Presentation Information
[B-19-07]Non-Contact Heart Rate Estimation Using Time-Series Deep Learning Models with a 60-GHz Pulse Sensor
◎△Taiyo Takahashi1, Takeshi Toda2 (1. Graduate School of Science & Technology,NIHON UNIVERSITY, 2. College of Science & Technology,NIHON UNIVERSITY)
Keywords:
Pulse Sensor,Heart Rate,Vital Signs,Healthcare,Radar
Conventionally, FMCW radar has been the mainstream approach for non-contact vital sign detection. In our previous study, we investigated heart rate estimation using deep learning with CNN and 2D ResNet applied to a low-power pulse-type sensor, achieving MAE of 9.54 BPM and 9.03 BPM, respectively. However, the subject composition of the training data and the measurement distance were limited, leaving the evaluation under more practical conditions as a future challenge. In this study, we revised the subject composition of the training data and introduced a time-series model, LSTM, under conditions with a reduced measurement distance, aiming to improve heart rate estimation accuracy.