<![CDATA[Archery requires high consistency and precise body posture, where small deviations can affect stability and accuracy. Recently, 2D human pose estimation has become an effective approach for analyzing sports techniques through automatic joint detection. This study proposes a 2D pose estimation system based on the MediaPipe framework to detect eight fundamental phases of archery technique and evaluate accuracy using the Normalized Mean Per Joint Position Error (N-MPJPE) metric. The dataset consists of annotated images representing the eight phases, which serve as ground-truth references. Accuracy is measured by calculating the normalized Euclidean distance between predicted joint positions and ground-truth coordinates across all phases. Experimental results show an average N-MPJPE of 0.71, indicating low joint-position deviation after scale normalization. Compared with prior studies reporting N-MPJPE values between 0.6 and 1.2, the proposed system demonstrates competitive accuracy for real-time 2D pose estimation. These results indicate that the system can reliably capture posture variations across archery phases and provide quantitative feedback on body alignment, making it a practical tool to support athletes and coaches in improving training quality and shooting performance.]]>
Copyrights © 2026
