利用動作感測器開發自動化功能性動作檢測系統

Develop an Automatic Functional Movement Screen System with IMU Sensor

李孟樺(Meng-Hua Lee)

指導教授 ： 吳汶蘭

高雄醫學大學/醫學院/運動醫學系碩士班/碩士(2020年)

摘要

目的： 本研究利用慣性測量單元(Inertial measurement unit，簡稱IMU)來測量深蹲、跨欄、直線前蹲、筆直抬腿、軀幹穩定俯臥撐、四肢旋轉穩定六項動作中位移角度，希望可藉由開發6軸IMU身體肢段位置動作提供最新的標準化測量，達到更快速、更準確的結果。方法： 本研究招募35名健康成年人(年齡:20-35歲)，以6軸BOOSTFIX IMU傳感器的數據與專業人士評分來評估兩者間的一致性，全身共配戴11顆感應器。每項動作數據會計算關節活動範圍，接著利用逐步回歸方式選擇最佳參數特徵擷取，搭配AdaBoost機器學習演算法為基礎的辨識系統，最後以準確率、F score以及kappa值來顯示研究成果。結果與結論：研究結果發現在進行逐步回歸分析，六項動作準確率為61%-79%，kappa值為0.06-0.47；未經逐步回歸分析的準確率為66%-91%，kappa值為0.18-0.85。本研究在使用機器學習方法來篩選參數及設定每項動作的閾值都比先前研究中以人工設定的閾值進行自動評分更為準確，因此利用機器學來評分的效果會更好。但是由於人體活動的差異甚大，即便本研究有篩選出最佳參數，現階段為止仍建議在測量時使用全身的IMU感測器來為FMS進行評分，以確保可以達到更好的評分效果。結論：本研究的結果顯示IMU感測器開發系統可以作為自動化功能性動作檢測篩選工具之選項。
關鍵詞: 功能性運動檢測、慣性測量單元、即時監測、kappa值、混淆矩陣、逐步回歸分析

關鍵字

功能性運動檢測 ； 慣性測量單元 ； 即時監測 ； kappa值 ； 混淆矩陣 ； 逐步回歸分析

並列摘要

Introduction: The Functional Movement Screen (FMS) is a popular movement screen used by strength and conditioning professionals to identify dysfunction in those at risk of, but not currently experiencing, signs or symptoms of a musculoskeletal injury. FMS screening methods, like other fitness tests, need to be manually tested and scored by professionally trained testers. This study hopes to develop artificial intelligence (AI) expert learning system to automatically judge functional movements. Methods: 30 healthy adults (15 men and 20 women, age range 20-35 yr) were fitted with a full-body IMU system and completed 6 FMS exercises: Deep Squat, Hurdle Step, In-line Lunge, Active Straight Leg Raise, Push up, and Rotary Stability. IMU system recorded full-body kinematics, at the same time, a professional athlete trainer graded each subject according to the standard protocols. Data normalization was performed in range of motion data collected from the IMU's sensors, and for reducing the number of input variables when developing predictive model, stepwise regression was used to select relevant feature subsets. Next, ensemble learning AdaBoost algorithm was used to construct our classifiers. Finally, the accuracy, kappa value, and F score were used for evaluating classification models. Results and Discussion: The model's accuracy of the six exercises using all the data was 66%-91%, and the kappa value was 0.18-0.85. After stepwise feature selection, the model's accuracy of the six exercises decreased to 61%-79%, and the kappa value was 0.06-0.47. It indicated that the retainment of over 100% for the features shows that prediction is better than using the reduced feature set. The prediction accuracy for all tests decrease from 10% to 20% using only the reduced feature set classifier. The results of F scores showed poor classification results in certain score groups with fewer samples. Our recommendations for future study are to collect more data and get more information to improve machine learning performance and achieve a more general predictor. Conclusion: The results of this study indicated that a IMU sensor-based system can potentially as an automation screening tools.

Key words: FMS, IMU sensor, Stepwise regression analysis, Confusion matrix, Kappa

並列關鍵字

FMS ； IMU sensor ； Stepwise regression analysis ； Confusion matrix ； Kappa

參考文獻

1. 王聖文, 台灣運動產業內部市場導向對團隊創造力及組織學習能力影響之研究. 運動知識學報, 2016(13): p. 188-198.

2. 林晉利, 臺灣體適能指導員證照現況及發展. 運動管理, 2017(38): p. 27-46.

3. 陳全壽, 劉宗翰, and 張振崗, 我國體適能政策指標之建議. 運動生理暨體能學報, 2004: p. 1-11.

4. 方進隆, 體適能與全人健康. 中華體育季刊, 1995. 9(3): p. 62-69.

5. Akpa, E.A., et al. Gift: glove for indoor fitness tracking system. in 2018 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops). 2018. IEEE.