Beat-to-beat heart rate estimation fusing multimodal video and sensor data

Abstract

Coverage and accuracy of unobtrusively measured biosignals are generally relatively low compared to clinical modalities. This can be improved by exploiting redundancies in multiple channels with methods of sensor fusion. In this paper, we demonstrate that two modalities, skin color variation and head motion, can be extracted from the video stream recorded with a webcam. Using a Bayesian approach, these signals are fused with a ballistocardiographic signal obtained from the seat of a chair with a mean absolute beat-to-beat estimation error below 25 milliseconds and an average coverage above 90% compared to an ECG reference.

On the Performance of Bed-Integrated Ballistocardiography in Long-Term Heart Rate Monitoring of Vascular Patients

“…As show in previous work, the inclusion of an adaptive prior could improve beat-to-beat estimation in low-quality signals [8]. Thus, a similar approach was applied.…”

Reducing false arrhythmia alarms using robust interval estimation and machine learning

“…This estimator was originally developed for the analysis of ballistocardiographic signals [2], but variations have since been successfully applied to unobtrusive vital sign estimation from multimodal sources [3] as well as robust detection of heart beats in multimodal ICU data [4]. Using a moving window, the self-similarity of the cardiac signals is analyzed.…”

“…While it was necessary to locate individual heart beats for the PhysioNet/Computing in Cardiology Challenge 2014 [5], features are extracted directly from the estimated intervals and used to train several machine learning approaches here. As it is essential to this year's challenge, the general concept of interval estimation is briefly reviewed in the next section, while a detailed explanation can be found in [3].…”

“…which ensures that the exact number of samples necessary for each candidate lag η is considered, see also [3]. Another metric to asses self-similarity is the average magnitude difference function (AMDF).…”

“…also uses the lag-adaptive window and is inverted so that it assumes larger values for lags that indicate more selfsimilarity [3]. As a third metric, the maximum amplitude pairs (MAP) function considers the amplitude of the signal and can thus be considered as indirect peak-detection, (ω(ν) + ω(ν − η)) .…”