An adaptive Kalman filter approach for cardiorespiratory signal extraction and fusion of noncontacting sensors
 Jerome Foussier^{1}Email author,
 Daniel Teichmann^{1},
 Jing Jia^{2},
 Berno Misgeld^{1} and
 Steffen Leonhardt^{1}
DOI: 10.1186/147269471437
© Foussier et al.; licensee BioMed Central Ltd. 2014
Received: 5 November 2013
Accepted: 29 April 2014
Published: 9 May 2014
Abstract
Background
Extracting cardiorespiratory signals from noninvasive and noncontacting sensor arrangements, i.e. magnetic induction sensors, is a challenging task. The respiratory and cardiac signals are mixed on top of a large and timevarying offset and are likely to be disturbed by measurement noise. Basic filtering techniques fail to extract relevant information for monitoring purposes.
Methods
We present a realtime filtering system based on an adaptive Kalman filter approach that separates signal offsets, respiratory and heart signals from three different sensor channels. It continuously estimates respiration and heart rates, which are fed back into the system model to enhance performance. Sensor and system noise covariance matrices are automatically adapted to the aimed application, thus improving the signal separation capabilities. We apply the filtering to two different subjects with different heart rates and sensor properties and compare the results to the nonadaptive version of the same Kalman filter. Also, the performance, depending on the initialization of the filters, is analyzed using three different configurations ranging from best to worst case.
Results
Extracted data are compared with reference heart rates derived from a standard pulsephotoplethysmographic sensor and respiration rates from a flowmeter. In the worst case for one of the subjects the adaptive filter obtains mean errors (standard deviations) of 0.2 min ^{−1} (0.3 min ^{−1}) and 0.7 bpm (1.7 bpm) (compared to 0.2 min ^{−1} (0.4 min ^{−1}) and 42.0 bpm (6.1 bpm) for the nonadaptive filter) for respiration and heart rate, respectively. In bad conditions the heart rate is only correctly measurable when the Kalman matrices are adapted to the target sensor signals. Also, the reduced mean error between the extracted offset and the raw sensor signal shows that adapting the Kalman filter continuously improves the ability to separate the desired signals from the raw sensor data. The average total computational time needed for the Kalman filters is under 25% of the total signal length rendering it possible to perform the filtering in realtime.
Conclusions
It is possible to measure in realtime heart and breathing rates using an adaptive Kalman filter approach. Adapting the Kalman filter matrices improves the estimation results and makes the filter universally deployable when measuring cardiorespiratory signals.
Keywords
Adaptive Kalman filter Sensor fusion Heart rate Breathing rate Signal processingBackground
The increase in both life quality expectancy and quality of life, together with improvements in medical support, has led to an increase in the mean age of the population in developed lands. The American Administration on Aging (AoA) predicts that, compared with the year 2000, the absolute number of people aged >65 years will double by 2030 and represent 19% of the total U.S. population [1]. The increase in the elderly population will lead to additional strains on the healthcare system. Therefore, there is an explicit need to transfer clinical measurement devices and technology to the patient’s home. The goal of personal health care is to relieve clinicians and the clinical infrastructure by means of technical improvements, but without reduction in diagnostic and rehabilitation performance, e.g. with telemonitoring at home [2]. This implies that new technology needs to be integrated into daily activity which, compared with a welldefined clinical environment, poses considerable challenges in terms of signal acquisition and processing. More complex processing algorithms are needed to overcome noise, artifacts and multisensor problems. One algorithmic approach is the use of the Kalman filtering technique.
