簡易檢索 / 詳目顯示

回結果列表
研究生： 施邑穎
Shih, Yi-Ying
論文名稱： 基於2D-to-3D Diffusion-Transformer網路之棒球投球骨架動作切分及動作分析
Skeleton-Based Motion Segmentation and Analysis Based on 2D-to-3D Diffusion-Transformer Network for Baseball Pitcher
指導教授： 連震杰
Lien, Jenn-Jier James
學位類別： 碩士
Master
系所名稱： 電機資訊學院 - 人工智慧科技碩士學位學程
Graduate Program of Artificial Intelligence
論文出版年： 2026
畢業學年度： 114
語文別： 英文
論文頁數： 74
中文關鍵詞： 棒球投球分析 、人體姿態估測 、三維人體姿態重建 、擴散模型 、生物力學分析 、投球四階段切分 、D3DP 、MixSTE
外文關鍵詞： Baseball Pitching Analysis, Human Pose Estimation, 3D Human Pose Reconstruction, Diffusion Model, Biomechanical Analysis, Four-Phase Pitching Segmentation, D3DP
相關次數： 點閱：12  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 棒球投球動作分析為運動科學與運動醫學的重要研究課題，透過分析投手投球過程中的動作時序與生物力學特徵，可作為技術訓練、動作優化及運動傷害預防的重要依據。然而，傳統投球分析多仰賴光學動作捕捉系統或穿戴式感測器，設備成本高且受限於實驗室環境，不利於實際訓練場域之應用。因此，本研究提出一套結合人體姿態估測、三維人體骨架重建、投球四階段切分及生物力學分析之智慧化棒球投球分析系統。本研究首先利用雙攝影機同步擷取投球影像，透過相機校正與三角測量法重建三維人體骨架，並結合 YOLOv11 人體偵測模型與 ViTPose 人體姿態估測模型取得二維人體骨架。接著，根據人體關節運動特徵，自動偵測 Foot Contact（FC）、Maximum External Rotation（MER）及 Ball Release（BR）三個關鍵幀，完成投球四階段切分，並計算肩髖分離角、肩關節外旋角、手腕速度、跨步距離及出手距離等生物力學參數，以提供客觀之投球動作分析。此外，為降低雙攝影機架設需求，本研究進一步探討單視角人體姿態提升技術於棒球投球分析之應用，採用 Diffusion-Based 3D Pose Prediction with Multi-Hypothesis Aggregation（D3DP）作為人體姿態提升模型，並利用自行建立之棒球投球資料集進行資料前處理與模型微調，使模型能適應高速投球動作。推論階段結合 MixSTE Denoiser、DDIM 去雜訊策略及多假設聚合機制，以提升單視角三維人體姿態重建之穩定性與準確性。綜合而言，本研究建立一套兼具雙視角人體姿態分析與單視角人體姿態提升之棒球投球分析系統，提供由人體姿態估測、三維人體骨架重建、投球四階段切分至生物力學分析之完整流程，可作為未來智慧運動分析、棒球訓練及教練輔助決策之參考依據。

    Baseball pitching motion analysis plays an important role in sports science and sports medicine, providing valuable information for performance enhancement, technique optimization, and injury prevention. However, conventional pitching analysis typically relies on optical motion capture systems or wearable sensors, which are expensive and confined to laboratory environments, limiting their applicability in practical training. Therefore, this study proposes an intelligent baseball pitching analysis system integrating human pose estimation, 3D human pose reconstruction, four-phase pitching segmentation, and biomechanical analysis. First, synchronized pitching videos are captured using a dual-camera setup. YOLOv11 and ViTPose are employed for 2D human pose estimation, followed by triangulation to reconstruct 3D human skeletons. Based on joint motion characteristics, three key events—Foot Contact (FC), Maximum External Rotation (MER), and Ball Release (BR)—are automatically detected to segment the pitching motion into four phases. Biomechanical parameters, including the shoulder–hip separation angle, shoulder external rotation angle, wrist velocity, and stride length, are then calculated for quantitative motion analysis. Furthermore, to reduce the dependence on a dual-camera setup, this study adopts Diffusion-Based 3D Pose Prediction with Multi-Hypothesis Aggregation (D3DP) and fine-tunes the model using a self-collected baseball pitching dataset to improve monocular 3D human pose reconstruction. In summary, this study develops a baseball pitching analysis system that combines dual-camera pose analysis with monocular human pose lifting, providing a complete pipeline for human pose estimation, 3D pose reconstruction, four-phase pitching segmentation, and biomechanical analysis. The proposed system has the potential to support intelligent sports analytics, baseball training, and coaching decision-making.

