Activity classification using a single chest mounted tri-axial accelerometer

Medical Engineering & Physics 33 (2011) 1127–1135

Contents lists available at ScienceDirect

Medical Engineering & Physics journal homepage: www.elsevier.com/locate/medengphy

Activity classification using a single chest mounted tri-axial accelerometer A. Godfrey a,∗ , A.K. Bourke b,c,d , G.M. Ólaighin b,c , P. van de Ven d , J. Nelson d a

Clinical Ageing Research Unit, Newcastle University, Campus for Ageing & Vitality, Newcastle Upon Tyne NE4 5PL, United Kingdom School of Engineering & Informatics, NUI Galway, University Road, Galway, Ireland National Centre for Biomedical Engineering Science, NUI Galway, University Road, Galway, Ireland d Wireless Access Research Group, Department of Electronic and Computer Engineering, University of Limerick, Limerick, Ireland b c

a r t i c l e

i n f o

Article history: Received 29 November 2010 Received in revised form 3 May 2011 Accepted 6 May 2011 Keywords: Physical activity Accelerometer Gyroscope Discrete wavelet transform ADL Postural transitions Scalar product Dot product

a b s t r a c t Accelerometer-based activity monitoring sensors have become the most suitable means for objective assessment of mobility trends within patient study groups. The use of minimal, low power, IC (integrated circuit) components within these sensors enable continuous (long-term) monitoring which provides more accurate mobility trends (over days or weeks), reduced cost, longer battery life, reduced size and weight of sensor. Using scripted activities of daily living (ADL) such as sitting, standing, walking, and numerous postural transitions performed under supervised conditions by young and elderly subjects, the ability to discriminate these ADL were investigated using a single tri-axial accelerometer, mounted on the trunk. Data analysis was performed using Matlab® to determine the accelerations performed during eight different ADL. Transitions and transition types were detected using the scalar (dot) product technique and vertical velocity estimates on a single tri-axial accelerometer was compared to a proven discrete wavelet transform method that incorporated accelerometers and gyroscopes. Activities and postural transitions were accurately detected by this simplified low-power kinematic sensor and activity detection algorithm with a sensitivity and specificity of 86–92% for young healthy subjects in a controlled setting and 83–89% for elderly healthy subjects in a home environment. Crown Copyright © 2011 Published by Elsevier Ltd on behalf of IPEM. All rights reserved.

1. Introduction The study of ambulatory human motion with accelerometer and (or) gyroscope-based mobility monitors has become increasingly extensive in many areas of biomedical electronic research. These research areas include the monitoring of the patients (old and young) suffering from: falls, Parkinson’s disease, venous ulceration, back pain, osteoarthritis, mental disorders, obesity, etc. [1–9]. Reduction in sensor size and cost [10] coupled with more powerful data mining techniques has enabled the researcher to apply both single and multiple sensor arrays to these groups. Coley et al. [11] adopted the use of a single gyroscope sensor to study shank kinematics during stair ascent/decent and walking compared to a reference motion capture system. A negative peak was observed during level walking and stair descent. The system returned a relative error of <8% in identifying periods of stair ascent

∗ Corresponding author. Tel.: +44 191 248 1245. E-mail addresses: [email protected], [email protected] (A. Godfrey).

during daily activities involving locomotion and postural changes [11]. However, the use of this system incorporated a complex wavelet approximation algorithm and while only one sensor (gyroscope) was used its power consumption was still quite substantial at 4 mA, which limited its recording period to 12-h. Another system incorporating gyroscopes for fall detection have used multiple gyroscope sensor configurations with individual power consumption of up to 6 mA (milliampere) each [12]. Previous ambulatory systems utilising just accelerometers have also been researched [13–15]. Karantonis et al. [14] adopted the use of a single tri-axial accelerometer at the waist for the detection between periods of activity and rest, postural orientation, walking and falls. The authors also aimed to provide a distinction between the upright postures of sitting and standing based on tilt angle differences [14]. However, this method is based on predetermined thresholds and as a result Karantonis et al. found that subjects could either be sitting or standing, depending on various other parameters (the type of accelerometers used by these authors had high power consumption, >4 mA). Laboratory testing carried out by these authors involved six subjects resulted in an overall accuracy of 90.8% across 12 tasks (283 tests) across a variety of movements [14]. Godfrey et al. [16] undertook a comprehensive

