Journal Logo


Accuracy of Posture Allocation Algorithms for Thigh- and Waist-Worn Accelerometers


Author Information
Medicine & Science in Sports & Exercise: June 2016 - Volume 48 - Issue 6 - p 1085-1090
doi: 10.1249/MSS.0000000000000865
  • Free


Sedentary behavior, defined as sitting or reclining with low energy expenditure during waking hours (22), has consistently been associated with morbidity and mortality (4,5,10–12,23,27,29) in adults. However, the majority of epidemiological studies to date have used either self-reported sedentary behavior measures or objective measures that infer sedentary behavior through lack of movement (5,29). Self-report questionnaires that assess sedentary behavior have consistently demonstrated poor validity and underestimate sedentary behavior (1). Objective measures that infer sedentary behavior through lack of movement may overestimate sedentary behavior (i.e., because of upright activities with very limited ambulation being recorded as sedentary) (18). A key factor in furthering our knowledge on sedentary behavior and health, levels, patterns, and determinants of sedentary behavior, and the effectiveness of sedentary behavior interventions is to use objective devices that directly measure the posture of sitting and distinguish between sitting and upright postures with limited movement (e.g., standing). This is important given that recent experimental research has demonstrated that even light activity such as standing still can have a positive effect on markers of health (6,16,26).

Three devices that are capable of postural classification are the activPAL (all models), the thigh-worn GENEActiv, and the ActiGraph (when worn on the waist or thigh). The activPAL and ActiGraph are small triaxial accelerometers that provide information on body posture (i.e., lying, sitting, and upright postures such as standing and stepping) using proprietary software algorithms created by the manufacturers. Alternatively, an open-source algorithm is available based on relative values of the x, y, and z vectors, which can be applied to raw acceleration data from a thigh-worn triaxial accelerometer to provide lying, sitting, and standing information (20). A key recommendation from the 2009 Objective Measurement of Physical Activity Meeting, cosponsored by the National Institutes of Health and the American College of Sports Medicine, was that monitor data should be collected and saved as raw signals, with data transformation carried out postprocessing to facilitate comparisons between output regardless of which monitor is used (2,7,9,28). This is only possible if open-source algorithms are available for data processing. The open-source algorithm for classifying posture from a thigh-worn monitor was initially developed by Activinsights (Activinsights Ltd., Cambridgeshire, UK) for the GENEActiv; to date, it has been validated using GENEActiv data, but not with data from other devices (20).

The activPAL device has been extensively validated in both laboratory and free-living studies (14,15,17–19,21); however, very little research has been published on the validity of the waist (3,8) and thigh-worn (24,25) ActiGraph inclinometer algorithm and the thigh-worn GENEActiv (20). Furthermore, the majority of validation studies, including those with the activPAL, have usually involved lying and sitting activities that are not fully representative of daily postures. For example, lying in daily life usually involves lying on the back or side with legs sometimes straight and sometimes bent. Sitting usually involves different leg positions such as crossed legs or tucked under a chair. Studies to date have not considered these types of activities in their validation methods. One exception is the recently published study by Steeves et al. (25) where participants wore the activPAL and the ActiGraph on the thigh while completing sitting activities with different leg positions (e.g., sitting with legs crossed at the knee). They found that the activPAL and ActiGraph were highly accurate for some (e.g., sitting with legs crossed), but not all (e.g., sitting on a laboratory stool), sitting activities. To expand our understanding of the accuracy of the devices that are capable of postural classification, it is important to include, in validation studies, a wide range of activities that are as representative of daily life as possible.

Therefore, the purpose of this study was to investigate the accuracy of the activPAL, the waist- and thigh-worn ActiGraph GT3X+ proprietary postural allocation algorithms, and the open-source thigh postural allocation algorithm (applied to GENEActiv and ActiGraph data). Accuracy for identifying a range of lying and sitting positions, representative of daily postures, and light-intensity upright activities was examined in a laboratory-based setting. Application of the open-source postural allocation algorithm to both the GENEActiv and ActiGraph data will enable the assessment of the generalizability of the open-source algorithm and comparison of the accuracy of the open-source and ActiGraph proprietary algorithms.