Since its introduction in the 1960s, the Kalman filter has become a well accepted and stateoftheart approach for many applications, especially in the technical domain. Initially, computational capacity was limited and costly, making it difficult to use Kalman filters in realtime applications. However, over the years computational power has increased, also outside the personal computer domain. Today, high performance devices are available, including microcontrollers, digital signal processors or specialized computational units such as Field Programmable Gate Arrays (FPGA). Therefore, it is now possible to move complex mathematical computations into such devices working in a selfsufficient way. Many examples of technical applications using Kalman filters in realtime have been described [3, 4]. In biomedical engineering, Kalman filters are often applied to smooth or extract physiological signals, such as respiration and cardiac activity [5, 6]. Also in the domain of Electrical Impedance Tomography (EIT), the Kalman filter is able to track fast changes in impedance [7]. The Kalman filter is adequate for the present work, as it is possible to integrate prior knowledge (e.g. about sensor or system noise or state transitions). In addition, the Kalman filter is capable to denoise, separate signals or fuse sensor data, all in one architecture. Compared to other filtering or signal separation methods the Kalman filter is able to perform in realtime with very little systemic delays.
To record evaluation data a noncontact measurement technique, called magnetic induction monitoring, was chosen. This technique is a relevant method since it comprises a variety of problems typical for noncontact monitoring of vital signs. The complete employed measurement setup is described in Section ‘Sensors and measurement setup’. Afterwards, the detailed working principle and the implementation of the Kalman filter is explained (Section ‘Kalman filter’). Then, heart and respiration rate extraction results of two different subjects are shown and discussed in Section ‘Results and discussion’. We oppose the nonadaptive, described in earlier work, to the developed adaptive Kalman filter and show the performance increase, especially when no a priori knowledge about the sensor system is present. Finally, the conclusions summarize the results and gives an outlook to future work (see Section ‘Conclusions’).
Methods
Sensors and measurement setup
The signal were acquired in voluntary selfexperiments of the first two authors (JF and DT). Both signed an informed consent to participate in this study. The local ethics committee decided that this type of study with this device was not in the scope of their responsibility (internal reference number EK 013/14). In addition it was stated that the selfmeasured data can be used for publication. Both signed an informed consent to participate in this study. Consider, that this work is not conceived to be an extensive study but rather a proof of concept. Magnetic induction monitoring is a noncontact technique for recording thoracic activity. The technique is based on magnetic coupling between the thorax and a nearby sensorcoil, as developed by Teichmann et al.[8, 9]. As the coil has no conductive contact with the skin it is called a noncontact technique. The sensorcoil is driven by an alternating current and sends out an alternating magnetic field. This field induces eddy currents within conductive objects in the vicinity of the coil. In turn, these eddy currents reinduce a secondary alternating magnetic field, which affects the primary one and thereby changes the reflective impedance of the coil. If the coil is placed near the thorax, cardiorespiratory activity modulates the impedance of the coil due to motions of inner organs and the thoracic wall and/or because of changes in conductivity, e.g. caused by more or less air in the lungs or blood shifts in the heart. In this way, respiration and pulse can be easily obtained by measuring the impedance changes of the sensorcoil.
In summary, the Kalman filter has to deal with the following signal properties:

A very high offset compared to the signal amplitude

Respiration is visible in the time domain

Heart activity is only slightly visible in the time domain (very low ratio of heart to respiration signal level)

Good signaltonoise ratio (SNR) for the respiratory signal, since noise is not noticeable in the time domain for this sensor

Higher harmonics of the respiratory signal may overlap the pulse signal in the frequency domain
Kalman filter
In 1960, R.E. Kalman reported a new method for linear filtering and solving problems related to prediction [13]. Generally, the socalled “Kalman filter” consists of mathematical equations that represent an efficient way to predict a future and/or unknown state of a system, based only on the use of the preceding step. The calculations are very efficient so that they can be performed and implemented in today’s standard default personal computers, in digital signal processors (DSP) and in microcontrollers, to work in realtime applications. When process and sensor noise have timedependent characteristics, the filter is also called a “timevarying Kalman filter”. The filter finds a prediction value $\widehat{\mathbf{x}}$ with minimum variance for a disturbed state vector x[14]. The transitions from one state to another are represented in the state transition matrix A when no disturbance is present.
