WO2013038326A1

WO2013038326A1 - Distortion reduced signal detection

Info

Publication number: WO2013038326A1
Application number: PCT/IB2012/054694
Authority: WO
Inventors: Gerard De Haan
Original assignee: Koninklijke Philips Electronics N.V.
Priority date: 2011-09-13
Filing date: 2012-09-10
Publication date: 2013-03-21

Abstract

The present invention relates to a device and a method for extracting information from remotely detected characteristic signals. A data stream (26) derivable from electromagnetic radiation (14) emitted or reflected by an object (12) is received. The data stream(26) comprises a continuous or discrete characteristic signal (80) including physiological information and a disturbing signal portion. The physiological information is representative of at least one at least partially periodic vital signal (20). The disturbing signal portion is representative of at least one of an object motion portion and/or a non-indicative reflection portion. A relevant frequency band (92) of the data stream (26) is split into at least two defined sub bands (34a, 34b, 34c) comprising determined portions of the characteristic signal (80), each of which representing a defined temporal frequency portion potentially being of interest. The sub bands (34a, 34b, 34c) are optimized so as to derive optimized sub bands (38a, 38b, 38c), the optimized sub bands (38a, 38b, 38c) being at least partially indicative of a presence of the vital signal (20). The at least two optimized sub bands (38a, 38b, 38c) are combined so as to compose an optimized processed signal (42). Consequently, the desired vital signal (20) can be accentuated in the optimized processed signal (42).

Description

Distortion reduced signal detection

FIELD OF THE INVENTION

The present invention relates to a device and method for extracting

information from remotely detected characteristic signals, wherein the characteristic signals are embedded in a data stream derivable from electromagnetic radiation, in particular wherein the data stream comprises a continuous or discrete characteristic signal including physiological information and a disturbing signal portion, the physiological information being representative of at least one at least partially periodic vital signal of a remote object of interest, the disturbing signal portion being representative of at least one of an object motion portion and/or a non-indicative reflection portion.. The invention further addresses distortion reduced signal detection.

BACKGROUND OF THE INVENTION

US 7,403,806 B2 discloses a method for use in pulse oximetry, comprising the steps of:

- receiving a time-based signal reflective of one or more optical signals incident on a detector of a pulse oximeter;

performing a correction with respect to a portion of said received time-based signal to reduce an effect of artifact, thereby providing a processed time-based signal;

performing a transform on said processed time-based signal to obtain transformed information relative to a second domain; and

processing the transformed information to obtain physiological information regarding a patient;

wherein said step of performing a correction comprises prefiltering said received time based signal to reduce an amplitude of a portion of said signal determined to be affected by an artifact to a value less than an amplitude of a portion of the signal determined to be clean with reference to the artifact.

The document relies on determined signal portions supposed to be "clean". It is suggested to accordingly attenuate signal portions being corrupted by artifact. However, it is not disclosed how to detect the required clean portions. Even for standard photoplethysmographic approaches it has to be expected that recorded signals are almost always somehow corrupted by distortion and/or noise. Standard photoplethysmography (PPG) comprises obtrusive measurement, e.g. via a transceiver unit being fixed to an object's earlobe or fingertip. Typically, the object to be observed, e.g., a patient, is asked for staying still or performing certain distinct repeatable movements during measurement. However, feeling uncomfortable, unfortunately patients often tend to move during observations.

Undesired object motion adversely affects signal quality.

Standard PPG requires defined artificial light sources to be directly attached to an indicative surface, e.g. tissue, of the object to be observed. In this manner, it is aimed at avoiding or reducing adverse effects, e.g. potentially disturbing incident radiation caused by other light sources, or undesired object motion with respect to the light source.

Correspondingly, also the receiver or detector, e.g. at least a photodiode, is closely fixed to the object's tissue of interest. In case the transceiver unit is firmly fixed to the patient so as to avoid patient movement with respected to the equipment, signal quality can be deteriorated as well, e.g. due to undesired tissue compression. As a consequence, even for standard PPG, required steady conditions cannot be guaranteed for a considerable period of time.

The method of US 7,403,806 B2 requires preceding steps directed to

(pre)determine which portion of the detected signal exactly is a "clean" signal portion. The underlying problem can be considered a "chicken and egg" problem. Having detected the clean portion not affected by disturbances or noise, the correction for artifacts is fairly simple. Consequently, the desired physiological information can be extracted from the signals. However, the determination of "clean" signal portions indeed comprises extracting, analyzing or, at least, separating the desired physiological information represented by "clean" signal portions from artifact containing "distorted" signal portions. Thus, the required signal portion detection poses major challenges for signal optimization measures. Prior to the suggested signal correction, the signal of interest has to be extracted at least to a certain extent so as to allow a determination of the "clean" portions. Thus, the suggested method is potentially inadequate for distortion-prone and/or noise-prone monitoring applications.

Moreover, obtrusive measurement is often experienced as being unpleasant. By contrast, remote PPG approaches apply unobtrusive measurement. Basically, remote photoplethysmography utilizes light sources, or, in general, radiation sources, disposed remote from the object of interest. Preferably, for some applications, even readily available existing light sources rather than defined special-purpose light sources are utilized. For instance, artificial light sources and/or natural light sources can be exploited. It could be further envisaged to utilize ambient (indirect) light. Consequently, it has to be assumed that due to widely changing illuminations conditions the detected signals (comprising the desired signals) generally provide a very small signal-to-noise ratio. Similarly, also a detector, e.g., a camera, can be disposed remote from the object of interest. In this manner, remote photoplethysmography systems and devices can be adapted for everyday applications, e.g., for unobtrusive inpatient and outpatient (home) monitoring or even leisure and fitness applications. Basically, it is intended that monitored objects can enjoy a certain degree of freedom of movement during remote PPG measurement.

Consequently, compared with standard (obtrusive) photoplethysmography, remote (unobtrusive) photoplethysmography is far more susceptible to distortion and noise. Undesired object motion with respect to the detector and/or the radiation source can excessively influence the detected signals. Furthermore, remote photoplethysmography systems are often subjected to varying illumination conditions. Therefore, it can be assumed that the detected signals are almost always corrupted by noise and distortion. Hence, quite often "clean" signal portions simply cannot be determined and utilized for processing remaining "corrupted" signal portions.

WO 2011/042858 Al discloses a further method and system addressing processing a signal including at least a component representative of a periodic phenomenon in a living being. The document also refers to remote photoplethysmography. WO

2011/021128 A2 discloses a method and a system for image analysis remote

photoplethysmographic (PPG) analysis. Additional basic approaches to remote

photoplethysmography are described in Verkruysse, W. et al (2008), "Remote

plethysmographic imaging using ambient light" in Optics Express, Optical Society of America, Washington, D.C., USA, vol. 16, no. 26, pp. 21434-21445.

For remote PPG, the recorded data, such as captured reflected or emitted electromagnetic radiation, especially recorded image frames, always comprises, beside of the desired signal to be extracted therefrom, further signal components deriving from overall disturbances, by way of example, such as noise due to changing luminance conditions or a movement of observed objects. Hence, a detailed precise extraction of the desired signals still poses major challenges for the processing of such data.

Although considerable progress in the field of computing performance has been made, it is still a challenge to provide for instant image recognition and image processing enabling immediate, so to say, on-line detection of desired vital signals. This applies in particular to mobile device applications commonly lacking of sufficient computing power. Furthermore, data transmission capacity can be restricted in several applications.

A possible approach to this challenge may be directed to providing well- prepared and steady ambient conditions when capturing a signal of interest in which the desired signal component is embedded so as to minimize disturbing signal components overlaying the signal. However, such laboratory conditions cannot be transferred to everyday field applications as high efforts and preparation work would be required therefor. As a matter of fact, for remote PPG monitoring applications, unsteady conditions and considerably large disturbances have to be "accepted".

After all, vital signal detection is made even more difficult when amplitudes and/or nominal values of disturbing signal components are much larger than amplitudes and/or nominal values of desired signal components to be extracted. Potentially, the magnitude of difference between the respective components can be expected to even comprise several orders. This applies in particular for remote PPG.

SUMMARY OF THE INVENTION

It is therefore an object of the present invention to provide a device and a method for extracting information from detected characteristic signals addressing the aforementioned challenges and providing further refinements facilitating obtaining the desired signals from remote objects with higher accuracy.

Furthermore, it would be advantageous to provide a device and method even adapted for enabling an extraction of the desired signals under considerably poor ambient conditions, e.g. small signal-to-noise ratio, varying luminance conditions and/or steady or even unsteady movements of the object to be observed. It would be further advantageous to provide a device adapted for being less susceptible to disturbances influencing the captured signals to be processed and analyzed.

In a first aspect of the present invention a device for extracting information from remotely detected characteristic signals is presented, the device comprising:

an interface for receiving a data stream derivable from electromagnetic radiation reflected by a remote object, the data stream comprising a continuous or discrete time-based characteristic signal including physiological information and a disturbing signal portion, the physiological information being representative of at least one at least partially periodic vital signal, the disturbing signal portion being representative of at least one of an object motion portion and/or a non-indicative reflection portion, the characteristic signal being associated with a signal space, the signal space comprising complementary channels for representing the characteristic signal, components of the characteristic signal being related to respective complementary channels of the signal space,

a decomposition means for pre-processing the data stream by splitting a relevant frequency band thereof into at least two defined sub bands comprising determined portions of the characteristic signal, each of which representing a defined temporal frequency portion potentially being of interest,

a processing means for optimizing the sub bands so as to derive respective optimized sub bands from the at least two sub bands, the optimized sub bands being at least partially indicative of a presence of the vital signal,

a composition means or combining the optimized sub bands so as to compose an optimized processed signal.