1350-4533/$ – see front matter. Crown Copyright © 2011 Published by Elsevier Ltd on behalf of IPEM. All rights reserved. doi:10.1016/j.medengphy.2011.05.002

1128

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

Fig. 1. Site of sensor attachment on the chest.

review of accelerometry and its application for human movement and activity detection. The authors direct the readers to that body of work where earlier systems and their description, accuracy, etc. are presented.

Recently, research has been carried out into investigating if accurate mobility monitoring and activity classification, including posture transition (without the need for thresholds) can be achieved using a single chest mounted sensor [17]. The sensor used in that study by Najafi et al. [17] consisted of a bi-axial accelerometer (current consumption 0.6 mA × 1) and single axis gyroscope (4.5 mA). The authors achieved high accuracies but the resulting activity algorithm was complex and computationally intensive. The aim of this pilot study is to determine, if using a single trunk mounted low power tri-axial accelerometer, can sufficient and accurate detection of postural activity (PA) and posture transition (PT) be achieved from a less complex activity algorithm. Once these have been detected, PTs will help in determining the type of activity and also better assist in understanding the problems occurring during daily activity (falls, postural stability, etc.) [17]. The improved arrangement would: (a) provide a more straightforward, low power sensor array (one tri-axial accelerometer) and (b) a reduction in the computational intensity of the activity classification algorithm.

Fig. 2. A flow chart summarizing the VESPA algorithm. This new algorithm eliminates the need for complex computations on gyroscope signals to eliminate drift introduced by integration. Simple vertical velocity estimates and scalar products provide a complete activity monitoring system on a single accelerometer-based sensor.

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

1129

Fig. 3. (i) sin() after applying the DWT with decomposition into 9 scales by a fifth order Coiflet wavelet (coif5). (ii) Original vertical acceleration signal (avs ). (a, iii) DWT(avs ) between the scales of 5 and 9 with a coif5 wavelet. The circles (green) represent the detection of the type of transition, here it is a SiSt. The nearest maximum and minimum to the time of posture transition tPT . (b, iii) Similar analysis/technique as (a, iii) but here showing a StSi. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

2. Methods 2.1. Sensor design In this study, the sagittal, frontal and transverse accelerations as well as pitch, yaw and roll angular velocities of the trunk, of each

subject were recorded during each activity. The inertial sensor consists of a tri-axial accelerometer sensor, consisting of two bi-axial Analog Devices ADXL2101 accelerometers which are sensitive to

1

ADXL210 and ADXRS300, Analog Devices BV, Limerick, Ireland.