    摘要 I ABSTRACT II 誌謝 III Table of Contents V List of Tables VII List of Figures VIII Chapter 1 Introduction 1 1.1 Motivation and Objective 1 1.2 Global Framework 2 1.3 Related Works 5 1.3.1 Human Motion Capture Technologies 6 1.3.2 3D Human Pose Estimation 8 1.4 Contribution 10 Chapter 2 System Setup and Hardware Specifications 13 2.1 System Setup 13 2.2 Hardware Specifications 14 2.3 User Interface for Display 15 Chapter 3 4-Phase Segmentation via 3-Keyframe Detection Based on 2D/3D Skeleton Joint Motion 19 3.1 4-Phase Segmentation Based on 2D Human Skeleton 19 3.1.1 Camera Calibration 20 3.1.2 2D Skeleton Point Estimation 24 3.1.3 4-Phase Segmentation via 3-Keyframe Detection Based on 2D Skeleton 27 3.2 4-Phase Segmentation Based on 3D Human Skeleton 30 3.2.1 3D Skeleton Point Estimation Based on Fundamental Matrix 30 3.2.2 4-Phase Segmentation via 3-Keyframe Detection Based on 3D Skeleton Joint Motion 33 3.3 Functions of Pitching Motion Action 35 Chapter 4 2D-TO-3D Skeleton Using Diffusion-Based with Multi-Hypothesis Aggregation 40 4.1 Diffusion-Based 3D Pose Estimation with Multi-Hypothesis Aggregation 40 4.1.1 Diffusion-Based 3D Pose Estimation (D3DP) of Pretraining Process 42 4.1.2 Diffusion-Based 3D Pose Estimation (D3DP) of Inference Process 44 4.1.3 Training Framework of D3DP 46 4.2 MixSTE Denoiser 48 Chapter 5 Experimental Results 52 5.1 Data Collection and Metrics 52 5.1.1 Data Collection 53 5.1.2 Metrics 54 5.2 Experimental Results of 4-Phase Segmentation 55 5.3 Experimental Results of D3DP 56 Chapter 6 Conclusion and Future Works 58 6.1 Conclusion 58 6.2 Future Works 59 REFERENCES 62