The present invention is based on the insight that predetermined frequency based data stream decomposition prior to subsequent processing measures can be a suitable approach to the optimization of the potentially heavily corrupted characteristic signals.

Moreover, it is understood that prior systems for non-remote obtrusive PPG rely on complex preparation measures requiring "laboratory" conditions and disturbances fairly well-known in advance. Consequently, these prior approaches face huge feasibility challenges when applied to remote unobtrusive PPG widely exposed to noise and disturbances.

Basically, a filtered (split) characteristic signal can be efficiently analyzed for desired vital signals expected to be at least partially present in the filtered portions. In this connection, signal processing can "focus" on the predetermined sub bands. For each sub band out-of-band disturbances are no longer significant. The desired signal, by way of example, heart beat or respiration rate, changes considerably slow over time. Generally, abrupt changes are unlikely to happen. Consequently, it can be assumed that the desired signal is at least substantially comprised in one of the at least two defined sub bands. Hence, the remaining sub band(s) can be expected to basically exhibiting a zero-signal with respect to the desired signal. It shall be understood that the various sub bands can comprise separate portions or can partially overlap each other.

According to a preferred embodiment the sub band optimization is applied separately to the respective sub bands.

Predefined pre-filtering can be particularly well suited for applications requiring instant data processing and signal detection. For these applications, approximate filtering methods based on ex post signal analysis and accordingly adapted filter parameters face several drawbacks. This applies in particular to application fields subjected to poor signal-to-noise ratios.

Consequently, applying predefined data decomposition, preselected frequency components can be drawn from the data stream and undergo separate data processing. Due to a potentially improved signal-to-noise ratio, subsequent signal analyses can be simplified.

It goes without saying that further signal optimization measures can be applied to the data stream comprising the characteristic signals. These measures can comprise motion compensation, pattern detection, e.g., face detection, or normalization measures.

Normalization can render signal components at least partially independent from overall disturbances. In this context, it is reminded that under everyday conditions the signals of interest are considerably small compared to the non-indicative disturbances.

The data stream can comprise a data sequence, e.g., a series of image frames comprising color information, such as RGB images. The image frames can represent the object of interest and further elements. Basically, the further elements are not indicative of the desired signals to be extracted from the data stream. The data stream can comprise instantly captured data or data already captured and stored in advance.

There exist several embodiments of the decomposition means, the processing means and the composition means. In a first, fairly simple embodiment all of them are embodied by a processing unit, in particular a processing unit of a personal computer or a mobile device, which is driven by respective logic commands. Such a processing unit may also comprise suitable input and output interfaces.

However, in the alternative, each of the decomposition means, the processing means and the composition means can be embodied by a separate processing unit driven or driveable by respective commands. Hence, each respective processing unit can be adapted to its special purpose. Consequently, a distribution of tasks may be applied, wherein distinct tasks are processed, for instance, executed on a single processor of a multi-processor processing unit, or, again referring to a personal computer, image processing-related tasks are executed on an image processor while other operational tasks are executed on a central processing unit. The decomposition means can be embodied by filter elements having discrete components, or by digital filters.

According to an advantageous embodiment, the device further comprises an analyzing means for enhancing and/or detecting the vital signal in the composed optimized processed signal. Preferably the analyzing means is further adapted for filtering the optimized processed signal and for enhancing a signal component at a bandwidth between 0.05 Hz and 10 Hz, preferably between 0.5 Hz and 3.5 Hz. This can be considered a particularly appropriate range for heart beat measurement. For respiration rate measurement, for instance, the range can comprise frequency values between about 0.1 Hz and about 1 Hz. For the detection of Traube-Hering-Mayer waves, the range can comprise frequency values between about 0.05 Hz and about 0.2 Hz. Post-filtering of the composed optimized processed signal can further improve the signal-to-noise ratio. In this way, even further disturbing signal components non- indicative of the desired vital signals can be removed from the data stream.

According to a further aspect, the decomposition means further comprises a band pass filtering means for suppressing a selected non-indicative frequency component of the characteristic signal and/or for enhancing a selected indicative frequency component of the characteristic signal.

For instance, the band pass filtering means can be adapted to pass frequencies within a range between about 40 BPM (beats per minute) and about 220 BPM to attenuate frequencies outside that range. This range is fairly applicable for heart rate detection. For respiration rate measurement, the range can comprise values between about 8 BPM and about 60 BPM. For sleep monitoring, bedside patient monitoring and related applications, the desired frequency range could be arranged in the area of between range 40 BPM and 120 BPM. Oh the other hand, for fitness monitoring, especially higher frequencies can be of interest, e.g. the range of between about 80 BPM and 220 BPM.

According to an even further aspect, the decomposition means comprises a transformation means for splitting the characteristic signal into constituent frequencies, preferably the transformation means is adapted for applying a Fourier transform to the characteristic signal, more preferably a discrete Fourier transform and/or a fast Fourier transform, so as to detect relevant sub bands of the characteristic signal.

In the alternative also predetermined (fixed) sub bands can be utilized and processed. For instance, the relevant frequency band, e.g. a range between 40 BPM and 220 BPM, can be split into two or more complementary sub bands. It goes without saying that the sub bands can be further shaped by applying (sub) band pass filters, low pass filters and/or high pass filters to the characteristic signals. It can be also envisaged to utilize several filters.

It shall be understood that a combination of band pass filtering and band split filtering can further improve the signal detection. For instance, a combination of a fixed band pass filter passing a certain band and at least two further fixed filters splitting the pass band can considerably enhance the desired signals in the (re)composed optimized processed signal.

Several signal processing filters exhibiting certain filter characteristics can be envisaged, e.g. high-pass filters, low-pass filters, all-pass filters, equalization filters, or even band stop filters. Filter characteristics, e.g. filter steepness or filter damping, can be chosen accordingly. For instance, the filters can comprise Butterworth filters, Chebyshev filters, Bessel filters, and raised-cosine filters.

According to a further embodiment, the signal space is an additive color signal space, wherein the complementary channels are additive color channels, wherein the characteristic signal is represented by at least three absolute components, wherein the at least three absolute components represent distinct color components indicated by the additive channels, wherein the additive channels are related to defined spectral portions.

For instance, an RGB signal space may be applied. Alternative signal spaces may comprise or be derived from CIE XYZ, HSV, HSL, sRGB and xvYCC signals. Also derivates thereof can be utilized. It should be noted that basically linear RGB signals can be utilized for the desired signal detection. Therefore, non-linear signal spaces (e.g. gamma corrected signals) can be transformed accordingly. It can be further envisaged to combine several distinct signal spaces, at least partially, so as to provide a broader spectral basis for the required analyzing processes. For instance, so-called RGBY signals can be applied as well. In an RGBY signal space in addition to red, green and blue also yellow signals can carry color information.

In case the input data stream is related to a subtractive color model, e.g., CMYK, the data can be transferred accordingly so as to arrive at an additive signal space.

Further spectral components can be utilized for extracting the desired vital signal(s) from the data stream. In this connection, also infrared radiation components can be applied. For instance a ratio between red and infrared signals can be highly indicative of the desired signals.

Hence, it is understood that basically the desired signals can be extracted from a characteristic signal comprising at least two potentially indicative components.

According to another aspect of the device, the processing means further comprises a converter means for separately transferring the sub band portions of the characteristic signal by converting the at least three absolute components related to respective additive channels to at least two difference components of the characteristic signal, wherein each of the at least two difference components can be derived through a respective arithmetic transformation considering at least two of the at least three absolute components, wherein the arithmetic transformation comprises additive and subtractive coefficients, the disturbing signal portion being at least partially suppressed in the transferred optimized sub band portions, wherein the arithmetic transformation comprises an at least partially subtraction of at least one of the at least three absolute components from the remaining absolute

components, and wherein the arithmetic transformation for each of the at least two difference components comprises coefficients at least substantially summing to zero.

In essence, it shall be understood that the sub bands can be processed independently of each other.

Object motion and changing illumination conditions pose major challenges for signal detection, in particular when instant signal detection is demanded. For instance, detected illumination changes can be caused by object motion. This applies in particular when object tracking is subjected to restrictions, such as time delay, or even when illumination is only consistent in a very small area. Furthermore, illumination conditions can deteriorate due to unsteady illumination sources, e.g., varying ambient light.

A considerable portion of illumination-related disturbances can be explained by specular reflection. Specular reflectance is the "perfect" reflection of incident radiation at an interface. Basically, an incident ray corresponds to a reflected ray. An angle of reflection equals an angle of incidence. In other words, specular reflection implies mirror-like reflection at surfaces and interfaces. Furthermore, the reflected ray is highly indicative of the source of electromagnetic radiation, namely the illumination source. This relationship has been utilized.

It shall be understood that mainly diffuse reflection provides the desired vital signals. Diffuse reflection substantially comprises body reflection rather than interface reflection. For instance, body reflection is influenced by slight changes of the color of an area of interest of the body. Color changes can be caused, inter alia, by vascular pulsation due to blood circulation. The desired vital signals can be derived therefrom. Furthermore, incident radiation can be absorbed to some extent. However, the detectable reflected signals most probably comprise a disturbing specular reflection portion. Specular reflection basically "mirrors" incident radiation without being influenced by object properties present under the interface, e.g., the top surface of the skin. Especially perspiring skin areas and oily or greasy skin areas are highly susceptible to specular reflections. Under certain circumstances, e.g., sports practice, workouts, physically demanding work, or even due to illness, a huge portion of electromagnetic radiation reflected by the object can be related to specular reflectance. Thus, the characteristic signal is supposed to have a poor, i.e. reasonably small, signal-to- noise ratio.

