Frequency Tracker for Unsupervised Heart Rate Estimation
We present frequency tracking for extracting heart rate trace from blood volume pulse (BVP) signal that can be used as an alternative for commonly used approach based on the mode of the BVP signal power spectral density. Our approach is based on particle filtering framework which provides smooth heart rate estimate, it is robust to motion-induced artifacts and noise. The method could be easily tuned and can be coupled with unsupervised BVP extraction approaches without the need for training. We evaluate our method on publicly available part of LGI dataset. Proposed algorithm shows competitive results comparing to argmax approach.