A convenience sample of 34 adults was recruited from Loughborough University and University of Leicester (staff and students) via word of mouth and e-mail. Participants needed to be ≥18 yr, English speaking, and without mobility issues, which would prevent full participation in the protocol of activities. Ethical approval was received from Loughborough University.


Participants visited the research centre at Loughborough University between March 2014 and August 2014. Participants provided written informed consent and basic demographic information (date of birth and sex). Body weight (Tanita, West Drayton, UK) and height (Leicester portable height measure) were measured to the nearest 0.1 kg and 0.5 cm, respectively. Participants were fitted with an activPAL3™, GENEActiv, and ActiGraph GT3X+ on the midline anterior aspect of their right thigh and an ActiGraph GT3X+ on their right hip. Participants were directly observed continuously (criterion measure) while completing a protocol consisting of 16 activities (Fig. 1), each performed for 5 min with a 30-s gap in between activities. Participants started with lying activities, and each participant completed the activities in the same order. The start and stop time for each of the activities was measured and recorded by the observer using the clock function on the same computer used to initialize the devices.

Flow of 16 lying, sitting, and upright activities.

Objective Sedentary and Activity Measures

activPAL3™ is a small (35 × 53 × 7 mm), lightweight (15 g) triaxial accelerometer, and via proprietary algorithms (Intelligent Activity Classification), accelerometer-derived information about thigh position and acceleration are used to determine body posture (i.e., sitting/lying and upright) and transition between these postures and stepping. Default settings were used during initialization (i.e., 20 Hz, 10-s minimum sitting and upright period). The activPAL was attached midline on the anterior aspect of the right thigh using Hypafix medical dressing.

ActiGraph GT3X+ (ActiGraph LLC, Pensacola, FL) is a small (45 × 33 × 15 mm), lightweight (19 g) triaxial accelerometer that can be worn on various body locations including the waist, wrist, ankle, and thigh. Through a proprietary postural algorithm, ActiGraph, when worn on the waist, is capable of describing positional information (lying, sitting, standing, and nonwear) during periods of inactivity because of gravitational forces acting on the orientation on the three axes. When the device is worn on the thigh, the lying and sitting categories are grouped together. ActiGraph devices were initialized to record at a frequency of 100 Hz, and the low frequency extension filter was selected. Participants wore two ActiGraph GT3X+ devices: one on an elastic belt around the waist on the right midaxillary line of the hip and one on an elastic belt on the midline on the anterior aspect of the right thigh (below the activPAL3™).

GENEActiv (Gravity Estimator of Normal Everyday Activity, Activinsights Ltd., Cambridgeshire, UK) is a small (43 × 40 × 13 mm), lightweight (16 g) triaxial accelerometer that can be worn on various body locations including the wrist, waist, ankle, upper arm, and thigh. When worn on the thigh, GENEActiv can assess posture based on the relative values of the x (mediolateral), y (vertical), and z (anteroposterior) vectors. The GENEActiv was initialized to record at a frequency of 100 Hz. Participants wore the GENEActiv on the midline on the anterior aspect of the right thigh using an elastic belt.

Data Reduction and Analysis

Proprietary algorithms

activPAL data were downloaded using activPAL Professional Research Edition v7.2.29 (PAL Technologies, Glasgow, UK), and 15-s epoch CSV files were created. ActiGraph data were downloaded using ActiLife v6.10.2 (ActiGraph, Pensacola, FL) and converted into 15-s epoch CSV files. Postural classification is determined proprietarily within the manufacturer’s software for the thigh-worn activPAL, thigh-worn ActiGraph, and waist-worn ActiGraph (APALPROP and T_AGRAPHPROP, and W_AGRAPHPROP, respectively).