Further adverse effects on the signal-to-noise ratio can occur when an object of interest has a dark skin tone. Basically, dark colors absorb a larger part of incident radiation than bright colors. Therefore, objects having a light skin tone absorb less radiation. As only reflected radiation can carry the desired signals, the signal-to-noise ratio is even further worsened for dark skin.

By way of example, the data stream can be captured by means of a charge- coupled device (CCD) sensor. Usually, a point of interest, e.g. captured by a single CCD pixel, or a pixel array, covers radiation portions attributable to non-indicative specular reflection and indicative diffuse reflection. Furthermore, when summarizing captured radiation of a pattern of pixels, most likely a combination of (diffuse) scattering reflection plus perfect (specular) reflection can be contained in the input data. Furthermore, object motion generally changes the average specular reflection fraction of an area of interest.

The refinement is further based on the insight that, when applying an additive signal space, the characteristic signal is basically composed of components related to distinct channels, or, so to say, axes. The additive signal space can be linked to a derivative signal space or signal model. Basically the derivative signal space utilizes a different approach for composing the characteristic signal. Amongst other possible components, the derivative signal model, or, signal representation, relies on difference components rather than absolute components. Advantageously, the difference components enable a signal representation wherein specular reflection can be suppressed, at least to a certain extent.

In other words, when transferring the characteristic signal to the derivative signal model, en passant, considerable parts of the disturbing signal portion can be eliminated from the characteristic signal. The characteristic signal can be at least partially compensated for object motion and/or a non-indicative body reflection. The signal-to-noise ratio can be improved in this way. Thus, downstream signal analysis can be simplified, even under considerably challenging conditions. Furthermore, a required data volume can be reduced as fewer "channels" are required for carrying the desired vital signals.

F r instance, the transformation can correspond to the following scheme

wherein (A_l A₂ )^t represents the difference components, wherein (R G B f represents the absolute components, wherein a_x , a₂, a₃ and b_x , b₂, b₃ represent the coefficients, and wherein a_x + a₂ + a₃ = 0 and b_x + b₂ + b₃ = 0 . The vector (R G B)^T can be represented in an additive signal space.

By way of example, in a preferred embodiment the coefficients may have the following values: a_x = \ , a₂ = -\ , a₃ = 0 and b_x = 1 , b₂ = 1 , b₃ = -2 .

In another preferred embodiment the coefficients may take the following values: a_x = 0,5 , a₂ = -0.5 , a₃ = 0 and b_x = 0.25 , b₂ = 0.25 , b₃ = -0.5 .

These embodiments enable a considerably improved signal-to-noise ratio facilitating further signal analyses. In this connection, an optional generalized supplemental expression for the terms A_x and A₂ could read as follows: Δ^* = cos(9 )A₁ + sin((p )A₂ and Δ^* ₂ = 8ΐη(φ)Δ₁ + cos((p )A₂ , wherein 0 < φ < 2π . Hence, adequate difference components can be chosen from a possible set of derived difference component terms Δ^* and Δ^* ₂ still meeting the requirements outlined above. Furthermore, for some embodiments it can be preferred that both difference component terms A_l and Δ₂ (or, Δ^* and Δ^* ₂ ) determined through the arithmetic transformation eventually have like signs.

The foregoing scheme can be further expanded to

in case it is preferred to maintain luminance information. The coefficients l_x,l₂, l₃ can fulfill the requirement l_x +l₂ + l₃ = 1 . For instance, the coefficients can have the following values: l_x = 0.33 , /₂ = 0.33 , /₃ = 0.33 . However, luminance information is no longer necessarily required for extracting the desired vital signals.

Needless to say, for some applications the symbol " = " can be readily replaced by " ~ " without departing from the scope of the present disclosure.

For instance, established PPG approaches basically can make use of a ratio of two distinct (absolute) signal components, e.g. the ratio between red and infrared signals, or the ratio between red and green signals. For further consideration, the ratio can be plotted over time. Slight periodic changes of the ratio can allow an estimation of the desired signals. Assuming that the applied absolute signals are simultaneously influenced (e.g., identically influenced by white illuminant) by varying illumination conditions (e.g., varying specular reflection), it is suggested to base the signal detection on a ratio of difference signals to be derived from the absolute signals. This aspect comes from the insight that non- indicative specular reflection is basically similarly present in the absolute signals, e.g. a red signal, a green signal and a blue signal. When at least two of these signals are compared, e.g., when a difference signal is derived therefrom, it can be assumed that specular reflection is at least substantially suppressed in the difference signal.

In other words, in terms of vector representation, the absolute signals, namely the at least three absolute components, can be considered as components of the vector representing the characteristic signal. Each of the at least two difference components can be obtained by applying the named transformation to at least two of the at least three absolute components. Needless to say, each of the at least three absolute components should be taken into account for the determination of at least one of the at least two difference components. Hence, when looked at together, the difference components can still represent each of the original absolute components, at least to a certain extent. The characteristic signal vector having at least three components can be replaced for further signal detection measures by a difference vector comprising a smaller number of components, e.g., decreased by one when compared with the components of the characteristic signal vector. The "axed" component represents the specular reflectance portion, at least to a certain extent.

According to yet another aspect, the processing means further comprises an extractor means for extracting the vital signal from the transferred optimized sub band portions under consideration of an additive or subtractive expression or a ratio of the at least two difference components.

In this connection, it shall be understood that the vital signal potentially is not present in some of the sub band portions. This applies in particular when the desired signal, e.g. the heart rate or the respiration rate, does not change much over time during data acquisition. However, clearly working out that the desired signal is not present in a respective sub band can be also considered a desirable result. Consequently, the remaining sub bands become even more indicative.

This embodiment can be further developed in that the extractor means is adapted to normalize the extracted vital signal under consideration of a deviation value thereof, preferably a standard deviation, over a moving window applied to a sequence of the transferred optimized sub band portions. Hence, the desired signals can be further enhanced by removing a statistical dispersion indicative of disturbing overall deviation. In particular, the amplitude of the signal of interest can be further "stabilized" in this way.

According to another aspect, the signal space is indicative of luminance information and chrominance information, the chrominance information being representable by the at least two difference components, wherein the luminance information is

representable by a luminance signal being substantially aligned with a luminance index element in the signal space, the luminance index element being substantially indicative of a selected source of electromagnetic radiation, wherein the at least two difference components are substantially orthogonal to the luminance index element, preferably the at least two difference components are substantially orthogonal to each other.

For detecting the desired signal(s) of interest, it is preferred that mainly the chrominance information is utilized. In this way, specular reflections substantially influencing luminance information can be "ignored". In other words, the use of chrominance information represented by the (color) difference signals can render the transferred signal substantially independent of the mainly disturbing luminance signal. It should be understood that preferably linear signals are utilized. Non- linear signals, e.g., gamma corrected signals, can be (re)transformed accordingly.

Preferably the luminance index element represents an expected or measured light source characteristic, e.g., a light source color or a color temperature of a radiation source.

The source of electromagnetic radiation can be embodied by artificial light sources, sun light, radiation sources emitting radiation having non- visible components, or combinations thereof. The radiation can be guided directly to the object of interest. Also indirect radiation, e.g. ambient light, is applicable.

For most applications it can be fairly assumed that the radiation source, namely the light source, emits basically plain white light. Hence, assuming an additive signal space composed of three color channels, the luminance index element can be represented by a diagonal vector traversing the signal space. For instance, a black point indicating the smallest luminance value can be embodied by a zero point of the signal space (0, 0, 0). The black point can coincide with the common initial point of the axes representing the additive components, e.g., red, green, and blue. Diagonally opposite of the black point a white point can be disposed in the signal space. The white point can denote the point of the largest luminance value. In case the signal space is a "unitary" signal space, the white point can be embodied by the point (1, 1, 1). The white point can further denote the end point of the luminance index element. So, given these assumptions, the luminance index element can be embodied by the vector (l 1 l)^r in the signal space.

The geometric attribution applies in particular when the respective components are handled in terms of vector representation. A diagonal plane traversing the signal space and being substantially orthogonal to the luminance index element can be considered a chrominance plane. The chrominance plane represents a "slice" of the signal space being substantially independent of luminance information.

According to another embodiment, it is further preferred that the processing means comprises a weighting means for weighting the at least two difference components so as to derive weighted optimized sub band portions from the transferred optimized sub band portions under consideration of at least two weighted difference components, preferably the weighting is directed to minimize a spread of the weighted optimized sub band portions.

The weighting means can contribute to further improvement of the signal detection. The weighting means can be comprised between the converter means and the extractor means. Also the weighting means can be embodied by the common processing unit.

The spread can be also referred to as statistical dispersion, statistical variability or variation. For instance, the spread can be represented by variance or standard deviation values.

As for the difference component approach, it is further preferred that the signal of interest is derived under consideration of a weighted sum (or difference) of the at least two difference signals. The weighting means can allow for instant determination of a weighting factor.

According to a further aspect of this embodiment, the weighting comprises a determination of a deviation value, preferably a standard deviation, of each of the at least two difference components, wherein the deviation value of each of the at least two difference components, is determined under consideration of temporal variations thereof over a moving window applied to a sequence of each of the at least two difference components.

