Optimization of accelerometers for measuring walking

Abstract

The devastating impact of obesity on global health is without question and it is generally agreed that low levels of physical activity, particularly sitting (i.e., sedentariness), are important in its pathogenesis. Therefore, the measure of physical activity such as walking is vital to its use in the research and clinical milieu.

This study investigated three accelerometry parameters (sampling rate, range, and data depth) on ten healthy subjects (BMI 18–31 kg/m²) walking on a calibrated treadmill at eight speeds (0, 0.95, 1.74, 2.48, 3.22, 4.04, 4.83, and 5.70 km/h) while wearing a three-axis accelerometer on the thigh (Crossbow Technology, San Jose, CA) in order to find an optimal system for the determination of walking speed, as well as a new data analysis strategy using a differentiation of the acceleration values (jerk). Twenty-four sampling rates (2–25 Hz in 1-Hz intervals) and seven acceleration ranges (± 1 g – ± 2.5 g at 0.25-g intervals) were used to create a 24 × 7 factorial design. Data was also truncated from two to seven digits in the mantissa.

This study found that although there is an improvement in walking speed prediction when sampling rate was set above 4 Hz (P<0.0002), there was no further improvement when the sampling rate is set higher. This study found that there is an increase in walking speed prediction accuracy when the range of acceleration is limited to ± 1 g (P<0.0024 for ± 2 g versus ± 1 g). This study found that increasing or decreasing data depth has no impact on walking speed prediction accuracy. Further, this study found that a model based on jerk was accurate at predicting walking speeds (r² 0.9 for all comparisons).

For measuring walking using a sensor on the thigh, there is no significant improvement gained by large sampling rates, data ranges, or data precision. A model based on the time rate of change of acceleration is a valid analysis tool for measuring walking.