Open-source algorithms

GENEActiv data were downloaded using GENActiv PC software v2.2, and the raw .bin files were converted into 15-s epoch CSV files. The 15-s epoch files were imported into a custom-built template in Excel, which computed the most likely posture based on the relative values of the x, y, and z vectors measured at the thigh (T_GACTIVOPEN). This method was developed by Activinsights for use with the GENEActiv when it is worn on the thigh and has been described previously (20). The method is open source and we have made the Excel template available on the Leicester-Loughborough Diet, Lifestyle and Physical Activity Biomedical Research Unit website (

The 100-Hz GT3X+ files from the thigh-worn ActiGraph were converted to 100-Hz CSV files containing x, y, and z vectors using ActiLife version 6.10.2. In order to match the format to the GENEActiv and to that required for the open-source algorithm, a purpose-built Excel template was used to convert the raw 100 Hz files to 15-s epoch files containing x, y, and z vectors (mean acceleration over the epoch). The 15-s epoch files were then imported into the custom-built Excel template for computation of the most likely posture (T_AGRAPHOPEN). The first and last 30 s of each activity were excluded from the analyses to protect against the potential of imperfect time synchronization and transition between activities.

For each participant, the percentage of epochs that were correctly coded as lying, sitting, and upright against direct observation was calculated for each of the 16 activities for each method of measurement (APALPROP, T_AGRAPHPROP, W_AGRAPHPROP, T_GACTIVOPEN, and T_AGRAPHOPEN). Percentages were then summarized and presented as means and 95% CI for each individual activity and activities grouped as lying, sitting, and upright. Analyses were conducted in IBM SPSS Statistics v20.0.


Participants consisted of 14 males and 20 females (mean age, 27.21 ± 5.94 yr (range, 20–40 yr); mean BMI, 23.82 ± 3.68 kg·m−2; range, 18.64–32.58 kg·m−2). Table 1 presents the mean percentage of time coded correctly, against direct observation, for each individual activity and activities grouped by type (i.e., lying, sitting, and upright) by each measurement method.

Mean (95% CI) percentage of activity correctly coded as lying, sitting, and upright for each activity and method against direct observation.

The APALPROP and T_GACTIVOPEN classified all lying activities correctly 100% of the time. The T_AGRAPHPROP and T_AGRAPHOPEN classified three of the four lying activities 100% of the time, with lying on the back with legs bent classified correctly 73% of the time (93% correctly classified for all lying activities) and 91% of the time (98% correctly classified for all lying activities), respectively. The W_AGRAPHPROP correctly classified lying activities between 67% and 77% of the time (72% overall for lying activities).

When examining sitting activities, the APALPROP correctly classified six out of seven sitting activities ≥97% of the time, with sitting with legs stretched out classified correctly 42% of the time (91% overall for all sitting activities). The T_GACTIVOPEN and T_AGRAPHOPEN correctly classified all sitting activities 100% of the time. The T_AGRAPHPROP correctly classified six out of seven sitting activities 100% of the time; sitting with legs stretched out was classified correctly 95% of the time (99% overall for all sitting activities). The W_AGRAPHPROP correctly classified sitting activities between 46% and 70% of the time (58% overall for sitting activities).

Four out of five upright activities were correctly classified 100% of the time by the APALPROP, with self-paced walking correctly classified 97% of the time (99% overall for all upright activities). The T_GACTIVOPEN, T_AGRAPHOPEN, and the T_AGRAPHPROP correctly classified upright activities ≥88% (93% overall for all upright activities), ≥97% (98% overall for all upright activities), and ≥91% (96% overall for all upright activities) of the time, respectively. The W_AGRAPHPROP correctly classified upright activities between 61% and 97% of the time (74% overall for upright activities).


This study adds to the literature by comparing the accuracy of several accelerometers, with proprietary and/or open-source postural allocation algorithms applied to the data, across a range of different postures and activities. This study demonstrated that all thigh-worn monitors were highly accurate in identifying lying, sitting, and upright postures, irrespective of whether proprietary (activPAL and ActiGraph) or open-source algorithms (GENEActiv and ActiGraph) were applied to the data. As noted recently by Steeves et al. (25), there is a need for improvements in algorithms to increase their ability to correctly classify a wider range of postures and activities. They further highlight that broader access to appropriate hardware and firmware to support postural and activity classifications would be a major advancement for the research community. The open-source algorithm applied in the current study demonstrated high accuracy across monitor brands and across the range of postures and activities typical during free living; this is a significant step forward.