1130

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

both static and dynamic acceleration with a range of ±10 g and consumes less than 0.6 mA each. The inertial sensor also included a tri-axial gyroscope sensor consisting of 3 uni-axial Analog Devices ADXSRS3001 rate gyroscopes. The ADXRS3001 is capable of measuring angular velocities in the range ±300◦ s−1 and provides an analog output voltage with a current consumption of 6 mA. This inertial sensor was positioned on the sternum, Fig. 1. A portable battery-powered data-logger (Biomedical MonitoringBM422 ) was used for data acquisition and attached at the subject’s waist by means of a belt attachment. The sensor signals were recorded at a frequency of 1000 Hz (hertz) and resolution of 12 bits. Each signal was low-pass filtered using a second-order low-pass Butterworth two-pass digital filter, with a cut-off frequency of 17-Hz. The sensor arrangement (sensor held both accelerometers and gyroscopes) allowed for more efficient data collection. Once the scripted activities were performed, accelerometer and gyroscope data from the sensor were downloaded from the data logger and analysed with a specially written Matlab® program. The accelerometer data collected were to be analysed separately by both methods (newly proposed algorithm and that used by Najafi et al. [17]) for direct comparison. The following section describes in more detail how the data were collected and analysed. 2.2. Experimental design The experimental design consists of two stages: Firstly, the chest-mounted sensor was evaluated on young healthy subjects (YHS). This involved the testing and comparison of a new single tri-axial accelerometer activity algorithm to a proven method of activity and postural transition detection using accelerometers and gyroscopes [17]. Previously, the proven (reference) method, performed over 349 PTs, achieved a sensitivity3 and specificity4 of 93% and 82% for sit-to-stand (SiSt) and 82% and 94% for stand-to-sit (StSi) transitions, respectively. Sensitivities and specificities were, respectively, 90.2% and 93.4% in sitting, 92.2% and 92.1% in “standing + walking” and finally 98.4% and 99.7% in lying. Secondly, once the new technique was compared against the reference method, the single tri-axial accelerometer sensor and new algorithm was tested on elderly healthy subjects (EHS) in their home environment. Stage 1: Ten YHS were recruited. These young adults ranged in age from 21 to 29 years (23.7 ± 2.2 years). All gave written informed consent and the University of Limerick Research Ethics Committee (ULREC) approved the protocol. All subjects wore an ambulatory recording system that included the kinematic sensor attached to the chest. Each subject performed 8 different activities of daily living (ADL), where each was repeated a total of 3 times at their own pace, under the supervision of one of the study investigators. These activities included: • a1: sitting down and standing up from an arm-chair, (height 42.6 ± 1.1 cm); • a2: sitting down and standing up from a kitchen chair, (height 46.2 ± 1.0 cm); • a3: sitting down and standing up from a toilet seat, (height 43.0 ± 0.8 cm); • a4: sitting down and standing up from a low stool, (height 39.2 ± 1.5 cm); • a5: getting in and out of a car seat, (height 52.0 ± 1.7 cm);

2 3 4

Biomedical Monitoring Ltd., Glasgow, Scotland. Measures proportion of actual positives which are correctly identified. Measures proportion of negatives which are correctly identified.

Fig. 4. (a) DWT of the vertical acceleration (avs ), the green ‘’ represents the peaks corresponding to actual walking steps. The red ‘’ corresponds to peaks that were excluded, as they were less than the timing criteria as outlined by Najafi et al. (<0.25 s) to be considered as steps. (b) Vertical velocity estimate obtained during a period of walking for a YHS. The green ‘’ represents the peaks corresponding to actual walking steps. No spurious peaks were detected. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

• a6: sitting down on and standing up from a bed, (height 53.5 ± 1.8 cm); • lying down and standing up from a bed, (height 53.5 ± 1.8 cm); • walking 10 m. These activities were to be detected using a previously validated method [17] and the new classification algorithm (VESPA), as proposed in this paper. The postural transition duration time for each transition was also calculated and compared for the two methods within this age group (results Fig. 6(a) and (b)). Stage 2: Upon the completion of Stage 1, ten EHS performed the same ADL procedures as the young adults, under the supervision of one of the study investigators, in their own homes. These were community-dwelling elderly subjects, 3 female and 7 male, were monitored. They ranged in age from 70 to 83 years (77.2 ± 4.3 years). All subjects gave written informed consent and the ULREC approved the measurement protocol. However, these elderly subjects wore only the kinematic sensor equipped with a single tri-axial accelerometer (ADXL210 × 2). The purpose of this stage, is to show that the validity of the new Velocity Estimate

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

1131

Fig. 5. (i) This the calculation of  from the scalar (dot) product (a• b) of the static vector, a, and the tri-axial accelerometer output, b, for change in body tilt – SiSt or StSi. (ii) Original vertical acceleration signal (avs ). (a) SiSt transition for a YHS – for comparison with Fig. 3(a) – shows the potential of the new VESPA algorithm over the previous method. (iii) Vertical estimate for the transition of SiSt – greater maximum positive peak in vertical estimate, |b| > |a| (a and b determined from t(PT) – nearest maxima). (b) StSi transition for a YHS – for comparison with Fig. 3(b). (iii) Vertical estimate for the transition of StSi – greater maximum negative peak in vertical estimate, |a| > |b| (a and b determined from t(PT) – nearest maxima).

