Efficient tracking of heart rate under physical exercise from photoplethysmographic signals

This paper deals with the estimation of the temporal evolution of heart rate (HR), in subjects undertaking physical exercise, using wrist-type photoplethysmographic (PPG) signals. The proposed algorithm employs a zero-padded DFT to provide a raw HR estimate with small frequency granularity and feeds it into a Kalman filter to accurately track HR evolution over time. Furthermore, heuristics are applied to sort out the estimates provided by spectral analysis and determine a reasonable preliminary frequency value. The algorithm was an entrant into the 2015 Signal Processing Cup [1] and has been tested, with very encouraging results, on the set of recorded waveforms (training traces) provided for that competition. In this paper we discuss its features and outline possible improvements. Measurement results on actual recorded traces are given and are shown to outperform those provided in [2] on all training traces provided for the competition.