The high validity of the activPAL monitor has been demonstrated in numerous laboratory studies (14,15); however, to our knowledge, this is only the second study using the activPAL including sitting postures with a variety of leg angles. Recently, Steeves et al. examined the accuracy of the activPAL for identifying different sitting postures (e.g., legs crossed at the knee, legs crossed at the ankle, legs crossed with the ankle on the opposite knee) and found that the activPAL was highly accurate for most sitting postures. In agreement with the current study, they found that the activPAL misclassified (15% of the time) sitting with legs outstretched but not to the extent of the current study (58%). This sitting position changes the angle of the thigh slightly (i.e., knee angle increases above 90° and front of thigh dips), and the misclassification suggests that the activPAL proprietary angular parameters for the classification of sitting require the thigh to be close to being parallel to the ground (25). Because the activPAL algorithm is proprietary, it is not possible to investigate whether accuracy can be improved by adjusting the parameters, as would be possible with an open-source algorithm. It is important to acknowledge that the extent to which this would affect misclassification of sitting time during a typical 7-d free-living data collection would depend on the prevalence of this type of sitting posture.

The use of the activPAL monitor in physical activity and sedentary behavior research is increasing rapidly (13) because of its ability to correctly identify posture (14–17). The high accuracy of the ActiGraph thigh proprietary algorithm and open-source algorithm applied to both ActiGraph and GENEActiv data observed in the current study suggests that these could also be an option for postural identification in research. This finding is consistent with a small body of previous research (24,25) that has shown the ActiGraph thigh proprietary algorithm to be highly accurate. Skotte et al. (24) under free-living conditions compared the hip- and thigh-worn ActiGraph postural allocation algorithms against a pressure logger to detect sitting posture. They found that the thigh algorithm was more precise than the hip algorithm. Furthermore, in a recent study by Steeves et al. (25), the ActiGraph thigh algorithm demonstrated 100% accuracy in detecting five different sitting postures, an accurate ability to identify standing and light movement at a whiteboard, and >95% accuracy for stepping activities. The ActiGraph thigh algorithm did however misclassify 14% of the time sitting on a laboratory stool as standing time (25). Although we, and others, have found the ActiGraph thigh algorithm to be highly accurate for the majority of activities, it is important to acknowledge the design limitations of this device. The device, although small and lightweight, is considerably thicker than the activPAL for example and has sharp edges where the elastic belt sits. This may make it visible under some clothing and uncomfortable on the thigh, possibly resulting in compliance issues when worn on the thigh. Ideally, a device needs to be both accurate and comfortable to wear. Before deciding upon a particular device, pilot testing with the target population would be advantageous.

In a small free living study, the accuracy and precision of the open-source algorithm applied to GENEActiv data have been demonstrated against the activPAL monitor (20); however, the current study is the first to compare against direct observation. This is also the first study to apply a transparent open-source algorithm to ActiGraph data and compare it to the manufacturer’s proprietary algorithm. The open-source algorithm applied to the ActiGraph thigh data performed slightly better than the ActiGraph proprietary algorithm for identifying lying and sitting activities, specifically on the individual activities of lying on the back with legs bent and sitting with legs stretched out, but had marginally lower accuracy for upright activities.

Few published studies have investigated the accuracy of the ActiGraph algorithm when worn on the waist (3,8,24). All studies reported poor accuracy of the algorithm, which corroborates the current findings. Given the poor accuracy of the waist algorithm for identifying lying, sitting, and upright activities, caution should be taken when considering using this device in research studies especially those with a focus on time spent sitting.

