Multisensor data fusion for enhanced respiratory rate estimation in thermal videos

• Published in: 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) Vol. 2016; pp. 1381 - 1384
• Main Authors: Pereira, Carina B., Xinchi Yu, Blazek, Vladimir, Venema, Boudewijn, Leonhardt, Steffen
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.08.2016
• Subjects: Adult; Algorithms; Cameras; Correlation; Data integration; Female; Hot Temperature; Humans; Male; Monitoring; Mouth; Movement; Nose; Respiration; Respiratory Rate - physiology; Signal Processing, Computer-Assisted; Videos; Videotape Recording
• ISSN: 1557-170X
• DOI: 10.1109/EMBC.2016.7590965

Scientific studies have demonstrated that an atypical respiratory rate (RR) is frequently one of the earliest and major indicators of physiological distress. However, it is also described in the literature as "the neglected vital parameter", mainly due to shortcomings of clinical available monitoring techniques, which require attachment of sensors to the patient's body. The current paper introduces a novel approach that uses multisensor data fusion for an enhanced RR estimation in thermal videos. It considers not only the temperature variation around nostrils and mouth, but the upward and downward movement of both shoulders. In order to analyze the performance of our approach, two experiments were carried out on five healthy candidates. While during phase A, the subjects breathed normally, during phase B they simulated different breathing patterns. Thoracic effort was the gold standard elected to validate our algorithm. Our results show an excellent agreement between infrared thermography (IRT) and ground truth. While in phase A a mean correlation of 0.983 and a root-mean-square error of 0.240 bpm (breaths per minute) was obtained, in phase B they hovered around 0.995 and 0.890 bpm, respectively. In sum, IRT may be a promising clinical alternative to conventional sensors. Additionally, multisensor data fusion contributes to an enhancement of RR estimation and robustness.