Template-Based Respiratory Monitoring and Tidal Volume Estimation

Vision-based monitoring has gained attention as a non-invasive alternative to conventional contact-based methods in intensive care units (ICUs). While prior studies highlight its potential for respiratory monitoring, the approach remains nascent, with challenges such as heuristically defined respiratory-relevant regions. This lack of anatomically consistent definitions can undermine the reliability of volume estimation techniques. Typically, patient-specific calibration data are required to address this limitation, yet such data are frequently unavailable in practical settings. In this study, we introduce a novel method for modeling the torso region to enable precise respiratory monitoring and volume estimation. Using a Kinect camera, we capture the patient's body surface, which is then aligned and registered with a standardized human template for consistent segmentation of the respiration-relevant region. Subsequently, an efficient and accurate volume computation algorithm extracts respiratory parameters and monitors tidal volume. We validated our approach on a cohort of volunteers, benchmarking it against gold-standard spirometry and expert annotations. Overall, the proposed method achieves over 90% accuracy in tidal volume estimation and 95% accuracy in respiratory rate estimation, without relying on patient-specific calibration. These results demonstrate its robustness and suitability for ICU respiratory monitoring.