1132

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

Fig. 6. (i) Posture duration (TD) times standing to sitting (StSi) for young healthy subjects as determined by both methods for PTs (mean with standard error). (ii) Posture duration (TD) times for sitting to standing (SiSt) young healthy subjects as determined by both methods for PTs (mean with standard error). (iii) Standing to sitting (StSi) and sitting to standing transition (SiSt) duration times for elderly healthy subjects as determined by the VESPA algorithm (mean with standard error).

and Scalar Product Activity (VESPA) algorithm on EHS where their movement or postural transition signatures (pattern) may differ from YHS.

performed by subjecting the accelerometer to a number of known static angles versus gravity [19]. 2.4. Signal processing

2.3. Calibration Calibration of the tri-axial accelerometer and gyroscope sensors was performed using previous methods as outlined by Bourke et al. [4]. The method of gyroscope calibration works by comparing the standard deviation of three gyroscope signals to a 3-element Gyroscope Static Threshold Vector (GSTV). The GSTV elements are the mean and standard deviation of each axis of the tri-axial gyroscope held static for a period of 5 s [18]. Accelerometer calibration is

The methods used to determine the subject’s activity were the use of multi-resolution analysis in the form of the discrete wavelet transform (for comparison purposes) and the new VESPA algorithm. 2.4.1. The discrete wavelet transform The purpose of the discrete wavelet transform (DWT) is a time-frequency representation of a signal. The DWT owes its

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

1133

Table 1 Sensitivity and specificity of PT and walking for the 10 YHS and 10 EHS using VESPA and the method by Najafi et al. (mean ± standard deviation). VESPA

Total PT

YHS (N = 10) EHS (N = 10) Najafi et al. [17] YHS (N = 6) EHS (N = 9)

Sensitivity (%)

Specificity (%)

StSi

SiSt

Lying

Walking

StSi

SiSt

42 ± 0 42 ± 0

92 ± 9 89 ± 8

85 ± 11 83 ± 11

100 ± 0 100 ± 0

100 ± 0 98 ± 1

85 ± 11 83 ± 11

92 ± 9 89 ± 8

57 ± 9 Not reported

82 ± 15

93 ± 7

100 ± 0

96 ± 1

94 ± 6

82 ± 15

functionality to the fast pyramid algorithm [20,21]. The pyramid algorithm has both forward and backward (inverse) algorithms to compute the wavelet transform. The backward algorithm reconstructs the original signal from the component wavelets [21]. The DWT is given in Eq. (1) in terms of its recovery transform, where d(k,l) is a sampling of the wavelet coefficients at discrete points k and l with the mother wavelet, [20]: x(t) =

∞ ∞  

d(k, )2−k/2 (2−k t − )

(1)

k=−∞=−∞

Here we apply the same techniques adopted by Najafi et al. to validate the new algorithm of activity monitoring with the use of a single tri-axial accelerometer – VESPA. Use of the DWT by Najafi et al. was two-fold: (a) Firstly, DWT (with a suitable mother wavelet between the required frequency bands/scales) eliminated low frequency drift which resulted when the gyroscope signal was integrated to determine change of trunk tilt (). (b) Secondly, the DWT was applied to the vertical acceleration (avs ) to determine the type of postural transition (StSi or SiSt) transition. A combination of these techniques subsequently determined the time, t(PT), and duration of the postural transitions, TD [17]. The wavelet toolbox of Matlab® was used to calculate the different wavelet transforms used in this study. 2.4.2. VESPA algorithm – velocity estimate It has previously been suggested by Degen et al. [22] and Bourke et al. [23] that velocity estimates provide an approximation most similar to velocity profiles. This can be derived from a tri-axial accelerometer (TA) by numerical integration of the norm of the TA signals after the magnitude of static acceleration (gravity) is subtracted, Eq. (2):

