Deep Learning Based Heart Rate Estimation Using Smart Shoes Sensor

Although heart rate is an important biomarker of the physical condition at active states of users, it is still difficult to be measured due to an ambient noise and movements. There have been several approaches proposed to obtain stable measurements in an active condition. However, these methods still need direct contact to users, and thus additional equipment to keep the contact are requested, resulting in inconvenience of its usage. This paper proposes a method to estimate the heart rate of the user using activity information from smart shoes sensors, which is relatively easy and robust to be recorded. For the accurate estimation of the heart rate, a new design of deep neural networks is proposed. The architecture extracts features of time-sequential patterns of sensor data with implementing CNN and LSTM model together. The model was validated with a `Leave-OneOut Cross-Validation method'. The results of the experiments are 10.21 ± 3.31 RMSE, 8.31 ± 2.81 MAE and 0.91 ± 0.09 correlation coefficient (Pearson) for the estimation of heart rate from smart shoes sensor data.