System and measurement noise, which are assumed to be white (zero mean) and statistically independent from each other, are represented by w and v, with their covariance matrices Q and R (both hermitian symmetric and positive semidefinite), respectively. External disturbances (e.g. systematic errors) are fed into the system with the control vector u and the matrix B, that describe the dynamics of the disturbance. Finally, the measurement matrix H describes the integration of a real measurement into the filter procedure.
The prediction step, also called “time update”, is described with (1) and (2). The aim of this step is to minimize the covariance of the estimation error which represents a degree of uncertainty of the estimation [14]. Note that both equations are only valid for k>0. The initial values of ${\widehat{\mathbf{x}}}_{0}$ and P_{0} have to be determined before the first iteration. The control input u_{ k } is not used in this model and is set to the zero vector, thus the matrix B_{ k } in (1) can be ignored.
The Kalman gain K_{ k } defined in (3) weights the innovation with respect to the measurement error covariance matrix R_{ k } and the estimation error covariance matrix ${P}_{k}^{}$ directly related to Q_{ k }. Higher values in R_{ k } give more importance to the real measurement, whereas higher values for Q_{ k } put more trust in the estimation of a state. The choice of R_{ k } and Q_{ k } is crucial for optimal filter performance. The correction step, also called “measurement update”, integrates the innovation of a new measurement z_{ k } to the estimated measurement in (4). Finally, the covariance matrix of the estimation error P_{ k } is updatedin (5).
The advantage of a Kalman filter is that it does not act as a pure filter, but also signal separation and fusion are realizable in a single implementation. Signal separation is done by defining several system states that are desired as separate outputs of the filter. Fusion is realized by expanding the measurement matrix H or H_{ k } to multiple sensors. With this, the Kalman technique is applicable to a wide field of applications, especially where no exact system model is known [14].
Signal extraction and separation
with ω_{s,k}=2π·f_{s,k} and ω_{f,k}=2π·f_{f,k} being the time changing angular frequencies of the breathing frequency f_{s,k} and the heart beat rate f_{f,k} at time step k, respectively.
The frequency measurement is based on a simple peak detector which extracts the location of maxima and minima in the signal and then computes the mean interval length between maximal and minimal points. The main advantage of a peak detector is the reduced computational complexity compared to other frequency measurement methods, e.g. based on the spectrum analysis. To accurately measure respiration and heart rate, the algorithm needs to buffer the output signal of the Kalman filter, where the denoted “Breathing” and “Heart” feedbacks in Figure 1 are represented by the internal Kalman states X_{s,k} and X_{f,k}, respectively. Note that it is not possible to estimate the heart rate directly from the raw sensor signal, due to the small heart signal amplitude and a possible frequency overlap of respiratory harmonics. A buffer of 20 s (approx. 1900 samples) for respiration and of 10 s (approx. 950 samples) for the heart signal were chosen to guarantee at least two consecutive respiratory cycles and several heart beats within the buffer. However, having such a large buffer causes systematic delays in the frequency estimation of half of the buffer length, i.e. 10 s and 5 s, respectively. Generally, because heart and respiration rates change slowly, we do not need to adapt the system model at 95 Hz. In fact, only at every tenth sample a frequency measurement is performed, resulting in an effective estimation rate of 9.5 Hz reducing computational effort by 90%. The remaining nine samples are filled with the last valid estimation to keep the same sampling rate of 95 Hz as the rest of the Kalman filter. This high sampling rate is needed to keep systematic errors at a minimum. At 95 Hz, the maximum absolute time resolution to measure the time between two signal peaks is $\frac{1}{95\phantom{\rule{1em}{0ex}}\text{Hz}}=10.5\phantom{\rule{1em}{0ex}}\text{ms}$, inducing a measurement error increasing with the heart rate with 0.0175% per beat per minute (bpm) (e.g. at 80 bpm the relative error is up to 1.4% equivalent to an absolute error of 1.12 bpm). The frequency measurement procedure has been validated with simulated sinusoidal signals covering heart rates from 60 bpm to 120 bpm and respiration rates from 12 min−1 to 15 min−1.