As mentioned above, photoplethysmography can make use of a ratio of two distinct (absolute) signal components, e.g. the ratio between red and green signals. For instance, a normalization can be applied wherein the red and green signals are divided by their respective (temporal) mean values. This approach applies in particular when the absolute components are unlikely to become zero. Under usual conditions neither a red (or mean red) signal nor a green (or, mean green) signal will become zero so that division by zero is unlikely to happen. Basically, such an approach can be applied to difference components as well.

However, under certain circumstances a ratio of difference components (as well as a ratio of absolute components) can lead to an erroneous division-by-zero term. This applies in particular when selected sub band portions are processed. For instance, a removal of non-indicative spectral portions may lead to processed difference components no longer exhibiting "steady" portions. Therefore, a mean value thereof potentially can become zero. Thus, common normalization under consideration of mean values can face a division-by-zero issue. Therefore, a simple consideration of the ratio between two difference components can face further challenges. A possible approach to this issue can be a transformation of the ratio (quotient) term for each sub band. For instance, logarithmic identities can be considered allowing an alternative representation of the logarithm of the quotient, namely a difference between the logarithm of the numerator of the quotient and the logarithm of the denominator of the quotient. Further considering the Taylor extension of the logarithms and assuming expected values of the logarithmic terms, an (inverse) additive combination of the two difference components still enables a detection of the desired signals. In this context, it is referred to respective equations stated below in connection with detailed description of an exemplary embodiment. It is reminded that the desired signals, e.g., the heart rate, basically can be extracted by analyzing slight temporal variations of the characteristic signals rather than absolute values thereof.

However, according to an alternative approach, the desired vital signals indeed can be extracted under consideration of a ratio between the at least two difference components. In this connection, when normalizing, facing the division-by-zero issue mentioned above, it is suggested to apply a normalization of the difference components relying on a division by their respective standard deviation rather than their respective mean value. This approach is based on the insight that amplitudes of the utilized difference components are at least partially proportional to their mean values.

Applying these approaches the respective sub bands can significantly enhance signal quality.

According to another aspect of the device, the at least one at least partially periodic vital signal is selected from the group consisting of heart rate, heart beat, respiration rate, heart rate variability, Traube-Hering-Mayer waves, and oxygen saturation.

In a further aspect of the present invention a method for extracting information from remotely detected characteristic signals is presented, comprising the steps: receiving a data stream derivable from electromagnetic radiation reflected by a remote object, the data stream comprising a continuous or discrete time-based characteristic signal including physiological information and a disturbing signal portion, the physiological information being representative of at least one at least partially periodic vital signal, the disturbing signal portion being representative of at least one of an object motion portion and/or a non-indicative reflection portion, the characteristic signal being associated with a signal space, the signal space comprising complementary channels for representing the characteristic signal, components of the characteristic signal being related to respective complementary channels of the signal space,

- pre-processing the data stream by splitting a relevant frequency band thereof into at least two defined sub bands comprising determined portions of the characteristic signal, each of which representing a defined temporal frequency portion potentially being of interest,

optimizing the sub bands so as to derive respective optimized sub bands from the at least two sub bands, the optimized sub bands being at least partially indicative of a presence of the vital signal,

combining the optimized sub bands so as to compose an optimized processed signal.

Advantageously, the method can be carried out utilizing the device for extracting information of the invention.

According to an even further aspect of the invention a computer program is presented, the computer program comprising program code means for causing a computer to carry out the steps of the method for extracting information of the invention when said computer program is carried out on a computer.

Preferred embodiments of the invention are defined in the dependent claims. It shall be understood that the claimed method has similar and/or identical preferred

embodiments as the claimed device and as defined in the dependent claims.

BRIEF DESCRIPTION OF THE DRAWINGS

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. In the following drawings

Fig. 1 shows a schematic illustration of a general layout of a device in which the present invention can be used, Fig. 2 shows a schematic illustration of a reflectance model utilizing a body reflection and interface reflection approach,

Fig. 3a shows an exemplary schematic illustration of a signal space comprising an index element representing a characteristic signal,

Fig. 3b shows a further exemplary simplified schematic illustration of a signal space by way of explanation,

Fig. 4 shows a schematic illustration of a system comprising signal decomposition and (sub) signal processing,

Fig 5a, 5b and 5c show frequency response diagrams indicating exemplary filter characteristics,

Fig. 6 shows a schematic illustration of a system comprising signal decomposition,

Fig. 7b and 7b show an object of interest being monitored when performing some fitness exercise,

Fig. 8 depicts three diagrams, each showing a spectrogram of physiological information obtained from an object of interest in a first exemplary case of application,

Fig. 9 depicts two diagrams, each showing a spectrogram of physiological information obtained from an object of interest in a second exemplary case of application, and

Fig. 10 shows an illustrative block diagram representing several steps of an embodiment of a method according to the invention.

DETAILED DESCRIPTION OF THE INVENTION

Fig. 1 shows a schematic illustration of device for extracting information which is denoted by a reference numeral 10. For instance, the device 10 can be utilized for recording image frames representing a remote object 12 for remote PPG monitoring. The image frames can be derived from electromagnetic radiation 14 reflected by the object 12. The object 12 can be a human being or animal, or, in general, a living being. Furthermore, the object 12 can be part of a human being highly indicative of a desired signal, e.g., a face portion, or, in general, a skin portion.

A source of radiation, such as sunlight 16a or an artificial radiation source 16b, also a combination of several radiation sources can affect the object 12. The radiation source 16a, 16b basically emits incident radiation 18a, 18b striking the object 12. For extracting information from the recorded data, e.g. a sequence of image frames, a defined part or portion of the object 12 can be detected by a sensor means 24. The sensor means 24 can be embodied, by way of example, by a camera adapted to capture information belonging to at least a spectral component of the electromagnetic radiation 14. It goes without saying that the device 10 also can be adapted to process input signals, namely an input data stream, already recorded in advance and, in the meantime, stored or buffered. As indicated above, the electromagnetic radiation 14 can contain a continuous or discrete characteristic signal which can be highly indicative of at least one at least partially periodic vital signal 20. The characteristic signal can be embodied by an (input) data stream 26. For data capturing, a potentially highly indicative portion of the object 12 can be masked with a pixel pattern 22. When agglomerating respective single pixel values of the additive components, a mean pixel value can be derived from the pixel pattern 22. In this way, the detected signals can be normalized and compensated for undesired object motion to some extent. The mean pixel value can be represented by a characteristic signal. In the following, the captured data stream 26 can be considered a representation of a certain area of interest of the object 12 which may cover a single pixel or, preferably, an agglomerated pixel area covering a plurality of pixels.

In Fig. 1 the vital signal 20 may allow several conclusions concerning heart rate, heart beat, heart rate variability, respiratory rate, or even oxygen saturation.

Known methods for obtaining such vital signals may comprise tactile heart rate monitoring, electrocardiography or pulse oximetry, for instance. To this end, however, obtrusive monitoring is required. As indicated above, an alternate approach is directed to unobtrusive remote measuring utilizing image processing methods.

The data stream 26 comprising the continuous or discrete characteristic signal can be delivered from the sensor means 24 to an interface 28. Needless to say, also a buffer means could be interposed between the sensor means 24 and the interface 28. Downstream of the interface 28 a decomposition means 32 is provided to which a data stream 30 can be delivered. The decomposition means 32 is adapted to split the data stream 26, 30 into several sub bands 34a, 34b, 34c. The sub bands 34a, 34b, 34c can represent (pre)defined frequency portions of the data stream 26, 30. The decomposition means 32 may comprise and/or utilize several filters, e.g. band pass filters, high pass filters and low pass filters.

The sub bands 34a, 34b, 34c of the data stream 26, 30 can be delivered to a processing means 36 and separately processed so as to obtain optimized sub bands 38a, 38b, 38c. Basically, greater prominence can be given to the desired signals in the optimized sub bands 38a, 38b, 38c. For instance, the processing of the sub bands 34a, 34b, 34c can comprise an arithmetic transformation resulting in difference signals rather than absolute signals. Difference signals are less indicative of disturbing components of the characteristic signals in the data stream 26.

Further, a composition means 40 can follow which is adapted for combining the optimized sub bands so as to compose an optimized processed signal 42. The optimized processed signal 42 can be delivered to an analyzing means which 44 can be utilized for further signal enhancement and/or detection measures. In this connection, the analyzing means 44 can be applied for isolating and enhancing the desired signal component even more indicative of the vital signal 20 of interest from the optimized processed signal 42 delivered thereto. For instance, the analyzing means 44 can be adapted for further processing the optimized processed signal 42, e.g., detection of a dominant signal peak, such as a heart rate indicative frequency peak. Eventually, an output signal 46 can be obtained.

The decomposition means 32, the processing means 36, the composition means 40, and the analyzing means 44 can be jointly embodied by a common processing unit 48, e.g. a central processing unit having a single processor or multiple processors. Also the interface 28 can be connected thereto in a common processing unit 48 housing the respective subcomponents. By way of example, the processing unit 44 can be embodied by a personal computer driven by respective logic commands. In case the sensor means 24 is also jointly connected to the interface 28 by means of hardware, a capturing unit arranged at a higher level may house the respective subcomponents.

However, in the alternative, it can be envisaged to combine a separate sensor means 24 with the processing unit 48. This connection can be established by means of cable links or by means of wireless links. In place of the sensor means 24 also a storage means comprising prerecorded data could be connected to the processing unit 48.

The following section describes an exemplary approach to remote photoplethysmography utilizing several aspects of the device and method of the invention. It should be understood that single steps and features of the shown approach can be extracted from the context of the approach. These steps and features can be therefore part of separate embodiments still covered by the scope of the invention.

