Validating linear mechanical simulation of human respiratory motion using optical flow for camera-based vital signs monitoring
L.W. te Hennepe (TU Delft - Mechanical Engineering)
J.J. van den Dobbelsteen – Mentor (TU Delft - Mechanical Engineering)
T. Horeman – Graduation committee member (TU Delft - Mechanical Engineering)
D.S. van Scheppingen – Mentor (Philips)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Problem: Camera-based vital sign monitoring offers an additional safety layer to general wards, but developing and validating such systems requires representative datasets of human respiration patterns. Current approaches rely on costly and time-consuming recruitment of human participants, limiting algorithmic development and benchmarking. A linear mechanical respiration simulator could reduce this burden, but only if it produces respiration patterns that the optical flow algorithm detects the same way it detects real human breathing. Methods: A respiration simulator was constructed and validated to faithfully reproduce human respiration patterns derived from a respiratory belt. A dataset of eight subjects was used across three scenarios where the subjects were lying still. Optical flow was extracted from videos of both the human subject and the respiration simulator performing the same respiration pattern using the Farneback optical flow algorithm. Correlations between human and simulator optical flow were evaluated separately for each scenario using Fisher's z-transformation and one-sample t-tests. Results: Equivalence between detected movement by the optical flow for the human subject and respiration simulator was confirmed across all three scenarios, with mean r-values of 0.938 (95% CI: [0.895, 0.963]), 0.916 (95% CI: [0.845, 0.955]), and 0.935 (95% CI: [0.885, 0.964]), all significantly different from zero (p < 0.001). The respiration simulator reproduced respiration belt input signals with high fidelity (range: 0.895-0.995), and optical flow tracking of simulator movement was robust across physiological amplitudes and frequencies (r > 0.952). Conclusions: A simple linear mechanical simulator can replicate human respiratory motion in a manner equivalently detectable by optical flow algorithms. This validates the use of such a simulator as a controllable, repeatable alternative to human participants for developing, training, and benchmarking camera-based respiratory monitoring systems.