For stability reasons, abrupt changes in the frequency measurement need to be avoided, e.g. false estimations of the frequencies or artifacts, directly fed into the state transition matrix A_{ k } of the adaptive Kalman filter. Therefore, the estimated frequencies representing the heart and the respiration rates are each lowpass filtered with first order infinite impulse response (IIR) Butterworth filters, designed with the Matlab R2011a (MathWorks) “Signal Processing Toolbox” with cutoff frequencies of 0.1 Hz and 0.05 Hz, respectively.
Note that all estimations that are described in this work have been computed every tenth sample and lowpass filtered in the same way as the heart rate estimation value as described previously (first order Butterworth filter with 0.1 Hz cutoff frequency). Also, the development of all the described estimations first has been applied to simulated signals, where noise and standard deviations of the heart and respiratory signals were known.
Signal filtering
with variances of the longterm trend ${\mathit{\sigma}}_{\mathbf{trend},k}={\left({\sigma}_{\mathit{\text{trend}},k}^{\mathbf{1}}\phantom{\rule{1em}{0ex}}{\sigma}_{\mathit{\text{trend}},k}^{\mathbf{2}}\phantom{\rule{1em}{0ex}}{\sigma}_{\mathit{\text{trend}},k}^{\mathbf{3}}\right)}^{T}$ of sensors 13.
with ${\widehat{\sigma}}_{\mathit{\text{resp}},k}^{\mathbf{i}}$ as standard deviation estimation of the respiration using the longterm segment l_{i,Δ} over the last 20 s (approx. 1900 samples). We defined that the trend variation corresponds to 1% of the respiratory variation. Lower entries in the matrix (10) mean that more trust is placed on the predicted model states than on the raw measurement. Higher frequencies also cause higher amplitudes in the differentiated Kalman state vector signals V_{f,k} and V_{s,k} and are proportional to ${\omega}_{f,k}^{2}$ and ${\omega}_{s,k}^{2}$, respectively. A high longterm variance in the sensor signal is assumed to be caused by respiration and heart beat. The values for the states X_{s,k} and X_{f,k} defined in eq. (6) normally are smaller than for V_{s,k} and V_{f,k}. With this, more trust is given to the system model than to the measurement resulting in smoothed sinusoidalshaped signals with angular frequencies of around ω_{s,k} and ω_{f,k}, respectively. This simplifies the detection of peaks in the signal and improves frequency estimation. In fact, the states V_{s,k} and V_{f,k} contain more actual sensor information, thus more noise and possible artifacts are present; these states would need a processing step before being used for frequency estimation.
Sensor fusion
If no a priori information is given, e.g. when the sensor location is unknown, the s c a l_{ f } and s c a l_{ s } settings should be chosen as (1 1 1)^{ T } to give the same scaling to all sensors.
However, the offset links cannot be changed as they act totally independently. In (3) and (4) the sensor fusion is performed, as all measurement channels are joined into the estimated state vector ${\widehat{\mathbf{x}}}_{\mathbf{k}}$ with the Kalman Gain K_{ k }.
Results and discussion
In this section, the performance of the nonadaptive and the adaptive Kalman filter is evaluated by comparing the estimated respiration rates and heart rates with the reference flow and PPG signals. Instead of the peak detection and filtering methods described in Section ‘Signal extraction and separation’ we applied a spectrum based frequency estimator to the reference flow and PPG signals to obtain much higher frequency resolution. This estimator is not adequate for realtime application since it is computationally much more complex. Note that in the first 22.5 s (i.e. the minimal length of the buffers used for the respiration frequency and signal variance estimations), frequency estimation neither for the respiration rate nor for the heart rate is computed. Especially for the adaptive Kalman filter, it is important to regulate the frequencies in the model only when both frequencies are really measurable thus only when the Kalman filter is settled. This avoids the filter reaching an irreversible incorrect state. Therefore, during further analysis, the first 22.5 s are ignored as generally no valid and comparable data are available.