As outline above, unobtrusive vital signal monitoring using a video camera, also referred to as remote photoplethysmography, has been demonstrated. Basically, algorithms can register the average skin-tone of an object, e.g., a person, which varies with the blood volume and blood oxygenation.

In a classical (also non-remote) photoplethysmography approach, the heartbeat can be detected in a ratio of red and green color components, for instance G(i)

Advantageously, the ratio can be normalized. Also a (normalized) ratio of red and infrared spectral components can be utilized. Basic photoplethysmography devices may comprise obtrusive attachments to be applied to a fingertip or an earlobe of an object to be observed. Hence, these approaches may imply an uncomfortable feeling when applied.

The named normalization can be directed to time based normalization. For instance, the red color components can be normalized by calculating:

k≤i≤n

and similarly for green, where n-k can be chosen such that at least a number of heartbeats is covered.

The normalization can be directed to make the heartbeat amplitude independent of the strength and color of the illuminant. The heartbeat signal itself results in:

i_{)= =}mm- i (2)

G, (i) G(i) R(i)

The last term (-1) indicates that for some applications mainly so-called pulsating portions of the signal are of interest while the direct component (or, constant offset value) can be deducted after normalization. In this way, illumination independent results can be achieved provided the observed red and green signals are the result of light passing through the skin. These spectral components are highly indicative of the desired signals. In a non-remote photoplethysmographic approach monitoring conditions are steady. Ambient light and distortions due to further illumination variations basically can be neglected.

Usually, non-remote photoplethysmographic devices comprise standard lights emitting radiation guided directly to a portion of interest of the object to be monitored. As the devices can be closely attached to respective skin areas, disturbing luminance variation caused by remote lights can be avoided.

Under these "laboratory" conditions the ratio of red and green is mainly determined by the color of the skin, which slightly fluctuates with the heartbeat, but can be considered constant for the mean ratio of red and green, as long as the spectrum of the (device inherent) illuminant is stable.

For remote PPG applications, large disturbance portions have to be expected. Steady or constant disturbances basically can be handled by the normalization of equation (2), at least partially. However, varying disturbances, e.g. due to object motion or varying illumination conditions, can even adversely influence a (time) normalized signal. The frequency domain of the signal components can be expected to be corrupted. Again referring to equation (2), it is reminded that a ratio of the time-dependent red and green values can be even further corrupted due to multiplicative noise effects introduced when instant and normalized values are multiplied and divided. In this way, even disturbances occurring in non-indicative frequency bands can potentially create disturbances in frequency bands considered to be highly indicative.

Thus, it can be appreciated to decompose the respective signal components (e.g., red, green and blue) prior to normalization and further signal processing measures.

Basically, signal decomposing can comprise band pass filtering, low pass filtering, high pass filtering and further filtering characteristics. It can be further envisaged to split a relevant frequency band so as to obtain at least two sub bands which can be further processed. The more accurate the at least two sub bands are defined, the less the respective normalization result is influenced by "frequency distorted" disturbances.

Basically, an object to be monitored can cause several indicative and non- indicative pulsations which can be present in the detected characteristic signals. For instance, heart rate, respiration rate as well as periodical movements can show frequencies all of which can be located in a similar frequency band. Especially for sports, fitness and leisure applications, object motion, e.g. periodic deflections of arms or legs, can cause a dominant frequency overlaying the desired vital signals.

In this connection, sub band processing can contribute to detecting and distinguishing relevant signals. It is understood that sub band processing is applicable for various data processing and/or optimization measures directed to detect and/or emphasize the desired signals.

In practice, when applying remote photoplethysmo graphic approaches, e.g. a camera based PPG system, further adverse effects can influence the signal-to-noise ratio. For instance, the light reflected from the skin basically comprises two components that can be described by the so-called dichromatic reflection model. In this connection, reference is made to Fig. 2 illustrating reflection of incident radiation 58 at an interface 50 between two media 51, 52. Reference numeral 51 denotes air through which incident radiation 58 is transmitted. Reference numeral 52 denotes a skin tissue to which incident radiation 58 is directed. The interface 50 is interposed between the air 51 and the skin tissue 52. The interface 50 can be considered as the top surface of the skin. The skin tissue 52 may comprise colorant 54 which slightly fluctuates with the signal of interest, e.g., the heart rate. The interface or top surface 50 may comprise a macroscopic surface normal 56 and microscopic surface normals 62, the latter attributable to microscopic surface unevenness. Hence, even incident radiation 58 subjected to (perfect) specular reflection at the interface 50 can be reflected at an reflection angle corresponding to the microscopic surface normal 62 rather than the macroscopic surface normal 56. The reflected radiation is denoted by reference numeral 64. A reflected radiation to be expected with knowledge of the macroscopic surface normal 56 is denoted by reference numeral 60. However, for the following elucidation the microscopic surface normal 62 can be equated with the macroscopic surface normal 56.

Furthermore, a considerable component of the incident radiation 58 is reflected by skin tissue colorant 54 rather than the interface 50. The reflection may comprise multiple reflections as indicated by reference numerals 66, 66', 66". As skin tissue colorant 54 is distributed inhomogeneously in the skin tissue and respective colors may vary over time, the so-called body reflection can be considered substantially diffuse reflection. Reflected radiation due to body reflection is denoted by reference numeral 68. Thus, beside of the specular reflection component 64 also a diffuse scattered reflection component 68 can be reflected by the object of interest.

Hence, a part of incident light or radiance is reflected by a diffuse reflection component, namely the body reflection component 68, which has traveled through the skin and represents skin colors including variations thereof due to the desired vital signals, e.g., heart rate. This reflection component is highly indicative of the signals of interest.

On the contrary, the specular reflection component 64 directly reflected at the top surface 50 of the skin is mainly indicative of the color of the illuminant and does not comprise considerable signals of interest.

Therefore, two fractions of radiance reflected by the object of interest may occur. In combination these fractions form the observed characteristic signals, e.g., the observed color. Illumination conditions may vary over time, e.g., due to object motion.

Consequently, also the characteristic signals may vary widely over time.

Two essential issues arose when specular reflection is to be taken into account for the desired signal detection. First, the (time based) normalization provided in equation (1) and (2) is no longer applicable and may vary over time since it also contains the (generally motion dependent specular) reflection component. The second issue is related to the fact that the amplitude of the signal of interest (e.g., HB, i.e. the heart beat) is no longer basically constant as it is proportional only to the fraction of the radiation that is diffusely reflected, namely the body reflection component 68, while the (time based) normalization also contains the specular reflection component 64. Thus, normalized signals can become distorted.

Therefore, remote camera based PPG systems are highly sensitive to motion and/or changing luminance conditions. In the following, an exemplary approach to significantly reduce the effect of specular reflections is outlined. Advantageously, sub band processing is included.

The approach is based on the insight that color difference signals, namely difference components, rather than color signals, namely absolute components, as disclosed in prior art methods, can be drawn for the detection of the vital signals. Hence, the adverse effect of specularly reflected radiation can be eliminated, at least to a certain extent.

Consequently, subsequent signal detection profits from a significantly improved signal-to- noise ratio.

Furthermore, the approach requires less information as luminance information indicative of the strength of the radiation source 16 can be neglected. However, as outlined above, luminance information can be kept for further processing but, at the same time, compressed with a significantly small bit rate without adverse effects on the vital signal detection.

An exemplary numerical description is elucidated in the following. Equation (2) can be rewritt n in the following form:

The logarithmic expression can be approximated by a Taylor expansion:

4

(x - 1)² , (x - 1)³ (x - 1)

log(x) = (x - 1) - + .... (4)

2 3 4

Hence, assuming that the arguments of the logarithmic terms in Eq. (3) are very close to one, equation (3) finally can be approximated by:

HB(i) ~ R_tt (i) - G„(i) (5)

Consequently, the desired signal of interest, e.g., the heart rate (or, heart beat), can be extracted from a small signal resulting from the difference, or, so to say the

"approximated ratio", of two signals. Both signals R_n and G_n may comprise a large deviation or variance. Therefore, (time based) normalization has to be addressed to with a high level of attention. In this context, it can be further envisaged to apply an alternative common normalization to the terms of equation (5):

HB(i) ^{» m ~ a ' Gii)}

norm

wherein the denominator norm could be derived, for instance, under consideration of an average difference of R and G. As difference signals are applied in this way, specular reflection is can be greatly reduced in the norm term:

norm = R(i) - G(i)

The coefficient a can be chosen so as to minimize the energy in the resulting signal spectrum, e.g. heart beat spectrum. Generally, the value of a can depend on the skin color. In this connection, it is reminded that the detected signals are expected to contain large noise portion while the vital signal of interest is expected to be carried by slightest periodical changes of the signals.

Basically, band splitting and optimization per sub band improves the above- mentioned signal processing measures.

With specular reflections the normalization of equation (1) probably contains errors as the skin color differs from the color of the illuminant, the latter representable by the specular reflection component 64. Skin color values can be "distorted", i.e., spatially and/or over time, by the color of the illuminant.

