Premium
An edge‐based scheme to support heart rate estimation for different physical exercises
Author(s) -
Pan Wenwen,
Ma Shuai
Publication year - 2020
Publication title -
internet technology letters
Language(s) - English
Resource type - Journals
ISSN - 2476-1508
DOI - 10.1002/itl2.192
Subject(s) - enhanced data rates for gsm evolution , computer science , wearable computer , particle filter , scheme (mathematics) , wearable technology , filter (signal processing) , recursive least squares filter , mean squared error , real time computing , cluster (spacecraft) , word error rate , algorithm , simulation , artificial intelligence , computer vision , statistics , mathematics , adaptive filter , embedded system , mathematical analysis , programming language
As an emerging computing paradigm for smart cities, edge computing has received widespread attention. A huge amount of healthcare data in smart city poses more severe issues to the resource‐constrained wearable platform. Meanwhile, this type of platform can not load some complex models to simultaneously improve the accuracy and real‐time performance of algorithms, particularly for heart rate (HR) estimation strongly contaminated by motion artifacts. In this paper, we propose an edge‐based scheme to estimate heart rate during different physical exercises, for example, walking and running with varying speed. This scheme utilizes two‐channel PPG signals and runs on two resource‐constrained edge clusters. The estimation of HR is divided into two stages and runs on edge cluster 1 (C1) and edge cluster 2 (C2), respectively. Firstly, the recursive least squares (RLS) filter running on the C1 is used to discard runaway errors and generate an initial estimate of HR. Second, on the C2, we use particle filters to track HR and can quickly recover the correct value of HR from the incorrect estimates. Our proposed algorithm is validated on a public dataset of 12 subjects recorded during running and walking, compared with some state‐of‐the‐art methods. Experimental results show that our algorithm has a better absolute average error and real‐time performance.