For the performance analysis between the adaptive and nonadaptive Kalman filter, we defined three different settings:

SE: When the Sensor Estimation settings is used, the whole raw sensor signal is first analyzed, giving the optimal estimations of variances. Note that this setting only works in a postprocessing way, but it is used to create the bestcase setting in this work.

DS: The Default Settings should adequately work in most cases.

BS: Bad Settings that give a very inappropriate definition of the system  the worstcase scenario when the user misconfigures the Kalman filter.
Initialization parameters using different settings for subject 1 and 2
SE  SE  DS  DS  BS  BS  

Subject  1  2  1  2  1  2 
f _{ f } [Hz]  1.3  1.7  1.5  1.5  1.0  1.0 
f _{ s } [Hz]  0.3  0.2  0.1  0.1  0.1  0.1 
${\widehat{\sigma}}_{\mathit{\text{trend}},0}^{\mathbf{1}}$  989.0  54.1  100.0  100.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{trend}},0}^{\mathbf{2}}$  159.2  819.9  100.0  100.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{trend}},0}^{\mathbf{3}}$  1057.7  1801.5  100.0  100.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{noise}},0}^{\mathbf{1}}$  53.6  6.7  10.0  10.0  1000.0  1000.0 
${\widehat{\sigma}}_{\mathit{\text{noise}},0}^{\mathbf{2}}$  11.2  12.2  10.0  10.0  1000.0  1000.0 
${\widehat{\sigma}}_{\mathit{\text{noise}},0}^{\mathbf{3}}$  64.2  15.6  10.0  10.0  1000.0  1000.0 
${\widehat{\sigma}}_{\mathit{\text{heart}},0}^{\mathbf{1}}$  1112.1  70.1  100.0  100.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{heart}},0}^{\mathbf{2}}$  195.8  198.8  100.0  100.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{heart}},0}^{\mathbf{3}}$  1605.8  454.8  100.0  100.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{resp}},0}^{\mathbf{1}}$  10195.7  191.2  10000.0  10000.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{resp}},0}^{\mathbf{2}}$  1114.4  4137.1  10000.0  10000.0  1.0  1.0 
${\widehat{\sigma}}_{\mathit{\text{resp}},0}^{\mathbf{3}}$  8498.0  5846.5  10000.0  10000.0  1.0  1.0 
$\mathit{\text{sca}}{l}_{f}^{\mathbf{1}}$  1.0  1.0  1.0  1.0  1.0  1.0 
$\mathit{\text{sca}}{l}_{f}^{\mathbf{2}}$  0.1  0.1  0.1  0.1  0.1  0.1 
$\mathit{\text{sca}}{l}_{f}^{\mathbf{3}}$  0.1  0.1  0.1  0.1  0.1  0.1 
$\mathit{\text{sca}}{l}_{s}^{\mathbf{1}}$  0.1  0.1  0.1  0.1  0.1  0.1 
$\mathit{\text{sca}}{l}_{s}^{\mathbf{2}}$  1.0  1.0  1.0  1.0  1.0  1.0 
$\mathit{\text{sca}}{l}_{s}^{\mathbf{3}}$  1.0  1.0  1.0  1.0  1.0  1.0 
Mean Error (standard deviation) between flow/PPG reference and estimated respiration/heart rates for the nonadaptive and adaptive (Ad.) Kalman filter
Subj.  Resp.  Resp. Ad.  Heart  Heart Ad.  