Fig. 3a depicts an exemplary signal space 72, e.g. an RGB color space. The signal space 72 comprises additive channels 74a, 74b, 74c indicative of spectral information, e.g., red, green and blue color channels. According to the reflection model outlined above, a detected characteristic signal 80 can be composed of the specular reflection component 64 and the body reflection component 68 of Fig. 2. The specular reflection component can be considered equivalent to a luminance signal indicated by an arrow 84. The body reflection component can be considered equivalent to a physiological information signal indicated by an arrow 86. The specular reflection component 64 and the body reflection component 68 span a reflection plane (not shown) in which also the detected characteristic signal 80 can be located. By way of example, the signal space 72 can be considered a "unitary" signal space, wherein components along the additive channels 74a, 74b, 74c can take values between zero and one. Further value ranges departing from the zero and one range can be envisaged and treated accordingly. The signal space 72 further comprises a chrominance plane 76 and a luminance index element 78. The luminance index element 78 is indicative of a source of incident radiation, e.g. a light source. The luminance index element 78 can be considered a diagonal vector traversing the signal space 72. This applies in particular when the radiation source 16 basically emits plain white light. Preferably the "color" of the radiation source 16 equals the white point (e.g. 1, 1 ,1) of the signal space 72. In case the specular reflection is only partially influencing a detected area of interest, e.g., the pixel pattern 22 of Fig. 1 , the luminance signal 84 is "shorter" than the luminance index element 78. Both vectors, the luminance signal 84 and the index element 78 are parallel and point in the same direction. The luminance signal 84 can be considered an expression of how much the detected area of interest, e.g., the pixel pattern 22, is influenced by specular reflection. For illustrative purposes, also a linear combinations, namely the composed characteristic signal 80, of the vector components 84, 86 is presented next to the respective signals space 72. In this context, it is reminded that Fig. 3a represents a three-dimensional (3D) representation. Consequently, also the added linear combinations represent 3D vectors rather than two-dimensional (2D) vectors.

The luminance index element 78 and the luminance signal 84 are basically perpendicular to the chrominance plane 76. The chrominance plane 76 is a diagonal plane in the signal space 72. For instance, the chrominance plane 76 can be described by the expression^ + G + B = 1 , wherein 0 < R < 1 , 0 < G < 1 and 0 < B≤ 1 . When aiming at an elimination of luminance information from the characteristic signal 80, graphically a projection to the chrominance plane 76 can be sought-for, refer also Fig. 3b.

The characteristic signal 80 is composed of absolute components 82a, 82b, 82c each of which related to a respective additive channel 74a, 74b, 74c. The absolute

components 82a, 82b, 82c can represent respective red, green and blue values.

For instance, the characteristic signal 80 can be composed according to the following expression:

wherein (R_CIJ G_CH B_CH )^T can correspond to an RGB value of a detected color pixel along the additive channels 74a, 74b, 74c, wherein (R_B G_B B_B )^T and (R_S G_S B_S )^T may denote directions of the body reflection component 68 and the specular reflection component 64, and wherein m_b(i) and m_s (i) can indicate magnitudes of the respective reflection components 64, 68. The term m_b (i)^■ (R_B G_B B_B )^T can be considered highly indicative of the desired signal. The term m_s (i)^■ (R_S G_S B_S )^T can be considered highly indicative of distortion due to specular reflection.

Besides that, a considerable part of incident radiation 58 can be absorbed by the object's skin tissue. In particular, dark skin color absorbs considerable parts of incident radiation.

In the following, the signal space 72 and its components can be considered a representation of a certain area of interest of the object 12 which may cover a single pixel or, preferably, an agglomerated pixel area covering a plurality of pixels, refer Fig. 1.

It would be advantageous to decompose the characteristic signals 80 so as to arrive at the desired physiological information signals 86. Substantially, the orientation and length of the desired physiological information signals 86 is unknown. While the orientation of the luminance signals 84 is basically known, the length of the luminance signals 84 is also unknown.

Therefore, a suitable approach relies on color difference signals instead of color signals. Since the specular reflection component is substantially identical in all color signals, e.g., roughly white illuminant, it can be considered absent in the difference of two color signals. It is reminded that the color signals may be represented by the respective values of the absolute components 82a, 82b, 82c, e.g., (R_CIJ G_CH B_CH )^T , of the characteristic signal

80.

Fig. 3b is referred to, exemplifying a difference component approach which can be advantageously combined with sub band based signal processing. Merely for illustration purposes Fig. 3b shows a two-dimensional (2D) signal space 72'. In others words, the signal space 72' can be considered a "slice" of the signal space 72. Therefore, the chrominance plane 76 is represented as a diagonal line being perpendicular to the luminance index element 78. Reference numerals 74a, 74b indicate two of the at least three additive channels, e.g., red and green out of the RGB signal space. In Fig. 3b characteristic signals 80', 80" are represented each of which comprising two distinct components, namely the luminance signals 84', 84" and the physiological information signal 86. The characteristic index elements 80', 80" are distorted due to varying luminance signals 84', 84". The luminance signals 84', 84" are parallel to the luminance index element 78. Reference numerals 82a, 82b indicate exemplary absolute components (e.g. RGB components) of the characteristic signal 80'. Next to the signal space 72' of Fig. 3b a simple arithmetic transformation or projection is illustrated. Projecting the characteristic signal 80 to the chrominance plane 76 can result in a difference component 88. For each of the characteristic signal 80', 80" and the genuine physiological information signal 86, the transformation results in the same difference component 88. Therefore, the potentially varying luminance signals 84', 84" indicative of disturbing specular reflection can be removed while the resulting difference component 88 is still indicative of the desired physiological information signal 86 which comprises the vital signal of interest. Accordingly, at least a second difference component 88 can be derived from the absolute components 82a, 82b, 82c. An expression (ratio or difference) of the difference components 88 can exhibit an enhanced vital signal of interest. Consequently, varying illumination conditions have no adverse effects on subsequent signal extraction measures.

It is understood that the amplitude of a single color difference signal can still be proportional to the strength of the illuminant. Therefore, at least two color difference signals 88, e.g., A_l and Δ₂ , are required for eliminating variation in the strength of the illumination, e.g., caused by object motion. Consequently, they have to be derived from at least three color signals. Therefore, an additive RGB space can be considered a proper choice since the characteristic signal 80 is composed of three absolute color components 82a, 82b, 82c. Preferred transformations and coefficients are outlined above.

When sub band optimization is applied, the normalization of the color difference signals A_l and Δ₂ , analog to equation (1), potentially cannot be carried out. Basically, avoiding out of band disturbances, the mean values of sub band signal components can be leveled to some extent. Furthermore, it can be assumed that the mean signals no longer exhibit the slight (at least partially periodic) changes of the original signals indicating the desired vital information. In other words, the signal means potentially can become zero. Hence, division by the temporal mean values can pose a division by zero issue. Therefore, an estimation of the ratio of the at least two difference components can result in computing problems. For this reason, the derivation provided in equations (3), (4) and (5) can be applied. Consequently, the mere ratio of the difference components can be replaced by a difference, e.g., HB(i) ~ A_l (i) - A₂ (i) .

A further refinement may comprise a minimization of the variance of a weighted sum of the two difference components. Hence, (time-based) normalization can be improved. This approach can compromise applying a weighting function to the at least two difference components:

HB(i) ~ (i) - w(i)A₂ (i) , (6) wherein the weight can be selected in order to minimize the variance of the vital signal of interest, e.g., the heart rate. Various approaches can be envisaged. A fairly simple method determines w(i) in such a way that the standard deviation of the two terms in equation (6) are basically equal:

Mi) (7)

std(A₂ )

In this way, overall disturbances can be removed from the desired signal to a certain extent. For instance, the standard deviation can be calculated in a temporal window around i. By way of example, the window can be chosen in the order of about one second. Hence, the number of frames to be covered by the moving window can be derived therefrom.

Furthermore, the resulting signal of interest, e.g. the heart rate, can be further normalized by applying its standard deviation. Advantageously, the standard deviation can be calculated utilizing the same window size interval as chosen for the weighting function.

Fig. 4 is referred to, showing a schematic block diagram of a device 10 for extracting information from detected characteristic signals. The input data stream is represented by a broad block arrow 26. The input data stream 26 comprises a considerable large frequency band. The input data stream 26 is delivered to the decomposition means 32. The decomposition means 32 comprises several filters 90, 94. The filter 90 can be considered a band pass filter. A band pass filtered signal is indicated by a block arrow 92 being thinner than the block arrow 26. Thus, several non- indicative signal portions have been blocked by the band pass filter 90. The filter 94 can be considered a band split filter. The band split filter 94 is adapted to split the band pass filtered signal 92 into at least two sub bands 34a, 34b, 34b. It should be understood that the filters 90, 94 are adapted to predetermined signal filtering. For instance, the band split filter 94 can comprise predefined filter characteristics 96a, 96b. Exemplarily, the filter characteristic 96a describes a high pass filter while the filter characteristic 96b describes a low pass filter. Furthermore, also band pass characteristics 96c, 96d can be utilized to split the band pass filtered signal 92 into the at least two sub bands 34a, 34b, 34b.

The at least two sub bands 34a, 34b, 34b represent respective frequency portions of the input data stream 26. The sub bands 34a, 34b, 34b can partially overlap each other, refer also Fig. 5c. The sub bands 34a, 34b, 34b define relevant frequency portions to which further signal processing measures can be applied individually.

Downstream the filter means 32 the processing means 36 is arranged for separately processing the sub bands 34a, 34b, 34b delivered thereto. The processing means 36 provides a converter means 98, a weighting means 100 and an extractor means 102 for this purpose. The converter means 98, the weighting means 100 and the extractor means 102 are adapted for signal optimization processing.

Hence, the optimized sub bands 38a, 38b, 38c can be obtained and delivered to the composition means 40 which is adapted to combine them so as to create the optimized processed signal 42.

