3D Reconstruction of Baseball Pitching Motion from Videos with Motion Blur
- 주제(키워드) physics-based simulation , video processing , motion reconstruction , reinforcement learning
- 주제(DDC) 006.31
- 발행기관 아주대학교 일반대학원
- 지도교수 Ri Yu
- 발행년도 2025
- 학위수여년월 2025. 8
- 학위명 석사
- 학과 및 전공 일반대학원 인공지능학과
- 실제URI http://www.dcollection.net/handler/ajou/000000034984
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
Baseball is one of the most popular sports worldwide, where pitching velocity and con- trol, collectively referred to as pitch quality, are key factors influencing the outcome of a game. With the continuous collection of high-quality video data from baseball games, a large amount of valuable information is available for reconstructing 3D pitching motions using pose estimation techniques. However, accurate motion reconstruction is hindered by challenges such as motion blur, particularly when the pitcher throws the ball at high speed toward the strike zone. To address this issue, we propose a two-stage learning approach, consisting of motion imitation and motion refinement, to sequentially reconstruct baseball pitching motions from imperfect pose estimates derived from video. Our framework com- bines physics-based simulations and deep reinforcement learning (DRL), along with domain- specific rewards, to guide the simulated character in throwing the ball to the target location at the desired speed while imitating the estimated pose and preserving the pitching style. We demonstrate the effectiveness of our framework through various experiments, showing that our method successfully reconstructs plausible, physically consistent pitching motions that closely resemble the original video, even when pose estimates are imperfect. Keywords: physics-based simulation, video processing, motion reconstruction.
more목차
1. Introduction 1
2. Related Work 4
2.1 Motion Reconstruction Using Monocular Video 4
2.2 Motion Deblurring 5
2.3 Physics-based Motion Control with Deep Reinforcement Learning 6
3. Proposed Method 8
3.1 Video Processing 9
3.2 Motion Reconstruction 12
3.3 Stage 1 : Pose Imitation 14
3.4 Stage 2 : Motion Refinement 16
4. Experimental Results and Discussions 20
4.1 Motion Reconstruction 20
4.2 Stage-wise Performance Comparison 22
4.3 Comparison of Single-Stage and Two-Stage Approaches 22
4.4 Ablation Study 24
5. Conclusion 27
Bibliography 29