The strengths of this study include the comparison of five different postural identification measurement methods (including application of an open-source algorithm), the range of lying, sitting, and upright activities that were chosen to be more representative of daily postures, and the use of direct observation as the criterion measure for comparisons. However, it is important to acknowledge that although activities and postures included were designed to mimic everyday behaviors, participants were instructed how to lie or sit and in a free-living environment may perform the same behaviors in a slightly different manner. Furthermore, our homogeneous sample of participants (i.e., narrow age range and 74% in the normal weight category) may limit generalizability of results.

In summary, we demonstrated that all thigh-worn monitors, irrespective of the type (proprietary or open source) of algorithm, were highly accurate. It is important to note that it is not the device or the algorithm per se that is accurate; it is the combination of the two. A major limitation of any proprietary algorithm, in addition to the lack of transparency, is that it is limited to a single device. In contrast, open-source methods are much more flexible for researchers to use (e.g., modifications can be made to angle thresholds for different population groups) and allow algorithms to be applied to different devices enabling assessments across devices to be made. The current study demonstrated accuracy of an open-source algorithm across monitor brands and across a range of postures and activities.

The research was supported by the National Institute for Health Research (NIHR) Diet, Lifestyle and Physical Activity Biomedical Research Unit based at the University Hospitals of Leicester and Loughborough University, the National Institute for Health Research Collaboration for Leadership in Applied Health Research and Care—East Midlands (NIHR CLAHRC-EM), and the Leicester Clinical Trials Unit.

Alex Rowlands provides consultancy services to Activinsights, the manufacturer of the GENEActiv. The authors declare that there are no other conflicts of interests.

C. Edwardson, T. Yates, and A. Rowlands conceived the study and T. Gorely, D. Esliger, M. Davies, and K. Khunti refined the study, C. Edwardson wrote the first draft of the manuscript, S. Bunnewell and J. Sanders carried out data acquisition, S. Bunnewell, S. O’Connell, and A. Rowlands analyzed the data. All authors reviewed/edited the manuscript and approved the final manuscript.

The views expressed are those of the authors and not necessarily those of the National Health Service, the NIHR, or the Department of Health. The results of the present study do not constitute endorsement by the American College of Sports Medicine.