[min^{−1}]  [min^{−1}]  [bpm]  [bpm]  
SE  1  0.0 (0.3)  0.0 (0.3)  0.3 (2.1)  0.2 (1.4) 
SE  2  0.3 (0.5)  0.2 (0.3)  0.8 (2.3)  0.7 (1.7) 
DS  1  0.1 (0.3)  0.0 (0.3)  1.2 (2.3)  0.2 (1.5) 
DS  2  0.3 (0.4)  0.2 (0.3)  0.3 (2.2)  0.8 (1.7) 
BS  1  0.1 (0.2)  0.0 (0.3)  11.2 (5.7)  0.2 (1.5) 
BS  2  0.2 (0.4)  0.2 (0.3)  42.0 (6.1)  0.7 (1.7) 
Mean Error (standard deviation) between estimated offset C_{ 1 , k } C_{ 3 , k }and raw sensor signals S1S3 for the nonadaptive and adaptive (Ad.) Kalman filter
Subj.  S1C_{1,k}  S1C_{1,k}Ad.  S2C_{2,k}  S2C_{2,k}Ad.  S3C_{3,k}  S3C_{3,k}Ad. 

1 (SE)  16 (4850)  1 (1259)  0 (644)  11 (571)  3 (4674)  84 (4027) 
2 (SE)  1 (111)  0 (88)  304 (4016)  263 (3960)  429 (5634)  352 (6028) 
1 (DS)  817 (4039)  2 (1253)  8300 (4172)  21 (574)  8302 (4119)  171 (4049) 
2 (DS)  882 (480)  0 (88)  8832 (4996)  246 (3934)  8832 (4967)  334 (6009) 
1 (BS)  861 (11310)  1 (1248)  203 (1264)  8 (569)  1693 (9014)  53 (4008) 
2 (BS)  3 (245)  0 (88)  698 (4239)  314 (3971)  498 (6483)  407 (6033) 
To assess whether the described procedure is able to work in realtime, computational time measurements were conducted. The filter procedure was implemented in Matlab R2011a (MathWorks) on a personal computer with a 3.10 GHz dual core processor and 4 GB RAM. In total the adaptive Kalman filter (including all initialization processes and filtering, but excluding reference frequency measurements) in average 35.9 s (24.79% of the total 144.8 s) and 40.5 s (24.80% of the total 163.4 s) processing times are required, respectively for subject 1 and 2. Without adaptation, the procedures take in average 24.2 s (16.73% of the total 144.8 s) and 24.7 s (15.08% of the total 163.4 s). Hence, performing one regular Kalman filter step requires about 1.7 ms out of 10.5 ms at 95 Hz sampling rate. The effective adaptation rate is reduced to 9.5 Hz (every tenth sample equal to 105 ms sampling time) and in average about 5060% more computational time is needed, i.e. approximately 2.6 ms per sample. Note, that the Kalman filter computations still are at 95 Hz sampling rate. The surplus of 9 ms within the 105 ms window (ten samples) compared to the nonadaptive filter only occurs during the first adaptation sample which takes 1.7 ms+9 ms=10.7 ms in total. This is slightly more than the sampling time of 10.5 ms resulting in a maximum lag of one sample every tenth sample. Therefore we conclude that the intrinsic computations of both filters are realtime compliant for separating respiratory and cardiac signal activity. However, as described in Section ‘Signal extraction and separation’, buffer sizes of 20 s and 10 s are still necessary to estimate breathing and heart rates, respectively. Note that, until now, no special code optimization has been performed, which would further reduce the required computational time and make it possible to integrate the procedures into microcontrollers or digital signal processing units.
