gms | German Medical Science

GMDS 2013: 58. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie

01. - 05.09.2013, Lübeck

Camera-based monitoring of cardio-respiratory signals – combining image and signal processing

Meeting Abstract

  • Sebastian Zaunseder - TU Dresden, Dresden, DE
  • Alexander Trumpp - TU Dresden, Dresden, DE
  • Georg Lempe - TU Dresden, Dresden, DE
  • Hagen Malberg - TU Dresden, Dresden, DE

GMDS 2013. 58. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V. (GMDS). Lübeck, 01.-05.09.2013. Düsseldorf: German Medical Science GMS Publishing House; 2013. DocAbstr.338

doi: 10.3205/13gmds269, urn:nbn:de:0183-13gmds2697

Published: August 27, 2013

© 2013 Zaunseder et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License ( You are free: to Share – to copy, distribute and transmit the work, provided the original author and source are credited.



Introduction: Lately it was shown that by using videos it is possible to acquire cardio-respiratory signals [1], [2]. The camera based assessment offers distinguished recording conditions – a non-contact acquisition at a distance is feasible – at the expense of a low signal-to-noise ratio. Therefore, real-world applications, i.e. recordings under out-of-laboratory conditions, will require the coordinated combination of image and signal processing techniques. This opens a field of joint image and signal processing for the future. This contribution investigates the applicability of a Bayesian skin classifier and a semi-automated method which combines distributed regions of interests (ROIs) by using independent component analysis (ICA) in order to extract the heart rate (HR) from videos.

Materials and Methods: The pulsatile blood volume in surface near vessels causes changes in the optical absorption which can be assessed from image sequences. To evaluate the subtle changes a ROI is defined from which the measurement signal is derived. Most often the measurement signal is formed by averaging the green channel of all pixels which belong to the ROI. In order to define a ROI in an automated way and extract the measurement signal consecutively we implemented a Bayesian skin classifier (method 1). Classification is done on the basis of a RGB skin color model which was derived from labeled (skin and non-skin areas) images. Secondly, we implemented an algorithm which combines the green channel of three predefined regions (both cheeks and the forehead [3]) by using ICA in order to obtain the measurement signal (method 2). Both methods yield a measurement signal from which the HR is derived in the frequency domain. The results were compared to the HR as derived from the plethysmogram. The evaluation of both methods was done using data acquired at our measurement site which allows for a synchronous recording of videos and reference signals [3]. We incorporated 20 healthy subjects at rest. Each record of 100 seconds was split up in 9 overlapping segments of 20 seconds resulting in overall 180 segments for which the HR was extracted.

Results: Bland Altman analysis qualitatively proves that both methods are similarly able to produce reliable estimates of the HR. 95% (methods 1) and 98% (method 2), respectively, of all measurements show a deviation of the HR from the reference HR smaller than 3 beats per minute (bpm). The limits of agreements (-12 bpm to 15 bpm and -9.6 bpm to 11 bpm for method 1 and method 2, respectively), however, are heavily influenced by single outliers which can be assumed to arise from slight movements already.

Discussion: Up to date the feasibility of using videos to extract cardio-respiratory parameters could be confirmed particularly in laboratory environments. Thereby, as in the current study, a suited combination of image and signal processing methods is crucial. As moderate motion and the illumination heavily affect the measurement this combination, however, must be enforced and refined in the future. Particularly the interaction should be analyzed in order to benefit from the strengths of both fields.


Blazek V, Wu T, Hoelscher D. Near-infrared CCD imaging: possibilities for noninvasive and contactless 2D mapping of dermal venous hemodynamics. Proceedings Optical Diagnostics of Biological Fluids. 2000; V: 2-9.
Fei J, Pavlidis I. Thermistor at a Distance: Unobtrusive Measurement of Breathing. IEEE Transactions on Biomedical Engineering. 2010;57:988-998.
Lempe G, Zaunseder S, Wirthgen T, Zipser S, Malberg H. [Camera-based assessment of cardio-respiratory signals] (German). Technisches Messen. 2013;80:179-184.