Robust real-time PPG-based heart rate monitoring

Many existing methods for tracking heart rate (HR) from pho-toplethysmographic (PPG) signals can be found in the literature, but they are tested only in static scenarios and fail when motion artifacts are strong. Recently, an algorithm called TROIKA was proposed for robust HR tracking, but the computational complexity of that algorithm is very high which makes it difficult to implement in small embedded devices, such as wrist-wearable HR monitors. In this article we present a new, fast family of methods robust against very strong motion artifacts using spectral subtraction or nonnegative matrix factorization (NMF) for signal enhancement and MA removal and online Viterbi decoding or particle filtering for HR tracking. On our test data set we obtain an average error of 1.3%.