 

vve =



a2x

+ a2y

+ a2z

− 9.81 ms

−2

dt

(2)

Degen et al. [22] removed drift from the signal by applying a multiplication factor of 0.9 (damping factor) to the positive acceleration values during integration. Kangas et al. [13] removed drift by subtracting a high-pass filtered root-sum-of-squares signal prior to integration and only integrating over a short period (the pit before the impact). However, Bourke et al. [23] removed integration drift by band-pass filtering the vertical profiles using a 2nd order Butterworth band-pass filter with upper and lower cut-off frequency of 15 Hz and 0.15 Hz respectively as it produced more accurate results. Similarly, this method is adopted in this study. The vertical velocity estimate (vve ) was used to detect walking and to differentiate between the postural transition of StSi or SiSt. If the absolute value of the negative peak around t(PT) was greater than the absolute value of the positive peak, the transition was deemed to be StSi. The opposite was the case for SiSt, see Fig. 5(a) and (b).

2.4.3. VESPA algorithm – scalar product This paper proposes a new method for trunk angle tilt estimation using a tri-axial accelerometer mounted at the chest. This novel method makes use of the scalar (dot) product. This method deals with vector multiplication as is defined by Eq. (3), where a, is a row vector and b is a column vector of equal length.

⎡

⎤

b1 n  .. ⎦ ⎣ = al bl = a1 b1 + · · · + an bn a · b = [a1 · · ·an ] . l=1 bn

(3)

For angle tilt detection the vector a, is a 1 × 3 matrix representing the mean output value from each axis of the tri-axial accelerometer during which the patient is standing and inactive. Similarly, b, is a 3 × N matrix, which represents the accelerometer output for each axis, where N denotes the length of the accelerometer signal, over the entire recording period. The scalar product of a and b is represented in Eq. (4). Thus, the resulting vector, or angle tilt (), is a 1 × N matrix represented in Eq. (5). The value represents the angle difference away from the static vector, a:



a · bN = [a1 , a2 , a3 ]

 =

3 

al bln · · ·

l=1

N = cos−1

3  l=1

b11 b21 b31

b12 b22 b32

··· ··· ···

b1N b2N b3N



N al blN

(4) n=1

(a · bN )



|a| · bN

(5)

The scalar product of the accelerometer signal results in the change of trunk tilt of the wearer. This method was compared directly with the method adopted by Najafi et al. [17] for activity classification. A full representation of the VESPA algorithm is presented in Fig. 2. It is proposed that the use of this scalar product method will provide a more simple and straightforward activity algorithm for the detection of sitting to standing and standing to sitting transitions. The added benefit is that this algorithm can be used on inertial sensors that incorporate a single tri-axial accelerometer, thus eliminating the need for gyroscopes. Practically this leads to a simple sensor design and will prolong the activity measurement capabilities of inertial sensors as gyroscopes draw more power from batteries. 3. Results 3.1. Stage 1 (YHS): discrete wavelet transform validation Lying was detected by the orientation of the vertical accelerometer axis (avs ) with a sensitivity of 100%. In the lying state the accelerometer axis measures 0g (g = gravity, 9.81 ms−2 ) while in the standing or sitting state the value is approximately –1g. Fig. 3(a) and (b) shows the resulting postural transition signals achieved from

1134

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

Table 2 Step count for a chest mounted device determined by (a) DWT method (Najafi et al.) applied to the vertical acceleration (avs ) and (b) VESPA derived step count from the vertical estimate. Subject

