Accuracy of step detection using a customized mobile phone app

Mobile phones offer unique opportunities to promote physical activity inexpensively. Phone apps with pedometer functions are widely available, but usually are not tested for accuracy. We developed an iPhone app that uses walking cadence to determine walking intensity, and provides feedback on progress towards government guidelines for moderate and vigorous physical activity. In this study, we tested the accuracy of the app's step-counting algorithm under conditions of varying speed, gradient, and placement. 32 adults (53% female; 29±13 yr) performed six treadmill walking trials at 53, 67 and 80 m/min, at 0% and 5% gradient. iPhones were worn in pouches at the hip and back, and also carried in the pocket. Criterion step counts were subsequently determined by hand-counter using a time-stamped video recording. iPhone step counts were compared to the criterion using repeated measures t-tests (p<.05) and Cohen's d. In the pocket position, steps were significantly and meaningfully over-counted (d=0.5-0.9) in all trials. In the hip and back positions, steps were significantly and meaningfully under-counted at 53 m/min (d=0.3-0.6), but accurately counted at 67 and 80 m/min, at level and 5% gradient (d=0.0-0.1). Similar to traditional pedometers, steps are under-counted by a mobile phone app at slow speeds, but accurately counted at moderate speeds and higher, when worn securely. When carried in the pocket, steps are over-counted regardless of speed and gradient. Further analysis of the raw acceleration signal and the time-stamped video recording will help identify reasons for inaccuracy and inform future signal-processing decisions in mobile phone accelerometer uses