本研究提出一套結合非穿戴式姿態擷取與物理引擎模擬之步態分析平台，旨在解決現行高精度步態量測系統於成本、部署與使用門檻上的限制，提供一種可應用於居家及非實驗室環境之可行方案。系統採用OpenPose搭配雙目攝影機進行二維人體關鍵點偵測，經由三角測距與濾波處理重建三維姿態，提升資料穩定性與可信度。在動態模擬方面，利用MuJoCo建構個體化多關節人體模型，並設計PD控制器重現實際步態動作，搭配貝氏最佳化方法自動調整地面接觸力學參數，提升模擬精度。
實驗結果顯示，本平台能有效重建三維關節運動與估測腳底反力，具備實用性與可擴展性，適用於復健評估、自動化監測與行為模擬等應用場景。未來可進一步整合三維姿態估測模型、強化學習控制器與邊緣運算架構，提升系統即時性與智能化程度，推動智慧健康照護與步態行為分析邁向更高層次。

This study proposes a gait analysis platform that integrates non-wearable posture acquisition with physics engine simulation, aiming to address the limitations of existing high-precision gait measurement systems in terms of cost, deployment, and operational barriers. The system provides a feasible solution applicable to home and non-laboratory environments. OpenPose, in conjunction with a stereo camera setup, is employed for two-dimensional human keypoint detection. Three-dimensional posture reconstruction is achieved through triangulation and filtering, thereby enhancing data stability and reliability. For dynamic simulation, an individualized multi-joint human model is built in MuJoCo, and a PD controller is designed to reproduce actual gait movements. Bayesian optimization is applied to automatically tune ground contact mechanics parameters, further improving simulation accuracy.
Experimental results demonstrate that the proposed platform can effectively reconstruct three-dimensional joint motions and estimate plantar ground reaction forces, exhibiting both practicality and scalability. It is applicable to scenarios such as rehabilitation assessment, automated monitoring, and behavioral simulation. Future developments may integrate three-dimensional pose estimation models, reinforcement learning controllers, and edge computing architectures to enhance real-time performance and intelligence, thereby advancing smart healthcare and gait behavior analysis to a higher level.

[1] H Zheng et al. Design of teaching system of physical yoga course in colleges and universities based on computer network. Security and Communication Networks, 2022(1):6591194, 2022.
[2] X H Yang. Utilizing depth data and 2d-pose estimation from part affinity fields to develop a tennis training system. Master’s thesis, NCKU, 2021.
[3] Motek Medical BV, Hogehilweg 18 - C, 1101 CD, Amsterdam, The Nether-lands. HBM2 - Reference Manual, 2018.
[4] Barkfin. Articulated dummy. https://grabcad.com/library/articulated-dummy-1, 2022. Accessed: 2025-05-15.
[5] A Thomas. The 4 planes of the body — how to remember each, 2024. Ac-cessed: 2025-07-01.
[6] E Todorov et al. Mujoco: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026–5033. IEEE, 2012.
[7] M Whittle. Gait analysis: an introduction. Butterworth-Heinemann, 2014.
[8] A G Cutti et al. ‘outwalk’: a protocol for clinical gait analysis based on inertial and magnetic sensors. Medical & biological engineering & computing, 48(1):17–25, 2010.
[9] G Prisco et al. Validity of wearable inertial sensors for gait analysis: a sys-tematic review. Diagnostics, 15(1):36, 2024.
[10] A Pfister et al. Comparative abilities of microsoft kinect and vicon 3d mo-tion capture for gait analysis. Journal of medical engineering & technology, 38(5):274–280, 2014.
[11] C Kranzinger et al. Classification of human motion data based on inertial measurement units in sports: a scoping review. Applied Sciences, 13(15):8684, 2023.
[12] Z Hu et al. Light-adaptive human body key point detection algorithm based on multi-source information fusion. Sensors, 24(10):3021, 2024.
[13] Z Cao et al. Openpose: Realtime multi-person 2d pose estimation using part affinity fields. IEEE transactions on pattern analysis and machine intelli-gence, 43(1):172–186, 2019.
[14] L Burnie et al. Commercially available pressure sensors for sport and health applications: A comparative review. The Foot, 56:102046, 2023.
[15] L Weidensager et al. Estimating vertical ground reaction forces from plantar pressure using interpretable high-dimensional approximation. Sports Engi-neering, 27(1):3, 2024.
[16] C E Rasmussen. Gaussian processes in machine learning. In Summer school on machine learning, pages 63–71. Springer, 2003.
[17] C Y Lee et al. Deeply-supervised nets. In Artificial intelligence and statistics, pages 562–570. Pmlr, 2015.
[18] Z Zhang. A flexible new technique for camera calibration. IEEE Transactions on pattern analysis and machine intelligence, 22(11):1330–1334, 2002.
[19] D W Marquardt. An algorithm for least-squares estimation of nonlinear pa-rameters. Journal of the society for Industrial and Applied Mathematics, 11(2):431–441, 1963.
[20] K Adel. Human joint range of motion – the complete guide, 2024. Accessed: 2025-05-15.
[21] S Plagenhoef et al. Anatomical data for analyzing human motion. Research Quarterly for Exercise and Sport, 54(2):169–178, 1983.
[22] B Acosta et al. Validating robotics simulators on real-world impacts. IEEE Robotics and Automation Letters, 7(3):6471–6478, 2022.
[23] E Todorov. A convex, smooth and invertible contact model for trajectory optimization. In 2011 IEEE International Conference on Robotics and Au-tomation, pages 1071–1076. IEEE, 2011.
[24] F E Udwadia and R E Kalaba. A new perspective on constrained motion. Pro-ceedings of the Royal Society of London. Series A: Mathematical and Physical Sciences, 439(1906):407–410, 1992.
[25] R Barrett et al. Templates for the solution of linear systems: building blocks for iterative methods. SIAM, 1994.
[26] J Nocedal and S J Wright. Numerical optimization. Springer, 2006.
[27] R Smith. Open dynamics engine. 2005.
[28] R E Kearney and I W Hunter. System identification of human joint dynamics. Critical reviews in biomedical engineering, 18(1):55–87, 1990.
[29] AIST Intelligent Systems Research Institute (ISRI). Jvrc1 model files for mu-joco. https://github.com/isri-aist/jvrc_mj_description, 2022. Ac-cessed: 2025-05-15.
[30] R Senden et al. Normative 3d gait data of healthy adults walking at three different speeds on an instrumented treadmill in virtual reality. Data in brief, 53:110230, 2024.
[31] X W Lin. Model-fusion-based precision motion control of linear motors. Mas-ter’s thesis, NCKU, 2013.
[32] Y C Zhang. Using bayesian optimization to search th best cnn architecture–an example with mammograms to detect breast abnormalities. Master’s thesis, NCKU, 2022.
[33] D Pavllo et al. 3d human pose estimation in video with temporal convolutions and semi-supervised training. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7753–7762, 2019.
[34] J Li et al. Hybrik-x: Hybrid analytical-neural inverse kinematics for whole-body mesh recovery. arXiv preprint arXiv:2304.05690, 2023.