DWT(avs ): YHSa

VESPA: YHSa

VESPA: EHSb

1 2 3 4 5 6 7 8 9 10 Total

23 28 32 35 38 25 30 27 24 35 297

16 26 26 33 35 25 28 31 25 34 279

26 27 21 24 11 23 15 51 29 26 253

a

YHS = Young Healthy Subjects. EHS = Elderly Healthy Subjects – this was a self-selected distance chosen by the patient. b

standing up (SiSt transition, Fig. 3(a)) and sitting down (StSi transition, Fig. 3(b)) from a kitchen chair for a young healthy subject by the application of the DWT and methodology adopted by Najafi et al. As discussed by Najafi et al. the nearest positive and negative peaks of DWT(avs ) to the local minimum point of sin() were chosen to represent the transition of StSi and SiSt. In total there were 360 PTs performed by the study subjects with a sensitivity and specificity of 92% and 88% for StSi and 89% and 93% for SiSt respectively (these findings were similar to that achieved by Najafi et al.). Walking was detected by applying the DWT to avs between the scales of 2 and 5 using a Daubechies mother wavelet of order 4 (db4) [24]. A threshold was then applied to the negative peaks of the signal, Fig. 4(a) [17]. Walking was correctly identified by this method with an accuracy of 100%. The total number of steps detected is presented in Table 2. Posture transition durations (TD) were also calculated from sin() as a measure of time between the two successive peaks, P1 and P2 , around the time of posture transition (tPT ), Eq. (6) [25]: TD = t(P1 ) − t(P2 )

(6)

The posture transition duration times for both methods (Najafi method and VESPA) are presented in Figs. 3 and 5. 3.2. Stage 1 (YHS): VESPA Lying was detected by the VESPA algorithm from the angle measured (trunk tilt, ) by the tri-axial accelerometer of the vertical accelerometer axis (avs ) with a sensitivity of 100%. Fig. 5(a) (SiSt) and (b) (StSi) shows the resulting postural detection method, similar to that adopted by Najafi et al. but applied to the scalar product method. However, in this study both types of transitions were distinguished using vve . This was achieved by examining the maximum positive and negative peak values of the vve around the time of a postural transition (t(PT)). A greater maximum positive peak value with a SiSt transition (Fig. 5(a, part iii)) and a greater maximum negative peak value was associated with a StSi transition (Fig. 5(b, part iii)). Of the 360 PTs performed by the young adults, sensitivity and specificity were 92% and 86% for StSi and 86% and 92% for SiSt respectively for VESPA. Table 1 shows the overall sensitivity and specificity of transition detection for the 10 YHS. Walking was detected with an accuracy of 100% by examining the negative peaks of the vertical velocity estimate, Fig. 4(b). The step estimation is compared to the method adopted by Najafi et al. in Table 2. Postural duration times were also calculated for VESPA based on similar peak detection during the time of a detected

posture transition (these duration times are presented in Fig. 6(a) and (b)).

3.3. Stage 2 (EHS): application of VESPA in a home environment Having obtained satisfactory results in Stage 1, the kinematic sensor with a single tri-axial accelerometer was applied to an elderly healthy group within their home environment for concept testing. Here the elderly subjects performed the same activities as the young subjects. Lying was detected with a sensitivity of 100%. StSi was detected with a sensitivity and specificity of 89% and 83%, while SiSt had a sensitivity and specificity of 83% and 89% respectively. Table 1 shows the overall sensitivity and specificity of transition detection for the 10 EHS. Walking was identified with an accuracy of 100% over an unmeasured controlled distance and the total number of steps recorded, Table 2. Posture duration times are presented in Fig. 6(c).