1. Atkin AJ, Gorely T, Clemes SA, et al. Methods of measuring sedentary behaviour in epidemiological research. Int J Epidemiol. 2012; 41(5): 1460–71.
2. Basset DR Jr, Rowlands AV, Trost SG. Calibration and validation of wearable monitors. Med Sci Sports Exerc. 2012; 44(1 Suppl): S32–8.
3. Berendsen BA, Hendriks MR, Meijer K, Plasqui G, Schaper NC, Savelberg HH. Which activity monitor to use? Validity, reproducibility and user friendliness of three activity monitors. BMC Public Health. 2014; 14: 749.
4. Biswas A, Oh PI, Faulkner GE, et al. Sedentary time and its association with risk for disease incidence, mortality, and hospitalization in adults: a systematic review and meta-analysis. Ann Intern Med. 2015; 162: 123–32.
5. Brocklebank LA, Falconer CL, Page AS, Perry R, Cooper AR. Accelerometer-measured sedentary time and cardiometabolic biomarkers: a systematic review. Prev Med. 2015; 76: 92–102.
6. Buckley JP, Mellor DD, Morris M, Joseph F. Standing-based office work shows encouraging signs of attenuating post-prandial glycaemic excursion. Occup Environ Med. 2014; 71: 109–11.
7. Butte NF, Ekelund U, Westerterp KR. Assessing physical activity using wearable monitors: measures of physical activity. Med Sci Sports Exerc. 2012; 44(1 Suppl): S5–12.
8. Carr LJ, Mahar MT. Accuracy of intensity and inclinometer output of three activity monitors for identification of sedentary behavior and light-intensity activity. J Obes. 2012; 2012: 460271.
9. Chen KY, Janz KF, Zhu W, Brychta RJ. Redefining the roles of sensors in objective physical activity monitoring. Med Sci Sports Exerc. 2012; 44(1 Suppl): S13–23.
10. Cong YJ, Gan Y, Sun HL, et al. Association of sedentary behaviour with colon and rectal cancer: a meta-analysis of observational studies. Br J Cancer. 2014; 110: 817–26.
11. de Rezende LF, Rey-López JP, Matsudo VK, do Carmo Luiz O. Sedentary behavior and health outcomes among older adults: a systematic review. BMC Public Health. 2014; 14: 333.
12. Edwardson CL, Gorely T, Davies MJ, et al. Association of sedentary behaviour with metabolic syndrome: a meta-analysis. PLoS One. 2012; 7: e34916.
13. Edwardson CL, Winkler EAH, Bodicoat DH, et al. Considerations when using the activPAL monitor in field based research with adult populations. J Sport Health Sci. In Press.
14. Grant PM, Ryan CG, Tigbe WW, Granat MH. The validation of a novel activity monitor in the measurement of posture and motion during everyday activities. Br J Sports Med. 2006; 40: 992–7.
15. Hart TL, McClain JJ, Tudor-Locke C. Controlled and free-living evaluation of objective measures of sedentary and active behaviours. J Phys Act Health. 2011; 8: 848–57.
16. Henson J, Davies MJ, Bodicoat DH, et al. Breaking up prolonged sitting with standing or light activity reduces post-prandial markers of diabetes risk. Diabetes Care. 2016; 39: 130–8. pii: dc151240. [Epub ahead of print].
17. Kozey-Keadle S, Libertine A, Lyden K, Staudenmayer J, Freedson PS. Validation of wearable monitors for assessing sedentary behavior. Med Sci Sports Exerc. 2011; 43(8): 1561–7.
18. Kozey-Keadle S, Libertine A, Staudenmayer J, Freedson P. The feasibility of reducing and measuring sedentary time among overweight, non-exercising office workers. J Obes. 2012; 2012: 282–303.
19. Lyden K, Kozey-Keadle SL, Staudenmayer JW, Freedson PS. Validity of two wearable monitors to estimate breaks from sedentary time. Med Sci Sports Exerc. 2012; 44(11): 2243–52.
20. Rowlands AV, Olds TS, Hillsdon M, et al. Assessing sedentary behavior with the GENEActiv: introducing the sedentary sphere. Med Sci Sports Exerc. 2014; 46(6): 1235–47.
21. Ryan CG, Grant PM, Tigbe WW, Granat MH. The validity and reliability of a novel activity monitor as a measure of walking. Br J Sports Med. 2006; 40: 779–84.
22. Sedentary Behaviour Research Network. Standardized use of the terms “sedentary” and “sedentary behaviours.” Appl Physiol Nutr Metab. 2012; 37: 540–2.
23. Shen D, Mao W, Liu T, et al. Sedentary behavior and incident cancer: a meta-analysis of prospective studies. PLoS One. 2014; 9: e105709.
24. Skotte J, Korshøj M, Kristiansen J, Hanisch C, Holtermann A. Detection of physical activity types using triaxial accelerometers. J Phys Act Health. 2014; 11(1): 76–84.
25. Steeves JA, Bowles HR, McClain JJ, et al. Ability of thigh-worn ActiGraph and activPAL monitors to classify posture and motion. Med Sci Sports Exerc. 2015; 47(5): 952–9.
26. Thorp AA, Kingwell BA, Sethi P, Hammond L, Owen N, Dunstan DW. Alternating bouts of sitting and standing attenuate postprandial glucose responses. Med Sci Sports Exerc. 2014; 46(11): 2053–61.
27. Thorp AA, Owen N, Neuhaus M, Dunstan DW. Sedentary behaviors and subsequent health outcomes in adults a systematic review of longitudinal studies 1996–2011. Am J Prev Med. 2011; 41: 207–15.
28. Welk GJ, McClain J, Ainsworth BE. Protocols for evaluating equivalency of accelerometry-based activity monitors. Med Sci Sports Exerc. 2012; 44(11): S39–49.
29. Wilmot EG, Edwardson CL, Achana FA, et al. Sedentary time in adults and the association with diabetes, cardiovascular disease and death: systematic review and meta-analysis. Diabetologia. 2012; 55: 2895–905.


© 2016 American College of Sports Medicine