Reference is made to Figs. 5a, 5b and 5c showing frequency response diagrams depicting exemplary filtering characteristics. Fig. 5a illustrates a first frequency response diagram having two axes 106, 108. Axis 106 represents frequency values while axis 108 represents amplitude values of plotted graphs. The amplitude axis 108 can be normalized while the frequency axis 106 plots frequency values in Hertz (Hz). A curly brace indicates a band pass filtered signal 92a which can be obtained by applying a band pass characteristic 110a. Furthermore, filter characteristics 112a', 112a" can be applied so as to define the at least two sub bands to be processed. Considering the window left by the band pass characteristic 110a, the filter characteristic 112a' can be considered a low pass filter while the filter characteristic 112a" can be considered a high pass filter.

Fig. 5b and Fig. 5c illustrate similar frequency response diagrams, each of which showing a band pass filter characteristic 110b, 110c potentially passing a relevant band pass frequency band 92b, 92c. Moreover, respective sub bands can be obtained by applying further filter characteristics 112b^*, 112b", 112b^*", 112b"" and 112c^*, 112c", 112c"'which basically differ from each other in their respective edge steepness (slew rates). In Fig. 5b non- indicative frequency components suppressed by the band pass filtering means 90 are indicated by curly braces 93. It shall be understood that a wide variety of filter characteristics can be applied so as to define suitable sub bands.

Fig. 6 shows a schematic block diagram of another device 10 for extracting information from detected characteristic signals. In comparison with Fig. 4, the device 10 of Fig. 6 is modified in that the filter means 32 utilizes a transformation means 114 for applying a Fourier analysis to the band pass filtered signal 92. Hence, constituent frequency portions 116a, 116b, 116c can be derived so as to determine relevant sub bands 34a, 34b, 34c for further processing. Figs. 7a and 7b are referred to, illustrating an object 12 of interest being monitored when performing some fitness exercise on a fitness device 118. The object 12 is monitored by a remotely arranged detector means 24. During exercise, the object 12 usually performs characteristic movements indicated by arrows 122a, 122b, 122c. These movements show prominent frequencies corrupting the detected signals when the respective portion of the object 12 is monitored. Furthermore, also the device 118 can undergo some prominent disturbing motion indicated by arrows 120a, 120b. Hence, the detected signals can be even more corrupted due to adverse ambient motion. Basically, motion artifacts and desired vital signals conflict since they can be embedded in the same frequency band. A detector means 24' in Fig. 7a exemplifies that also detector means motion (e.g. camera motion) can cause large deviations.

For applications where the object 12 of interest is periodically moving, especially for certain fitness applications, object motion can be classified to certain spatial axes 124a, 124b. For instance, referring to the side view of Fig. 7a, it has been recognized that back and forth motion of both arms or both legs 122a, 122b "doubles" the respective arm or leg motion frequency with respect to the background. By contrast, sideward bending 122c of the object 12 and up and down motion of the object's torso basically lead to a "one-to-one" motion frequency. It goes without saying that these phenomena are highly dependent from the orientation of the sensor means 24 with respect to the object 12. However, for data processing and band split operations, generally two related prominent motion related frequencies can be expected and treated accordingly. Hence, conflicting frequencies can be detected and attenuated so as to further enhance the vital signal of interest.

Reference is made to Fig. 8 and Fig. 9 showing exemplary spectrograms illustrating results of remote photoplethysmographic analyses utilizing several approaches. In the diagrams, f denotes frequency while t denotes time. The frequency axis can represent Hz (Hertz) values while the time axis may stand for the number of processed image frames.

The spectrograms 126a, 126b, 126c of Fig. 8 exemplify the same situation, namely results obtained from a person performing some workout on a fitness device. Under these circumstances, objection motion renders the detection challenging. Furthermore, as the skin typically becomes sweaty during a workout, strenuous activities basically may imply further adverse effects on the detected characteristic index elements the desired signal is to be derived from.

The spectrogram 126a represents a remote PPG approach relying on difference components rather than absolute components, but without data stream decomposition prior to signal enhancement measures. Instead, the whole frequency band is processed. The spectrogram 126a merely exhibits one dominant frequency 128. However, this dominant frequency 128 is indicative of undesired object motion, e.g., the fitness exercise motion, rather than the desired signal(s) of interest.

Based on the same input data, the spectrogram 126b represents an approach further utilizing band pass filtering of the input data stream. For instance, the decompositions means passes relevant heartbeat frequencies (e.g., between about 40 Hz to about 220 Hz). As a result, two dominant frequencies 128, 130 can be detected. Besides the dominant object motion frequency 128 also the desired vital signal of interest 130, namely the heart rate, can be detected.

The spectrogram 126c is further enhanced in that the input data is band pass filtered and the passed frequency band is further split into two sub bands which are optimized individually. These refinements can lead to an even further enhanced dominant frequency 130 while the motion-related dominant frequency 128 is suppressed. Consequently, the approach relying on data (frequency related) decomposition prior to data processing and/or optimization enhances the signal-to-noise ratio, even under poor conditions.

Fig. 9 provides two spectrograms 132a, 132b exemplifying remote

photoplethysmographic analyses directed to an object having very dark skin.

The spectrogram 132a utilizes remote photoplethysmography applying defined pre-filtering approaches outlined above. It can be clearly seen that a dominant frequency 130 indicative of the desired signals has been enhanced so as to allow further signal analyses.

The spectrogram 132b is based on the same input data. However, no pre- filtering is applied to the data stream prior to signal enhancement measures. Hence, the frequency band is processed as a whole. The spectrogram 132b merely exhibits a dominant frequency 128 which is indicative of object motion. The heart rate related frequency 130 of the spectrogram 132a is hidden in noise.

Having demonstrated several alternative exemplary approaches covered by the invention, Fig. 10 is referred to, schematically illustrating a method for extracting

information from characteristic signals.

Initially, in a step 136 an input data stream or sequence comprising several frames 138a, 138b, 138c is received. A time axis is indicated by an arrow t. The data stream can be delivered from the sensor means 24 or a data buffer or storage means. The data stream can be embodied, by way of example, by a sequence of image frames varying over time. The image frames can comprise RGB based pixel data. The data steam comprises a representation of an object of interest.

In a subsequent step 140 portions of interest 142a, 142b, 142c can be selected in the data stream. The portions of interest 142a, 142b, 142c may comprise skin portions of an object on interest, e.g., a face portion of a human being to be observed. Non-indicative portions, e.g., clothes, hair, or further non- indicative surroundings, can be removed from the data stream. According to an exemplary embodiment, the portions of interest 142a, 142b, 142c can be selected and tracked by means of face detection. Furthermore, step 140 can comprise motion compensation measures directed to object motion and/or sensor means motion. Consequently, the problem of extracting the desired information can be facilitated.

In a further step 144 a detected pattern, e.g., the portions of interest 142a, 142b, 142c of data stream image frames, are normalized. Suitable approaches have been outlined above. By way of example, a pixel array having a certain dimension can be summarized in a single entity representative of mean values of image characteristics of the whole pixel array. The resulting normalized signal is indicated by 146a, 146b, 146c. When applying the RGB color space, the normalized entity can comprise (spatial) mean red, green and blue values. An exemplary representation of the normalized signal over time is indicated for illustrative purposes by reference numeral 146'. The normalized signal 146' comprises indicative and non-indicative portions. The non-indicative portion can be attributed to specular reflection of incident electromagnetic radiation, at least partially. The indicative portion can be attributed to diffuse body reflection of incident electromagnetic radiation, at least partially.

In an even further step 148, the data stream is decomposed. Decomposition can comprise band bass filtering, high pass filtering, low pass filtering and applying further suitable filter characteristics to the data stream. For instance, a non-indicative (frequency) portion 150 can be disregarded during further signal optimization. As a result, defined sub band portions 152a, 152b can be further processed separately.

In preceding step 154, the normalized signal and (sub band) signals are split up into additive components 156a, 156b, 156c from which they are composed. By way of example, the (sub band) additive components 156a, 156b, 156c can represent red, green and blue values. It goes without saying, that the additive composition can be inherent to the signal representation of the data stream (e.g. RGB signals). Therefore, alternatively, step 154 can be considered an illustrative step facilitating understanding. Viewed in terms of vector representation, vectors representing the sub bands of the normalized signal 146' in the signal space, e.g., RGB, are split up into their components.

In a further subsequent step 158, an arithmetic transformation is applied to the (sub band) additive components 156a, 156b, 156c. The arithmetic transformation results in (sub band) difference components 162a, 162b. The (sub band) difference components 162a, 162b comprise chrominance information rather than luminance information. The arithmetic transformation utilizes coefficients substantially summing to zero. By way of example, the (sub band) difference component 162a is derived through a transformation of the additive channels 156b, 156c. The transformation can be embodied by an addition comprising positive and negative coefficients summing to zero. The transformation results in "difference" values. Therefore, the transformation is indicated by a subtractive operator 160a. Similarly, (sub band) difference component 162b can be derived through a transformation of the (sub band) additive components 156a, 156b, 156c. The transformation is indicated by a subtractive operator 160b. Possible formulas and coefficients have been outlined above.

Eventually, the non- indicative (specular) reflection portion is suppressed in the

(sub band) difference components 162a, 162b as luminance information is "deducted", at least to some extent. In this way, a specular reflection portion and/or undesired object motion portion can be minimized or even removed from the initial signal.

In a further step 164 a deviation value or variance value, e.g., the standard deviation σ or possible derivates thereof, is determined for each of the (sub band) difference components 162a, 162b. To this end, moving windows 166a, 166b are applied to the time signal of the (sub band) difference components 162a, 162b.

