dc.description.abstract | Many studies have indicated that moderate-intensity walking is highly beneficial for the body. Consequently, many people set daily step or time goals for walking. Accurate step counting and reducing inaccuracies are highly valuable for users in devising appropriate exercise plans and achieving their physical health objectives.
This study proposes the Adaptive Autocorrelation Step Counting (APASC) algorithm, which is a real-time computational algorithm. The algorithm utilizes autocorrelation functions to infer the periodicity and regularity of signals. It further assesses the oscillation range, period, regularity, and continuity of acceleration values to determine whether a signal corresponds to a walking state. Subsequently, the algorithm calculates the step count within the identified regions classified as walking states. Additionally, the algorithm is capable of measuring the duration of walking exercises at different intensity levels, providing users with valuable references.
In this study, human movements were categorized into four types: normal walking, non-normal walking, irregular non-walking, and regular non-walking. The experimental dataset had a total duration of approximately 22 hours, comprising 510 data sets. These data were utilized to modify and validate the algorithm, aiming to improve its accuracy across these four experimental categories.
Finally, the APASC algorithm proposed in this study achieved an error rate of 1.62% in the normal walking experiment and 11.16% in the non-normal walking experiment. For the irregular non-walking experiment, which lasted approximately 13.75 hours, the algorithm overestimated the step count by 290 steps. For the regular non-walking experiment, which lasted approximately 1.5 hours, the algorithm overestimated the step count by 2460 steps. | en_US |