[home] [Personal Program] [Help]

---

Session: Poster session II
Session starts: Thursday 24 January, 16:00

---

Simone Boerema (Roessingh Research and Development)
Hermie Hermens (Roessingh Research and Development)

Abstract:
INTRODUCTION Accelerometry-based activity sensors are nowadays widely deployed in ambulatory monitoring of physical activity. Field experiments are characterized by very little control over usage of the sensor [1]. A quick and reliable calibration procedure is required [2], focused on determining the relation between the accelerometer output and physical activity across a range of activity levels. The purpose of this study was to design an easy spot-check to monitor the validity and reliability of activity sensors. METHOD Four sensors were securely fastened to a mechanical oscillator (Vibration Exciter, type 4809, Brüel & Kjær) and moved at various frequencies (6.67Hz; 13.45Hz; 19.88Hz) within the range of human physical activity. For each of the three sensor axes, the sensors were simultaneously moved for five minutes per defined frequency. The acceleration of each movement was expressed by its RMS value (RMSinput). Raw sensor output (sample frequency 200 Hz) was converted to IMA counts per minute (cpm, in [ms-2]), according to Bouten et al. (1997) [3]. Linear regression analysis was used to examine the relationship between the RMS and IMA. Factors taken into account were the four sensors and the three axes. The device tested in this study was the ProMove2 (Inertia Technology), containing the 3D MEMS inertial sensor (type: LIS3LV02DL, ST Microsystems). RESULT The RMS output of the sensors was within 6% of the RMSinput at each frequency, indicating a high accuracy of the sensors. There were no significant differences between the sensors in IMA values at 6.67Hz and 13.45Hz. Differences found in IMA values at all frequencies and all the different axes, were not significant when added to the regression model. The resulting regression model is: IMA*10-3 [ms-2] = 783*RMS [ms-2] (R2=0.980), indicating that only RMSinput is significant for the output of the sensors. The tested intensity levels are, compared to normal living, in the range of low intensity activities (e.g. reading a book ±200cpm, compared to ±320cpm at 6.67Hz in the experiment) to moderate intensity activities (e.g. shopping ±1800cpm, compared to ±1740cpm at 19.88Hz). DISCUSSION AND CONCLUSION The developed method provides an easy to perform procedure to test an individual activity sensor as well as to determine the inter-variability of a set of sensors. This is especially useful when multiple sensors are used in field experiments.