Abstract
Step counters are popular for quantifying walking. However, they may not measure the number of steps taken by elderly persons. In this study, we proposed a step count algorithm based on a filter bank and threshold processing to improve the accuracy of an accelerometer-type step counter for elderly persons. The accuracy of the proposed algorithm was compared with the observed steps taken during walking. The waist acceleration signals during self-paced walking were recorded for 74 attendees (age 82.7±8.6 yr, height 148.1±7.6 cm, weight 52.3±7.8 kg) participating in gait training at a rehabilitation services center using a triaxial accelerometer. The participants walked approximately 20 m (10 m in each direction and a turning arc). After seven different band-pass filters were applied to the magnitude of the acceleration vector, a signal related to the step cycle was selected from the outputs of those filters. Then, the number of steps was estimated from this signal using a predetermined threshold. The percent error was calculated as (estimated steps-observed steps)/observed steps×100. On verifying the algorithm using the data for the 74 elderly subjects, the steps taken by 57 (77.0%) were estimated with less than 10% error. We suggest that our step count algorithm is suitable for estimating the number of steps taken by elderly persons.
