Improved Touchless Respiratory Rate Sensing

TL;DR

This paper introduces a novel method for remote respiratory rate measurement using video streams, enhancing accuracy and enabling real-time monitoring for human-computer interaction applications.

Contribution

The paper presents a new 1D profile creation technique and motion signal grouping method that significantly improve pixel intensity change-based respiratory rate estimation.

Findings

Achieved MAE of 0.7, 0.6, and 1.4 BPM on different datasets.

Enabled real-time continuous respiratory rate monitoring.

Enhanced accuracy over previous pixel intensity change methods.

Abstract

Recently, remote respiratory rate measurement techniques gained much attention as they were developed to overcome the limitations of device-based classical methods and manual counting. Many approaches for RR extraction from the video stream of the visible light camera were proposed, including the pixel intensity changes method. In this paper, we propose a new method for 1D profile creation for pixel intensity changes-based method, which significantly increases the algorithm's performance. Additional accuracy gain is obtained via a new method of motion signals grouping presented in this work. We introduce several changes to the standard pipeline, which enables real-time continuous RR monitoring and allows applications in the human-computer interaction systems. Evaluation results on two internal and one public datasets showed 0.7 BPM, 0.6 BPM, and 1.4 BPM MAE, respectively.