Although the signals were acquired on healthy young men in this proof of concept, a target application might be the telemonitoring of elderly at home. The main advantage of this technique is the contactless and unobtrusive way the measurement are performed. The sensors may be arranged in a shirt, a bed or in a chair just to give some examples. Nevertheless, it has to be examined how daily activities (e.g. cooking, vacuuming, mowing the lawn or walking around) affect the performance of detecting breathing and heart rates. The main drawback of the magnetic induction measurement method is that movements relative to the coils cause large artifacts in the signal on top of the desired signal. These may be detected or in a good case even be compensated. We do not expect that the performance is age dependent. Rather the position of the sensors is more crucial for optimal performance. During resting activities (e.g. watching TV, sitting in a chair reading or sleeping) where the number of body movements is small, the developed filter should be able to perform as well as described since it adapts automatically to the measurement conditions.
Conclusions
The above analysis shows that estimations of respiration and heart rate based on the implemented adaptive Kalman filter perform very well in a realtime acquisition scenario employing contactless sensors measuring cardiorespiratory signals. The direct evaluation of the respiration and heart rate only based on the Kalman filter states and the direct feedback into the time variant state transition matrix A_{ k } improved the results compared with the nonadaptive procedure. It is also important that the filter does not require a specific signal shape as long as it contains periodic content for respiration and heart activity; therefore, the filter is suitable for all biomedical signals containing two periodic motions on top of large signal offsets. The delays caused by the buffer windows of 20 s for measurement of the respiration rate and of 10 s for the heart rate are necessary to correctly estimate the frequencies with the described peak detection method. The frequency measurement resolution of 0.0175% per bpm inherently generates errors, especially with increased heart rates. For shorter delays and a better frequency resolution, the frequency estimation method needs to be improved, e.g. by upsampling the signal in the buffer or employing more complex detection algorithms but without neglecting the realtime ability.
In general, this implementation of the adaptive Kalman filter is kept as simple as possible; this allows to port the code to embedded processing units, e.g. microcontrollers or digital signal processors. It is also possible to expand the system to more than three sensors by adding one row and one column in the state transition matrix A_{ k }, the system noise covariance matrix Q_{ k } and the measurement noise covariance matrix R_{ k } and only one row in the measurement matrix H.
Since this analysis is a proof of concept, a larger study with additional subjects needs to be performed to validate all of the above methods.
Authors’ information
Jérôme Foussier was born in Cologne, Germany, in 1984. He holds a Dipl.Ing. degree in Electrical Engineering from RWTH Aachen University, Germany. Currently, he is pursuing the Dr.Ing. (Ph.D.) degree at the Chair of Medical Information Technology, RWTH Aachen University, where he is also working as a Research Assistant. His research interests include signal processing and classification as well as physiological measurement techniques.
Daniel Teichmann was born in Essen, Germany, in 1982. He holds a Dipl.Ing. degree in Electrical Engineering from RWTH Aachen University, Germany. Currently, he is pursuing the Dr.Ing. (Ph.D.) degree at the Chair of Medical Information Technology, RWTH Aachen University, where he is also working as a Research Assistant. His research interests include noncontact monitoring techniques and signal processing.
Jing Jia was born in Shanghai, China, in 1985. She holds a Dipl.Ing. degree in Electrical Engineering as well as in Economic Science from RWTH Aachen University, Germany. Currently, she is working at ’Philips Medizin Systeme Böblingen GmbH’ in Böblingen, Germany.
Berno J.E. Misgeld received the Dipl.Ing. (FH) in Electrical and Automation Engineering from University of Applied Sciences, Aachen, Germany, and the M.Sc. degree from Coventry University, Coventry, U.K., in 2003, respectively. In 2007 he received the Dr.Ing. degree in Biomedical and Control Engineering from RuhrUniversity Bochum, Bochum, Germany. From 2006 to 2011 he was a research and development engineer for guidance and flight control systems at DiehlBGTDefence, Ueberlingen, Germany. Since 2011 he is a Senior Scientific Engineer in Biomechatronical Systems and Rehabilitation Robotics at the Chair of Medical Information Technology at RWTH Aachen University, Aachen, Germany. His research interests include feedback control and filtering with application to biomedical systems, robotics and medicine.
