Step counting for slow and intermittent ambulation based on a smartwatch accelerometer

The ambulatory monitoring of human movement can provide valuable information regarding the degree of functional ability and general level of activity of individuals. Since walking is a basic every day movement and an important element to introduce into one's daily routine, automatic step detection or step counting is very important in developing ambulatory monitoring systems. This paper is concerned with the development and the preliminary validation of a step counter (SC) that is especially designed for conditions of slow and intermittent ambulation. The SC was based on processing the accelerometer data measured by a Gear 2 smartwatch using a custom wearable app, named ADAM, running on the Gear 2. A dataset of 8 users, for a total of 80 trials, was used to tune ADAM. Finally, ADAM was compared with the native SC running in the Gear 2 smartwatch, and with the SC implemented in a waist-worn pedometer (Geonaute ONSTEP 400) (dataset of 8 users, for a total of 80 trials). The three SCs performed quite similarly in conditions of normal walking over long paths (1--3% of mean absolute relative error); ADAM outperformed the two other SCs in conditions of slow and intermittent ambulation; the error incurred by ADAM was limited to 5%, significantly lower than errors of 20--30% incurred by the two other SCs.