本研究主要目的為開發上肢部分動作分析之機械學習數學模型。此動作分析方法是用於解析使用者在進行各項動作時的上部身體姿勢，旨在消除多餘動作及矯正錯誤姿勢，來達到減輕疲勞、避免受傷及身體復健的功效，可用於醫療保健及身體康復。而除了探討臨床議題外，精細動作分析亦為當下運動科學研究之主要範疇。藉由量測、紀錄並分析解構特定運動動作特徵，提供其動作連動性、肌肉使用順序、旋轉角度以及施力狀態等資訊給予教練以及醫生對運動員狀況進行評估。提供運動員各種最適當化的各項肌肉與骨頭之操作指標，以期避免運動傷害並在不傷害身體之前提下增強其運動表現。目前在運動科學中的動作分析研究主要是使用光學動作捕捉系統及最近開始盛行的可穿戴慣性測量單元(Inertial measurement unit，IMU)完成。但上述技術所使用的設備都較為昂貴，且光學動作捕捉系統還需在專門的實驗室環境中操作。因此本研究之主要研究架構建立於降低科學動作分析的設備成本及使用者操作位置與運動範圍限制的需求上。本研究方法主旨在於完成即時性的人體上部運動分析，其中僅使用三個加速度感測器及兩台攝影機完成。架構上使用加速度感測系統偵測人體肩肘的移動及轉動軌跡，同時兩台視野正交的攝影機會捕捉人體特定定位點之動作。而我們發展出利用非線性迴歸模型，先分別將攝影機捕捉的影像動作以及加速度感測數據分別建模。爾後將加速度模型映射至影像模型空間中，以獲得加速度模型與影像模型之同構關係。最後利用特定定位點的物理長度來校正影像空間中的像素距離，因此便可以獲得加速度數據點之間的物理長度關係。因此運用本研究當中的機械學習提供之身體上部分運動模型，可以讓醫生判斷患者的肌肉及骨骼狀況，也可用於提供治療師作為是否為正確的運動姿勢的依據。

The main purpose of this research is to develop a machine learning mathematical model for the analysis of upper limb movements. This action analysis method is used to analyze the upper body posture of the user when performing various actions. It aims to eliminate redundant actions and correct wrong postures to achieve the effect of reducing fatigue, avoiding injury, and physical rehabilitation. It can be used in medical care and physical recovery. In addition to discussing clinical issues, fine motor analysis is also a major area of current sports science research. Measuring, recording, analyzing, and deconstructing the characteristics of specific sports movements, provides information such as movement linkage, muscle use sequence, rotation angle, and state of exertion for coaches and doctors to evaluate the athletes' condition. Athletes can be provided with the most appropriate operating indicators of various muscles and bones to help them avoid sports injuries and enhance their performance within the physiological endurance. At present, motion analysis research in sports science is mainly done using optical motion capture systems and the recently popular wearable inertial measurement unit (IMU). However, the equipment used in the above techniques is relatively expensive, and the optical motion capture system also needs to be operated in a specialized laboratory environment. Therefore, the main research framework of this study is based on the need to reduce the equipment cost of scientific motion analysis and for ameliorating the limitation of the user's operating position and range of motion. The main purpose of this research method is to perform real-time upper body motion analysis using only three accelerometers and two cameras. The acceleration sensing system is used to detect the movement and rotation trajectory of the human shoulder and elbow, and at the same time, two cameras with orthogonal fields of view will capture the movement of the specific positioning point of the human body. We developed a nonlinear regression model to model the image motion captured by the camera and the acceleration sensing data separately. Then, the acceleration model is mapped into the image model space to obtain the isomorphic relationship between the acceleration model and the image model. Finally, the pixel distance in the image space is corrected by the physical length of the specific positioning point, so the physical length relationship between the acceleration data points can be obtained. Therefore, the motion model of the upper body provided by the machine learning in this study can allow doctors to judge the patient's muscle and bone condition, and can also be used to provide therapists with a basis for correct exercise posture.

