Machine Learning Driven Heart Rate Detection with Camera Photoplethysmography in Time Domain
Measuring bio signals such as the heart rate in non medical applications is gaining an increasing importance. With camera based photoplethysmography (PPG) it is possible to measure the heart rate remotely with built in webcams of every tablet and laptop.
- PDF / 874,772 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 57 Downloads / 217 Views
Abstract. Measuring bio signals such as the heart rate in non medical applications is gaining an increasing importance. With camera based photoplethysmography (PPG) it is possible to measure the heart rate remotely with built in webcams of every tablet and laptop. Recent research with machine learning based methods showed great success compared to signal processing based methods. In this paper, we use k-nearest neighbor (kNN) and multilayer perceptron (MLP) with an alternative representation of the input vector. Estimating the quality of peaks with a Gaussian distribution could further improve the detection. Overall we could improve the root mean square error (RMSE) from 23.97 to 8.62. Keywords: Photoplethysmography (PPG) · remote Photoplethysmography (rPPG) · Camera · Webcam · k-nearest neighbor · Neural network · Gaussian distribution
1
Introduction
Most tablets and laptops are equipped with a front camera and are often used for hours every day. For health care applications this is an interesting means to monitor the health of a person. Several works in the last ten years showed that camera based photoplethysmography (PPG) can be used to remotely measure bio signals such as the heart rate. This allows a long term monitoring system which does not interfere with the user in his/her daily work and at the same time does not need a daily scheduled, explicit measurement. Advances in signal processing based measurement methods for camera based PPG as well as camera technology enables better detection rates but until now not reliable for serious applications. Machine learning technology methods show great success in replicating systematic occurrences. However, until now only few learning based approaches were presented. One of the early works was presented by Lamonaca et al. [7]. They used a neural network to evaluate the blood pressure from facial videos recorded with a smartphone camera and its flashlight. Hsu et al. [6] used support c Springer International Publishing AG 2016 F. Schwenker et al. (Eds.): ANNPR 2016, LNAI 9896, pp. 324–334, 2016. DOI: 10.1007/978-3-319-46182-3 27
Machine Learning Driven Heart Rate Detection in Time Domain
325
vector regression (SVR) in the frequency domain to detect the heart rate. They showed three times better results than a pure signal processing based method. Maaoui et al. [8] used a support vector machine (SVM) and seven features from time and frequency domain with the aim of detecting the stress level. The remainder of this paper is organized as follows. In Sect. 2 the generation of the signal for the detection is explained. This includes skin extraction, signal filtering and detection of the heart rate. Two machine learning algorithms, knearest neighbors (kNN) and multilayer perceptron (MLP), are described in Sect. 3 and analyzed in Sect. 4 on the Open EmoRec II dataset [11]. A conclusion follows in Sect. 5.
2 2.1
Signal Extraction Region of Interest (ROI) Detection
The interesting content of a video for this work is the face of a participant within which we measure the h
Data Loading...