4. Discussion and conclusion Previous activity classification studies have focused on the multiple sensor arrangements but this often involved more complex algorithm development to determine the correct activity. More recent studies have incorporated single site sensor attachment, however, those sensors incorporated more high-power devices, (such as gyroscopes) thereby limiting the monitoring duration capability and offline analysis of the study (due to complex algorithms). Clinically, the application of body worn sensors without the possibility of long term patient monitoring can be limiting due to the added burden/cost placed upon patients and researchers to replace sensors and/or batteries. While modern day sensors with a single site of body attachment can record for up to seven or more days, their activity detection algorithms can be minimal due to their need to save battery life. The aim of this pilot study was to show that a suitable low-power device with a more complete and uncomplicated activity classification algorithm was possible with a single chest worn tri-axial accelerometer device. The single tri-axial accelerometer kinematic sensor and simplified activity algorithm was compared directly with a proven sensor configuration worn on the chest (incorporating a bi-axial accelerometer and uni-axial gyroscope) that utilised a more complicated activity algorithm. Firstly, the proven activity detection technique was replicated successfully and was found to have sufficient accuracy with the test group, young healthy adults (YHS) for the activities performed, Fig. 3. The activities of lying, standing, sitting, walking, and postural transitions (StSi and SiSt) were detected with success rates similar to those of Najafi et al. [17]. The new sensor arrangement and simplified algorithm (classified as VESPA) were then applied to the same activities as performed by the YHS. The data is presented in Figs. 4(b) and 5 for direct comparison to Figs. 3 and 4(a) which show the potential for the new system configuration and activity algorithm. The activity of lying and walking was detected with an accuracy of 100% under control conditions. This is an improvement in walking detection over the method used by Najafi et el. where spurious peak detection may have resulted in misclassification (Fig. 4(a)). Importantly, the activities of walking and step counting were more clearly identifiable using VESPA, with the elimination of spurious peaks (Table 1 and Fig. 4) when compared to the method adopted by Najafi et al. [17]. This new kinematic sensor and simplified activity algorithm technique was then applied to data recorded from an elderly group in their own home environment. The same activities as performed by the EHS were also identified successfully with a sensitivity and specificity of 89% and 83%, while SiSt had a sensitivity and specificity of 83%

A. Godfrey et al. / Medical Engineering & Physics 33 (2011) 1127–1135

and 89% respectively. Posture duration transition times were also similar between the two systems for both age groups examined. Limitations of the current study include a 1-s calibration before the performance of each activity. This required the participants to remain still for at least 1-s to provide the quantities required by the calibration technique [4]. There is also a lack of continuous, long-term ADL for both groups and accurate recordings for at home testing. Future work based upon the techniques developed and presented here will involve the recruitment of a larger cohort within these groups, over an extended recording period (7–14 days) to test the algorithm under continuous ambulatory monitoring. The proposed low-power (<1.2 mA) sensor (single tri-axial accelerometer) and new activity algorithm (VESPA) while still at an early stage of development has a number of advantages over previous methods: (a) Simple and easy to use activity algorithm and as a result the new VESPA algorithm is less processor intensive. (b) Use of a low-power body worn sensor (single tri-axial accelerometer on the chest) to improve long term ambulatory monitoring. CAALYX (Complete Ambient Assisted Living Experiment) is an Integrated Project supported by the European Community under the Sixth Framework Programme (IST-2006-045215). Conflict of interest The authors wish to acknowledge the assistance of the volunteers, Analog Devices, BV for providing the ADXL210 accelerometers and ADXRS300 gyroscopes and the CAALYX FP6 project caalyx.eu [26], for their financial support. References [1] Coleman KJ, Saelens BE, Wiedrich-Smith MD, Finn JD, Epstein LH. Relationships between TriTrac-R3D vectors, heart rate, and self-report in obese children. Med Sci Sports Exerc 1997;29(11):1535–42. [2] Clarke-Moloney M, Godfrey A, O’Connor V, Meagher H, Burke PE, Kavanagh EG, et al. Mobility in patients with venous leg ulceration. Eur J Vasc Endovasc Surg 2007;33(4):488–93. [3] Bourke AK, O’Brien JV, Lyons GM. Evaluation of a threshold-based tri-axial accelerometer fall detection algorithm. Gait Posture 2007;26(2):194–9. [4] Bourke AK, O’Donovan KJ, Ólaighin GM. The identification of vertical velocity profiles using an inertial sensor to investigate pre-impact detection of falls. Med Eng Phys 2008;30(7):937–46. [5] Moore ST, MacDougall HG, Gracies JM, Cohen HS, Ondo WG. Long-term monitoring of gait in Parkinson’s disease. Gait Posture 2007;26(2):200–7.

