| 摘要: | 許多文獻指出中強度的行走對身體非常有益，因此很多人都會設定每日行走一定的步數或時間目標。若能準確的測量步數，並且降低誤算情況的發生，對於使用者制定合適的運動計劃和達到身體健康目標非常有價值。
本研究提出自適性自相關函數計步(Adaptive Autocorrelation Step Counting, APASC)演算法，該演算法為一即時運算演算法，以自相關函數推算訊號的週期和規律性，並以加速度值振盪範圍、週期、規律性和連續性判斷訊號是否為行走狀態，最後計算被判定為行走狀態區域的步數，同時，演算法也能統計不同強度行走運動的時長，提供使用者參考。
本研究將人的動作分為4個種類，分別為一般行走、非一般行走、不規律非行走和規律非行走，實驗數據總長約22小時，共包含510組數據，運用這些數據對演算法進行修改和驗證，以提高演算法在這4類實驗中的準確度。
最終，本研究提出的APASC演算法在一般行走的實驗中，誤差為1.62%；在非一般行走的實驗中，誤差為11.16%。對於總長約13.75小時的不規律非行走實驗，誤算290步；對於總長約1.5小時的規律非行走實驗，誤算2460步。
;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. |