[1] Lapinski, Michael et al. “A Wide-Range, Wireless Wearable Inertial Motion Sensing System for Capturing Fast Athletic Biomechanics in Overhead Pitching.” Sensors (Basel, Switzerland) vol. 19,17 3637. 21 Aug. 2019, doi:10.3390/s19173637
[2] Vellios, Evan E et al. “Technology Used in the Prevention and Treatment of Shoulder and Elbow Injuries in the Overhead Athlete.” Current reviews in musculoskeletal medicine vol. 13,4 (2020): 472-478. doi:10.1007/s12178-020-09645-9
[3] Rigoni, Michael et al. “Assessment of Shoulder Range of Motion Using a Wireless Inertial Motion Capture Device-A Validation Study.” Sensors (Basel, Switzerland) vol. 19,8 1781. 13 Apr. 2019, doi:10.3390/s19081781
[4] Rawashdeh, Samir A et al. “Wearable IMU for Shoulder Injury Prevention in Overhead Sports.” Sensors (Basel, Switzerland) vol. 16,11 1847. 3 Nov. 2016, doi:10.3390/s16111847
[5] Ancans, Armands et al. “Wearable Sensor Clothing for Body Movement Measurement during Physical Activities in Healthcare.” Sensors (Basel, Switzerland) vol. 21,6 2068. 16 Mar. 2021, doi:10.3390/s21062068
[6] Toshev, Alexander et al. “DeepPose: Human Pose Estimation via Deep Neural Networks.” IEEE Conference on Computer Vision and Pattern Recognition vol. 25 Sep. 2014, doi: 10.1109/CVPR.2014.214
[7] Adesida, Yewande et al. “Exploring the Role of Wearable Technology in Sport Kinematics and Kinetics: A Systematic Review.” Sensors (Basel, Switzerland) vol. 19,7 1597. 2 Apr. 2019, doi:10.3390/s19071597
[8] Lui, Jordan, and Carlo Menon. “Would a thermal sensor improve arm motion classification accuracy of a single wrist-mounted inertial device?.” Biomedical engineering online vol. 18,1 53. 7 May. 2019, doi:10.1186/s12938-019-0677-7
[9] Filippeschi, Alessandro et al. “Survey of Motion Tracking Methods Based on Inertial Sensors: A Focus on Upper Limb Human Motion.” Sensors (Basel, Switzerland) vol. 17,6 1257. 1 Jun. 2017, doi:10.3390/s17061257
[10] González-Alonso, Javier et al. “Custom IMU-Based Wearable System for Robust 2.4 GHz Wireless Human Body Parts Orientation Tracking and 3D Movement Visualization on an Avatar.” Sensors (Basel, Switzerland) vol. 21,19 6642. 6 Oct. 2021, doi:10.3390/s21196642
[11] Sabatini, Angelo Maria. “Estimating three-dimensional orientation of human body parts by inertial/magnetic sensing.” Sensors (Basel, Switzerland) vol. 11,2 (2011): 1489-525. doi:10.3390/s110201489
[12] Horenstein, Rachel E et al. “Using Magneto-Inertial Measurement Units to Pervasively Measure Hip Joint Motion during Sports.” Sensors (Basel, Switzerland) vol. 20,17 4970. 2 Sep. 2020, doi:10.3390/s20174970
[13] Miezal, Markus et al. “On Inertial Body Tracking in the Presence of Model Calibration Errors.” Sensors (Basel, Switzerland) vol. 16,7 1132. 22 Jul. 2016, doi:10.3390/s16071132
[14] Slade, Patrick et al. “An Open-Source and Wearable System for Measuring 3D Human Motion in Real-Time.” IEEE transactions on bio-medical engineering vol. 69,2 (2022): 678-688. doi:10.1109/TBME.2021.3103201
[15] Gong, Wenjuan et al. “Human Pose Estimation from Monocular Images: A Comprehensive Survey.” Sensors (Basel, Switzerland) vol. 16,12 1966. 25 Nov. 2016, doi: 10.3390/s16121966
[16] Wang, Hao et al. “LHPE-nets: A lightweight 2D and 3D human pose estimation model with well-structural deep networks and multi-view pose sample simplification method.” PLoS One vol.17.2 (2022): e0264302. doi: 10.1371/journal.pone.0264302
[17] GU Tian-qi, ZHANG Lei, JI Shi-jun, TAN Xiao-dan, HU Ming. Curve fitting method for closed discrete points. Journal of Jilin University Engineering and Technology Edition, vol. 45, pp.437-441, 2015.
[18] ADXL335 三軸重力加速度感測器規格書：https://www.mouser.tw/datasheet/2/609/ADXL335-1503897.pdf