1135

[6] Culhane KM, Lyons GM, Hilton D, Grace PA, Lyons D. Long-term mobility monitoring of older adults using accelerometers in a clinical environment. Clin Rehabil 2004;18(3):335–43. [7] Grant M, Dall P, Granat M. The feasibility of measuring activity patterns of women with osteoporosis. In: Procedings of the 4th world congress of the international society of physical medicine and rehabilitation. 2007. [8] Ryan C, Gray H, Newton M, Granat MH. An investigation of the effects of psychological distress on physical activity levels in individuals with chronic lower back pain. In: Procedings of the 4th world congress of the international society of physical medicine and rehabilitation. 2007. [9] Leonard M, Godfrey A, Silberhorn M, Conroy M, Donnelly S, Meagher D, et al. Motion analysis in delirium: a novel method of clarifying motoric subtypes. Neurocase 2007;13(4):272–7. [10] Culhane KM, O’Connor M, Lyons D, Lyons GM. Accelerometers in rehabilitation medicine for older adults. Age Ageing 2005;34(6):556–60. [11] Coley B, Najafi B, Paraschiv-Ionescu A, Aminian K. Stair climbing detection during daily physical activity using a miniature gyroscope. Gait Posture 2005;22(4):287–94. [12] Nyan MN, Tay FE, Tan AW, Seah KH. Distinguishing fall activities from normal activities by angular rate characteristics and high-speed camera characterization. Med Eng Phys 2006;28(8):842–9. [13] Kangas M, Konttila A, Lindgren P, Winblad I, Jamsa T. Comparison of lowcomplexity fall detection algorithms for body attached accelerometers. Gait Posture 2008;28(2):285–91. [14] Karantonis DM, Narayanan MR, Mathie M, Lovell NH, Celler BG. Implementation of a real-time human movement classifier using a triaxial accelerometer for ambulatory monitoring. IEEE Trans Inf Technol Biomed 2006;10(1):156–67. [15] Lyons GM, Culhane KM, Hilton D, Grace PA, Lyons D. A description of an accelerometer-based mobility monitoring technique. Med Eng Phys 2005;27(6):497–504. [16] Godfrey A, Conway R, Meagher D, Ólaighin GM. Direct measurement of human movement by accelerometry. Med Eng Phys 2008;30(10):1364–86. [17] Najafi B, Aminian K, Paraschiv-Ionescu A, Loew F, Bula CJ, Robert P. Ambulatory system for human motion analysis using a kinematic sensor: monitoring of daily physical activity in the elderly. IEEE Trans Biomed Eng 2003;50(6):711–23. [18] Ferraris F, Grimaldi U, Parvis M. Procedure for effortless in-field calibration of three-axis rate gyros and accelerometers. Sens Mater 1995;7(5): 311–30. [19] Lötters J, Schipper J, Veltink PH, Olthuis W, Bergveld P. Procedure for in-use calibration of triaxial accelerometers in medical applications. Sens Actuators A: Phys 1998;68:221–8. [20] Semmlow J. Biosignal and biomedical image processing: matlab based applications. New York: Marcel Dekker; 2004. [21] Bruce A, Donoho D, Gao H-Y. Wavelet analysis. IEEE Spectr 1996;33(10):26–35. [22] Degen T, Jaeckel H, Rufer M, Wyss S. SPEEDY: a fall detector in a wrist watch. In: Proceedings of the seventh IEEE international symposium on wearable computers, 2003. 2003. [23] Bourke AK, O’Donovan KJ, Nelson J, Ólaighin GM. Fall-detection through vertical velocity thresholding using a tri-axial accelerometer characterized using an optical motion-capture system. Conf Proc IEEE Eng Med Biol Soc 2008:2832–5. [24] Daubechies I. Ten lectures on wavelets (CBMS-NSF regional conference series in applied mathematics). Philadelphia, PA, USA: Society for Industrial and Applied Mathematics; 1992. [25] Najafi B, Aminian K, Loew F, Blanc Y, Robert PA. 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 2002;49(8):843–51. [26] Boulos MNK, Rocha A, Martins A, Vicente ME, Bolz A, Feld R, et al. CAALYX: a new generation of location-based services in healthcare. Int J Health Geogr 2007;6(9):1–6.