    [1] B. Alexey, C. Wang, and H.Y.M. Liao, “Yolov4: Optimal Speed And Accuracy of Object Detection,” arXiv preprint arXiv:2004, 2020.
    [2] K. Aouaidjia, B. Sheng, P. Li, J. Kim and D.D. Feng, “Efficient Body Motion Quantification and Similarity Evaluation using 3-D Joints Skeleton Coordinates,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 51, no. 5, pp. 2774-2788, 2019.
    [3] C.L. Camp, S. Loushin, S. Nezlek, A.P. Fiegen, D. Christoffer, and K. Kaufman, “Are Wearable Sensors Valid and Reliable for Studying the Baseball Pitching Motion? An Independent Comparison with Marker-Based Motion Capture,” The American Journal of Sports Medicine, 49(11), pp. 3094-3101, 2021.
    [4] Y. Chen, and Y. Wang, “Tracknetv3: Enhancing Shuttlecock Tracking with Augmentations and Trajectory Rectification,” Proceedings of the 5th ACM International Conference on Multimedia in Asia, 2023.
    [5] Y. Cheng, B. Yang, B. Wang, W. Yan, and R.T. Tan, “Occlusion-Aware Networks for 3D Human Pose Estimation in Video,” IEEE/CVF International Conference on Computer Vision (ICCV), pp. 723–732, 2019.
    [6] T. Chou, and W. Chu, “Automatic Baseball Pitch Overlay,” International Conference on Multimedia Retrieval (ICMR ‘21), pp. 599-602, 2021.
    [7] A. Elaoud, W. Barhoumi, E. Zagrouba, and B. Agrebi, “Skeleton-Based Comparison of Throwing Motion for Handball Players,” Journal of Ambient Intelligence and Humanized Computing, 11(1), pp. 419-431, 2020.
    [8] R. Escamilla, G. Fleisig, S. Barrentine, J. Andrews, and C. Moorman 3rd, “Kinematic and Kinetic Comparisons between American and Korean Professional Baseball Pitchers,” Sports Biomechanics, 1(2), pp. 213-228, 2002.
    [9] G.S. Fleisig, S.W. Barrentine, N. Zheng, R.F. Escamilla, and J.R. Andrews, “Kinematic and Kinetic Comparison of Baseball Pitching Among Various Levels of Development,” Journal of Biomechanics, 32(12), pp. 1371-1375, 1999.
    [10] M.R.I. Hossain, and J.J. Little, “Exploiting Temporal Information for 3D Human Pose Estimation,” Proceedings of the European Conference on Computer Vision (ECCV), pp. 68–84, 2018.
    [11] C. Ionescu, D. Papava, V. Olaru, and C. Sminchisescu, “Human3.6m: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(7), pp. 1325–1339, 2013.
    [12] W. Li, M. Liu, H. Liu, P. Wang, J. Cai, and N. Sebe, “Hourglass Tokenizer for Efficient Transformer-Based 3D Human Pose Estimation,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 604-613, 2024.
    [13] J. Liu, N. Saquib, Z. Chen, H. Fu, R.H. Kazi, L. Wei, and C. Tai, “Posecoach: A Customizable Analysis and Visualization System for Video-Based Running Coaching,” IEEE Transactions on Visualization and Computer Graphics, pp. 3180-3195, 2022.
    [14] J. Liu, M. Liu, H. Liu, and W. Li, “TCPFormer: Learning Temporal Correlation with Implicit Pose Proxy for 3D Human Pose Estimation,” Proceedings of the AAAI Conference on Artificial Intelligence, pp. 5478-5486, 2025.
    [15] M.E. Loaiza, A.B. Raposo, and M. Gattass, “Multi-Camera Calibration Based on An Invariant Pattern,” Computers & Graphics 35.2, pp. 198-207, 2011.
    [16] J.E. Manzi, B. Dowling, Z. Wang, K.N. Kunze, J. Estrada, M.C. Fu, K.L. McElheny, J.B. Carr, and J.S. Dines, “Association of Pitch Timing And Throwing Arm Kinetics in High School and Professional Pitchers”, The American Journal of Sports Medicine, 49(12), pp. 3386-3394, 2021.
    [17] T. von Marcard, R. Henschel, M. Black, B. Rosenhahn, and G. Pons-Moll, “Recovering Accurate 3D Human Pose in The Wild Using Imus and A Moving Camera,” European Conference on Computer Vision (ECCV), pp. 601-617, 2018.
    [18] D. Mehta, H. Rhodin, D. Casas, P. Fua, O. Sotnychenko, W. Xu, and C. Theobalt, “Monocular 3D Human Pose Estimation in The Wild Using Improved Cnn Supervision,” International Conference on 3D Vision (3DV), pp. 506–516, 2017.
    [19] D. Pavllo, C. Feichtenhofer, D. Grangier, and M. Auli, “3D Human Pose Estimation in Video with Temporal Convolutions and Semi-supervised Training,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 7753–7762, 2019.
    [20] J. Peng, Y. Zhou, and P.Y. Mok, “Ktpformer: Kinematics and Trajectory Prior Knowledge-enhanced Transformer for 3D Human Pose Estimation,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1123-1132, 2024.
    [21] W. Shan, Z. Liu, X. Zhang, Z. Wang, K. Han, S. Wang, S. Ma, and W. Gao, “Diffusion-based 3D Human Pose Estimation with Multi-hypothesis Aggregation,” Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 14761-14771, 2023.
    [22] B. Tseng, H. Chiang, J. Chen, and H. Hsieh, “Pitching-Motion: Pose-Based Pitch Trajectory Overlay System,” IEEE 26th International Conference on Advanced Communications Technology (ICACT), pp. 116-121, 2024.
    [23] N. Wojke, A. Bewley, and D. Paulus, “Simple Online and Realtime Tracking with a Deep Association Metric,” IEEE International Conference on Image Processing (ICIP), pp. 3645-3649, 2017.
    [24] H. Ye, W. Zhu, C. Wang, R. Wu, and Y. Wang, “Faster Voxelpose: Real-Time 3D Human Pose Estimation by Orthographic Projection.” European Conference on Computer Vision. Cham: Springer Nature Switzerland, pp. 142-159, 2022.
    [25] J. Zhang, Z. Tu, J. Yang, Y. Chen ,and J. Yuan, “Mixste: Seq2seq Mixed Spatiotemporal Encoder for 3D Human Pose Estimation in Video,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 13232–13242, 2022.
    [26] H. Zheng, H. Li, W. Dai, Z. Zheng, C. Li, J. Zou, and H. Xiong, “HiPART: Hierarchical Pose AutoRegressive Transformer for Occluded 3D Human Pose Estimation,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 16807-16817, 2025.

    下載圖示
    校外：立即公開
    QR CODE