An improved step counting algorithm using classification and double autocorrelation

摘要

The objective of this paper was to develop an end-to-end algorithm that would improve the step counting accuracy in regular walking/running data and also meet the ANSI/CTA-2056 standards. The ANSI/CTA-2056 standards are to achieve an error rate of less than 10 on treadmill data on at least 20 participants. Our UWF-algorithm (UWFv1) has an improved step counting accuracy and also performs well below the acceptable ANSI/CTA-2056 error rate, using both treadmill data and non-treadmill data, hence our UWFv1 algorithm also meets the ANSI-CTA-2056 standards. For the end-to-end algorithm, the random forest model, trained on a feature engineered dataset, was chosen as the walking/running detection classifier, and double autocorrelation was recommended in the process of determining the step counts.