Steffen Leonhardt was born in Frankfurt, Germany, in 1961. He holds an M.S. in Computer Engineering from SUNY at Buffalo, NY, USA, a Dipl.Ing. in Electrical Engineering and a Dr.Ing. degree in Control Engineering from the Technische Universität Darmstadt, Germany, and a M.D. in Medicine from J. W. Goethe University, Frankfurt, Germany. He has 5 years of R&D management experience in medical engineering industry and was appointed Full Professor and Head of the Philips endowed Chair of Medical Information Technology at RWTH Aachen University, Germany, in 2003. His research interests include physiological measurement techniques, personal health care systems and feedback control systems in medicine.
Declarations
Acknowledgements
We would like to thank Laraine VisserIsles for her insightful comments and the language review of the manuscript.
Authors’ Affiliations
References
 Administration on Aging:Aging statistics  department of health & human services. 2012, [http://www.aoa.gov/aoaroot/aging\_statistics/index.aspx],
 Meystre S:The current state of telemonitoring: a comment on the literature. Telemed J eHealth. 2005, 11: 6369. 10.1089/tmj.2005.11.63.View ArticlePubMed
 Moreno V, Pigazo A (eds.): Kalman Filter: Recent Advances and Applications. 2009, Rijeka (Croatia): InTech
 Chui CK, Chen G: Kalman Filtering with RealTime Applications. Springer Ser. Info. Sci., Vol. 17. 2009, BerlinHeidelberg (Germany): Springer
 Spincemaille P, Nguyen TD, Prince MR, Wang Y:Kalman filtering for realtime navigator processing. Magn Reson Med. 2008, 60: 158168. 10.1002/mrm.21649.View ArticlePubMed
 Sayadi O, Shamsollahi MB:ECG denoising and compression using a modified extended Kalman filter structure. IEEE Trans on Biomed Eng. 2008, 55 (9): 22408.View Article
 Vauhkonen M, Kaipio JP, Karjalainen Pa:A Kalman filter approach to track fast impedance changes in electrical impedance tomography. IEEE Trans on Biomed Eng. 1998, 45 (4): 486493. 10.1109/10.664204.View Article
 Teichmann D, Foussier J, Leonhardt S:Respiration monitoring based on magnetic induction using a single coil. Biomedical Circuits and Systems Conference (BioCAS), 2. 2010, Paphos, Cyprus: IEEE, 3740.View Article
 Teichmann D, Foussier J, Jia J, Leonhardt S, Walter M:Noncontact monitoring of cardiorespiratory activity by electromagnetic coupling. IEEE Trans on Biomed Eng. 2013, 60 (8): 21422152.View Article
 Teichmann D, Foussier J, Buscher M, Walter M, Leonhardt S:Textile integration of a magnetic induction sensor for monitoring of cardiorespiratory activity. IFMBE Proceedings of World Congress on Medical Physics and Biomedical Engineering, May 2631, 2012, Beijing, China, Volume 39. 2013, BerlinHeidelberg (Germany): Springer, 13501353.View Article
 Teichmann D, Kuhn A, Leonhardt S, Walter M:Human motion classification based on a textile integrated and wearable sensor array. Physiol Meas. 2013, 34 (9): 963975. 10.1088/09673334/34/9/963.View ArticlePubMed
 Foussier J, Teichmann D, Jia J, Leonhardt S:Fusion of noncontacting sensors and vital parameter extraction using Kalman filtering. Proceedings of the World Congress on Engineering 2011, WCE 2011, July 6  8 2011, London, U.K., Volume II. 2011, HongKong (China): IAENG, 15921596.
 Kalman R:A new approach to linear filtering and prediction problems. Trans ASMEJ Basic Eng. 1960, 82: 3545.View Article
 Welch G, Bishop G:An introduction to the Kalman filter. Tech. rep1995,
 The prepublication history for this paper can be accessed here:http://www.biomedcentral.com/14726947/14/37/prepub
Prepublication history
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.