In an even further step 168 the calculated deviation values are utilized for carrying out a weighting function. The weighting can be applied to the (sub band) difference components 162a, 162b. Additionally, a (sub band) signal 169 can be composed taking into account the (weighted) (sub band) difference components 162a, 162b. The weighting can be directed to minimize a variance of the composed (sub band) signal 169. The composed (sub band) signal 169 is highly indicative of the desired signals, e.g., heart rate or heat rate variability.

In a subsequent signal composition step 170 the relevant (processed) sub band portions 152a, 152b are combined so as to generate an optimized processed signal 172. It shall be understood that the optimized processed signal 172 can comprise the whole frequency band of the input data or the whole frequency band of band pass filtered input data (disregarding the non- indicative portion 150). In a further step 174, further analyzing measures are applied to the optimized processed signal 172. Finally, the desired signals can be extracted therefrom. For instance, a temporal pulsation in the optimized processed signal 172 is sought-for. The analyzing measures can comprise spectral analysis or frequency analysis. Reference numeral 176 depicts an exemplary spectral representation of the optimized processed signal 172. The spectral representation exposes a dominant frequency. A frequency axis is indicated by an arrow f. Furthermore, a time-based representation of a signal of interest 178 might be of interest.

By way of example, the present invention can be applied in the field of health care, e.g. unobtrusive remote patient monitoring, general surveillances, security monitoring and so-called lifestyle applications, such as fitness equipment, or the like. Applications may include monitoring of oxygen saturation (pulse oxymetry), heart rate, respiration rate, blood pressure, cardiac output, and changes of blood perfusion, pulse wave analysis, assessment of autonomic functions and detection of peripheral vascular diseases.

Needles to say, in an embodiment of a method in accordance with the invention several of the steps provided can be carried out in changed order, or even concurrently, unless explicitly indicated otherwise. In this connection, it is emphasized that the determined data decomposition prior to fundamental data processing is an essential aspect of the present disclosure. Further, some of the steps could be skipped as well without departing from the scope of the invention. This applies in particular to several alternative signal processing steps.

While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive; the invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

In the claims, the words "comprising" and "including" do not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single element or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

A computer program may be stored/distributed on a suitable non-transitory medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

Any reference signs in the claims should not be construed as limiting the scope.

Claims

CLAIMS:

1. Device for extracting information from remotely detected characteristic signals, comprising:

an interface (28) for receiving a data stream (26) derivable from electromagnetic radiation (14) reflected by a remote object (12), the data stream (26) comprising a continuous or discrete time-based characteristic signal (80) including physiological information and a disturbing signal portion, the physiological information being representative of at least one at least partially periodic vital signal (20), the disturbing signal portion being representative of at least one of an object motion portion and/or a non- indicative reflection portion, the characteristic signal (80) being associated with a signal space (72), the signal space (72) comprising complementary channels (74a, 74b, 74c) for representing the characteristic signal (80), components (82a, 82b, 82c) of the characteristic signal (80) being related to respective complementary channels (74a, 74b, 74c) of the signal space (72),

a decomposition means (32) for pre-processing the data stream (26, 30) by splitting a relevant frequency band (92) thereof into at least two defined sub bands (34a, 34b, 34c) comprising determined portions of the characteristic signal (80), each of which representing a defined temporal frequency portion potentially being of interest,

a processing means (36) for optimizing the sub bands (34a, 34b, 34c) so as to derive respective optimized sub bands (38a, 38b, 38c) from the at least two sub bands (34a, 34b, 34c), the optimized sub bands (38a, 38b, 38c) being at least partially indicative of a presence of the vital signal (20),

a composition means (40) for combining the optimized sub bands (38a, 38b, 38c) so as to compose an optimized processed signal (42).

2. Device as claimed in claim 1, further comprising an analyzing means (44) for enhancing and/or detecting the vital signal (20) in the composed optimized processed signal (42).

3. Device as claimed in claim 1, wherein the decomposition means (32) further comprises a band pass filtering means (90) for suppressing a selected non-indicative frequency component (93) of the characteristic signal (80) and/or for enhancing a selected indicative frequency component (92) of the characteristic signal (80).

4. Device as claimed in claim 1, wherein the decomposition means (32) further comprises a transformation means (114) for spitting the characteristic signal (80) into constituent frequencies (116), preferably the transformation means (114) is adapted for applying a Fourier transform to the characteristic signal, more preferably a discrete Fourier transform and/or a fast Fourier transform, so as to detect relevant sub bands (34a, 34b, 34c) of the characteristic signal (80).

5. Device as claimed in claim 1, wherein the signal space (72) is an additive color signal space, wherein the complementary channels (74a, 74b, 74c) are additive color channels, wherein the characteristic signal (80) is represented by at least three absolute components (82a, 82b, 82c), wherein the at least three absolute components (82a, 82b, 82c) represent distinct color components indicated by the additive channels (74a, 74b, 74c), wherein the additive channels (74a, 74b, 74c) are related to defined spectral portions.

6. Device as claimed in claim 5, wherein the processing means (36) further comprises a converter means (98) for separately transferring the sub band portions (34a, 34b, 34c) of the characteristic signal (80) by converting the at least three absolute components (82a, 82b, 82c) related to respective additive channels (74a, 74b, 74c) to at least two difference components (88; 162a, 162b) of the characteristic signal (80), wherein each of the at least two difference components (88; 162a, 162b) can be derived through a respective arithmetic transformation considering at least two of the at least three absolute components (82a, 82b, 82c), wherein the arithmetic transformation comprises additive and subtractive coefficients (160a, 160b), the disturbing signal portion being at least partially suppressed in the transferred optimized sub band portions (38a, 38b, 38c), wherein the arithmetic transformation comprises an at least partially subtraction of at least one of the at least three absolute components (82a, 82b, 82c) from the remaining absolute components, and wherein the arithmetic transformation for each of the at least two difference components (88; 162a, 162b) comprises coefficients at least substantially summing to zero.

7. Device as claimed in claim 5, wherein the processing means (3) further comprises an extractor means (102) for extracting the vital signal (20) from the transferred optimized sub band portions (38a, 38b, 38c) under consideration of an additive or subtractive expression or a ratio of the at least two difference components (88; 162a, 162b).

8. Device as claimed in claim 7, wherein the extractor means (102) is further adapted to normalize the extracted vital signal under consideration of a deviation value thereof, preferably a standard deviation, over a moving window applied to a sequence of the transferred optimized sub band portions (38a, 38b, 38c).

9. Device as claimed in claim 6, wherein the signal space (72) is indicative of luminance information and chrominance information, the chrominance information being representable by the at least two difference components (88; 162a, 162b), wherein the luminance information is representable by a luminance signal (84) being substantially aligned with a luminance index element (78) in the signal space (72), the luminance index element (78) being substantially indicative of a selected source (16) of electromagnetic radiation, wherein the at least two difference components (88; 162a, 162b) are substantially orthogonal to the luminance index element (78), preferably the at least two difference components (88; 162a, 162b) are substantially orthogonal to each other.

10. Device as claimed in claim 6, wherein the processing means (36) further comprises a weighting means (100) for weighting the at least two difference components (88; 162a, 162b) so as to derive weighted optimized sub band portions from the transferred optimized sub band portions (38a, 38b, 38c) under consideration of at least two weighted difference components, preferably the weighting is directed to minimize a spread of the weighted optimized sub band portions.

11. Device as claimed in claim 10, wherein the weighting comprises a

determination of a deviation value, preferably a standard deviation, of each of the at least two difference components (88; 162a, 162b), and wherein the deviation value of each of the at least two difference components (88; 162a, 162b) is determined under consideration of temporal variations thereof over a moving window (166a, 166b) applied to a sequence of each of the at least two difference components (162a, 162b).

12. Device as claimed in claim 1, wherein the at least one at least partially periodic vital signal (20) is selected from the group consisting of heart rate, heart beat, respiration rate, heart rate variability, Traube-Hering-Mayer waves, and oxygen saturation.

13. Method for extracting information from remotely detected characteristic signals, comprising the steps:

receiving a data stream (26) derivable from electromagnetic radiation (14) reflected by a remote object (12), the data stream (26) comprising a continuous or discrete time-based characteristic signal (80) including physiological information and a disturbing signal portion, the physiological information being representative of at least one at least partially periodic vital signal (20), the disturbing signal portion being representative of at least one of an object motion portion and/or a non-indicative reflection portion, the characteristic signal (80) being associated with a signal space (72), the signal space (72) comprising complementary channels (74a, 74b, 74c) for representing the characteristic signal (80), components (82a, 82b, 82c) of the characteristic signal (80) being related to respective complementary channels (74a, 74b, 74c) of the signal space (72),

pre-processing the data stream (26, 30) by splitting a relevant frequency band (92) thereof into at least two defined sub bands (34a, 34b, 34c) comprising determined portions of the characteristic signal (80), each of which representing a defined temporal frequency portion potentially being of interest,

optimizing the sub bands (34a, 34b, 34c) so as to derive respective optimized sub bands (38a, 38b, 38c) from the at least two sub bands (34a, 34b, 34c), the optimized sub bands (38a, 38b, 38c) being at least partially indicative of a presence of the vital signal (20), combining the optimized sub bands (38a, 38b, 38c) so as to compose an optimized processed signal (42).

14. Computer program comprising program code means for causing a computer to carry out the steps of the method as claimed in claim 13 when said computer program is carried out on a computer.