applications using spine,” in Systems, Man and Cybernetics, 2008. SMC 2008. IEEE International Conference on, pp. 2810–2815, Oct. 2008. [3] C. Lombriser, D. Roggen, M. St¨ager, and G. Tr¨oster, “Titan: A tiny task network for dynamically reconfigurable heterogeneous sensor networks,” in 15. Fachtagung Kommunikation in Verteilten Systemen (KiVS), Informatik aktuell, pp. 127–138, Springer, 0 2007. [4] C. Lombriser, N. B. Bharatula, D. Roggen, and G. Tr¨oster, “On-body activity recognition in a dynamic sensor network,” in 2nd International Conference on Body Area Networks (BodyNets), 0 2007. [5] V. Shnayder, B.-r. Chen, K. Lorincz, T. R. F. F. Jones, and M. Welsh, “Sensor networks for medical care,” in Technical Report TR-08-05, Division of Engineering and Applied Sciences., Harvard University, 2005.

[6] N. Oliver and F. Flores-Mangas, “Healthgear: A real-time wearable system for monitoring and analyzing physiological signals,” in BSN ’06: Proceedings of the International Workshop on Wearable and Implantable Body Sensor Networks, (Washington, DC, USA), pp. 61–64, IEEE Computer Society, 2006. [7] R. Chakravorty, “A programmable service architecture for mobile medical care,” in PERCOMW ’06: Proceedings of the 4th annual IEEE international conference on Pervasive Computing and Communications Workshops, (Washington, DC, USA), p. 532, IEEE Computer Society, 2006. [8] A. Milenkovi´c, C. Otto, and E. Jovanov, “Wireless sensor networks for personal health monitoring: Issues and an implementation,” Comput. Commun., vol. 29, no. 13-14, pp. 2521–2533, 2006. [9] A. Wood, G. Virone, T. Doan, Q. Cao, L. Selavo, Y. Wu, L. Fang, Z. He, S. Lin, and J. Stankovic, “ALARM-NET: Wireless sensor networks for assisted-living and residential monitoring,” University of Virginia Computer Science Department Technical Report, 2006. [10] J. Burke, D. Estrin, M. Hansen, A. Parker, N. Ramanathan, S. Reddy, and M. B. Srivastava, “Participatory sensing,” in SENSYS’06 Workshop on World-Sensor-Web: Mobile Device Centric Sensor Networks and Applications, Boulder, CO, USA 2006. [11] T. Choudhury, S. Consolvo, B. Harrison, J. Hightower, A. LaMarca, L. LeGrand, A. Rahimi, A. Rea, G. Bordello, B. Hemingway, P. Klasnja, K. Koscher, J. Landay, J. Lester, D. Wyatt, and D. Haehnel, “The mobile sensing platform: An embedded activity recognition system,” Pervasive Computing, IEEE, vol. 7, pp. 32–41, April-June 2008. [12] A. Dinh, D. Teng, L. Chen, Y. Shi, C. McCrosky, J. Basran, and V. Del Bello-Hass, “Implementation of a physical activity monitoring system for the elderly people with built-in vital sign and fall detection,” pp. 1226 –1231, april 2009. [13] A. Dinh, D. Teng, L. Chen, S.-B. Ko, Y. Shi, C. McCrosky, J. Basran, and V. Del Bello-Hass, “A wearable device for physical activity monitoring with built-in heart rate variability,” pp. 1 –4, june 2009. [14] http://spine.tilab.com/ 2010. [15] E. Guenterberg, H. Ghasemzadeh, and R. Jafari, “A distributed hidden markov model for fine-grained annotation in body sensor networks,” in BSN ’09: Proceedings of the 2009 Sixth International Workshop on Wearable and Implantable Body Sensor Networks, (Washington, DC, USA), pp. 339–344, IEEE Computer Society, 2009. [16] P. Kuryloski, A. Giani, R. Giannantonio, K. S. Gilani, V.-P. Seppa, E. Seto, R. Gravina, V. Shia, C. Wang, P. Yan, A. Yang, J. Hyttinen, S. S. Sastry, S. Wicker, and R. Bajcsy, “Dexternet: An open platform for heterogeneous body sensor networks and its applications,” November 2008. [17] V.-P. Sepp¨a, J. V¨ais¨anen, O. Lahtinen, and J. Hyttinen, “Assessment of breathing parameters during running with a wearable bioimpedance device,” in 4th European Conference of the International Federation for Medical and Biological Engineering, vol. 22 of IFMBE Proceedings, pp. 1088–1091, Springer Berlin Heidelberg, 2009. [18] I. H. Witten and E. Frank, Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations (The Morgan Kaufmann Series in Data Management Systems). Morgan Kaufmann, 1st ed., October 1999. [19] M.-F. Ho, K.-Z. Chen, and C.-L. Huang, “Gait analysis for human walking paths and identities recognition,” in Multimedia and Expo, 2009. ICME 2009. IEEE International Conference on, pp. 1054–1057, 28 2009-July 3 2009. [20] U. D. of Health and O. o. t. S. G. Human Services, “The surgeon generals vision for a healthy and fit nation.,” January 2010. [21] U. D. of Health and H. Services, “Dietary guidelines for americans.” U.S. Government Printing Office, 2005 January. [22] K. B. Adamo, S. A. Prince, A. C. Tricco, S. Connor-Gorber, and M. Tremblay, “A comparison of indirect versus direct measures for assessing physical activity in the pediatric population: A systematic review,” International Journal of Pediatric Obesity, vol. 4, pp. 2 – 27, March 2009. [23] D. Mizell, “Using gravity to estimate accelerometer orientation,” in ISWC ’03: Proceedings of the 7th IEEE International Symposium on Wearable Computers, (Washington, DC, USA), p. 252, IEEE Computer Society, 2003. [24] http://www.theactigraph.com/ 2010.

Suggest Documents