Earticle

초록

영어

This study proposes a system that non-invasively predicts electrocardiogram (ECG) and photoplethysmography (PPG) biosignals through a multimodal deep learning model utilizing driver image information and time-series characteristics. As the first step toward responding to cardiac emergencies while driving, we developed a multimodal architecture combining Vision Transformer (ViT) and LSTM. The proposed system optimizes prediction performance through feature engineering that includes facial region analysis, RGB and HSV color feature extraction, and utilization of time-series patterns in biosignals. Experimental results demonstrate that the proposed model achieves excellent performance with R² of 0.998 and MAE of 0.014 for ECG prediction, and R² of 0.992 and MAE of 0.018 for PPG prediction. These results suggest that the proposed multimodal approach is effective in predicting driver biosignals. This study establishes the foundation for a non-invasive monitoring system that can provide notifications for cardiac emergencies while driving and suggests the possibility of expansion into a real-time anomaly detection and early warning system in the future.

목차

Abstract
1. 서론
2. 연구 방법
2.1 멀티모달 아키텍처
2.2 특징 공학(Feature Engineering)
2.3 데이터셋 및 학습 방법
3. 연구 결과
3.1 학습 과정 분석
3.2 모델 성능 평가
4. 결론
후기
References

키워드

저자

• 김경이 [ Kyoung-yee Kim | Professor, Dept. of AI Software, Inje University ] Corresponding Author
• 최재형 [ Jae-hyung Choi | M.S., Dept. of AI Convergence Robotics, Inje University ]
• 김성우 [ Seong-woo Kim | Same affiliation as above ]
• 박세훈 [ Se-hoon Park | Same affiliation as above ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  한국기계항공기술학회 학술대회논문집(구 한국기계기술학회 학술대회논문집) [Proceedings of KSMT Annual Meeting]

• 간기

  부정기

• 수록기간

  2000~2026

• 십진분류

  KDC 550 DDC 620