문맥 인식 시간적 차분 합성곱 기반 비접촉식 심박수 추정에 관한 연구

Abstract

This paper proposes a novel remote heart rate estimation model based on a 3D Video SwiftFormer architecture integrated with Context-Aware Temporal Difference Convolution (CTDC). The model effectively captures spatial-temporal features and incorporates a soft-DTW loss to enhance signal alignment. Experiments on the iBVP dataset show that the proposed method outperforms existing approaches in both accuracy and efficiency, demonstrating its potential for real-time applications.