태권도 겨루기 경기력 관련 데이터 수집을 위한 인공지능 활용연구-SAM2와 HPE의 적용

• 영문명: Research on data collection methods related to taekwondo competitions using AI- Application of SAM2 and HPE
• 발행기관: 세계태권도문화학회
• 저자명: 김우진(Woo jin Kim) 정훈(Hun Jung) 민병석(Byeong Seok Min) 김주원(Joo Won Kim)
• 간행물 정보: 『세계태권도문화학회지』16권 2호, 1~18쪽, 전체 18쪽
• 주제분류: 예술체육 > 체육
• 파일형태: PDF
• 발행일자: 2025.04.30

국문 초록

In this study, we suggest a technique for transforming and analyzing video data from Taekwondo competitions into a structured format. The objects in the video are segmented using SAM2, and the joint coordinates of the participants are extracted using HPE. To enhance the interpretability of the model's predictions, we also implement XAI (Explainable AI) techniques. The research demonstrated that SAM2 efficiently segmented essential components of the taekwondo competition film, while HPE properly extracted the coordinates of 18 critical joints of the participants. Analysis of SHAP (SHapley Additive exPlanations) values indicated that alterations in the X-axis coordinates, particularly of the ankle, significantly influenced scoring predictions.

목차

Ⅰ. 서 론

Ⅱ. 연구방법

Ⅲ. 결 과

Ⅳ. 논 의

Ⅴ. 결론 및 제언

참고문헌

키워드

Taekowndo Human Pose Estimation (HPE) Segment Anything Model 2 (SAM2) SHapley Additive exPlanations(SHAP) Explainable AI(XAI)