SYSTEM AND METHOD FOR DETECTING THE HEART BEAT RATE OF A PERSON IN A VEHICLE, AND SYSTEM AND METHOD FOR DETECTING FATIGUE

FIELD OF THE INVENTION

The invention relates to the monitoring of physical parameters of a person, such as a driver of a vehicle. More specifically, the invention relates to the monitoring of the heart beat rate or cardiac frequency of a person, and also to fatigue detection based on a detected heart beat rate.

STATE OF THE ART

Traditionally, the heart beat rate (also referred to as heart rate (HR) in this present text) or cardiac frequency has been monitored mechanically, for example, by sensing the pulsations of a blood vessel, and electronically using electrodes attached to the body. Also, as the electrical pulses corresponding to the heart beat also generate a low frequency magnetic field (equivalent to a dipolar magnetic moment of a few μAm ² ), techniques have been developed for measuring heart beat related parameters magnetically. Basically, these techniques are based on SQUID magnetometry, and have proven to be useful for medical so-called magnetocardiography (MCG) (cf., for example, US-B-6745063). However, SQUID magnetometry requires the use of complex and bulky devices and involves cryogenics. Some authors (Nathan A. Stutzke, et al., "Low-frequency noise measurements on commercial magnetoresistive magnetic field sensors", JOURNAL OF APPLIED PHYSICS 97, 10Q107 (2005)) have analysed the use of magnetoresistive field sensors, including spin valves, and concluded than the detectivity is too low to make such detectors useful for MCG applications.

On the other hand, in the field of automotive vehicles there has been an increasing interest in the detection of parameters useful for determining the physical state of the driver of the vehicle, for example, so as to detect a medical emergency condition or simply to detect fatigue of the driver. For

example, US-A-6946965 describes a prior art driver fatigue detector basically based on the detection of a lack of reaction of the driver to a stimulus, and EP-A-1477117 describes a driver fatigue detector based on the detection of a blinking motion of the eyelids of the driver. JP-A-11-151230 discloses a driver state measuring instrument which detects a physical condition of the driver using electrodes. The heart beat rate is detected by using electrical contacts on the steering wheel, and the variability of the heart rate is analysed to determine the physical condition of the driver. However, problems occur when the driver, for example, removes a hand from the steering wheel.

The analysis of the heart rate variability (HRV) is a known technique used to evaluate the cardiovascular changes produced during the awake- asleep cycle (cf . Task Force of The European Society of Cardiology and The North American Society of pacing and Electrophysiology, "Heart rate variability. Standards of measurement, physiological interpretation, and clinical use", Guidelines, European Heart Journal 1996; 17: 354-381).

Two major objective changes of the HRV between the awake and asleep states are well described in the literature:

(a) the heart rate (HR) decreases between 10 and 20%, between the moment when the person is completely awake and the moment when the person is completely asleep, but before reaching the first REM stage of the sleep;

(b) there are changes in the HRV (for example, the ratio between the spectral power density of the LF band (0,04-0,15Hz) and the HF band (0,15- 0,4Hz), LF/HF, decreases 50-70%) between the moment when the person is completely awake and the moment when the person is completely asleep, but before reaching the first REM stage of the sleep. (Cf.: Melinda Carrington, Michelle Walsh, 'The influence of sleep onset on the diurnal variation in cardiac activity and cardiac control", Journal of Sleep Research (2003) 12, 213-221; M. Nakagawa, T. Iwao, "Circadian rhythm of the signal averaged electrocardiogram and its relation to heart rate variability in healthy

subjects", Heart (1998) 79, 493-496; Andrzej BHan, Agnieszka Witczak, "Circadian rhythm of spectral indices of heart rate variability in healthy subjects", Journal of electrocardiology (2005) 38, 239-243; Helen J. Burgess, Jan Kleiman, "Cardiac activity during sleep onset", Psychophysiology (1999) 36, 298-306).

However, the literature focuses on the behaviour during the two states, but not on the transition between these states.

DESCRIPTION OF THE INVENTION A first aspect of the invention relates to a system for detecting the heart beat rate (that is, the cardiac frequency) of a person in a vehicle (for example, the driver or a passenger). The system comprises: at least one magnetic field sensor mounted inside the vehicle in a position close to a person's seat in the vehicle; and signal processing circuitry arranged to receive an output signal from said at least one magnetic field sensor, and to extract from said output signal data (such as specific values, or a signal indicative of said values) indicative of a heart beat rate.

The use of one or more magnetic field sensors makes it possible to overcome the disadvantages involved with prior art systems requiring a direct contact between the user and the equipment used to measure the heart beat rate (for example, direct ohmic contact necessary for obtaining ECGs).

Said at least one magnetic field sensor can, for example, be mounted in a seat belt for the person in the vehicle, or in the person's seat. Said at least one magnetic field sensor can comprise at least two magnetic field sensors, for example, two magnetic field sensors, both mounted in a seat belt for the person in the vehicle, both mounted in the person's seat, or one mounted in the person's seat and the other one mounted in the seat belt for the person. Said at least two magnetic field sensors can be arranged to be placed substantially symmetrically with respect to the person's heart when the

person is sitting in the vehicle, and/or said at least two magnetic field sensors can be arranged at different heights. The signal processing circuitry can be arranged to subtract an output signal from one of the magnetic field sensors from an output signal from another of said magnetic field sensor, so as to obtain a resulting signal less influenced by magnetic fields not originated by the heart of the driver.

The magnetic field sensors and the signal processing circuitry can be arranged so as to produce a subtraction of components of output signals from the magnetic field sensors that are related to external magnetic fields not originated by the heart of the driver, so as to obtain a resulting signal less influenced by magnetic fields not originated by the heart of the driver. This can, for example, be achieved by arranging two magnetic field sensors with their sensing axes in the same direction but opposed sense, and thereafter summing the output signal from these magnetic field sensors, using a summing circuit producing effective subtraction of signal components having different signs. Of course, the system must be arranged so as to prevent the components originated by the heart to be effectively subtracted from each other.

These arrangements make it possible to obtain a signal that can be used to detect the heart beat rate. It must be kept in mind that a motor vehicle is a difficult environment when one tries to perform low magnetic field measurements. The car itself has sources that generate magnetic fields (hard contribution) and has a lot of soft magnetic materials than distort the Earth's magnetic field (soft contribution). For devices using the Earth's magnetic field (high precision compasses, magnetic blind angle object detectors, etc.) located inside or near a car, these two contributions can be corrected. The standard procedure is based on turning the vehicle 360°, for example, a few complete turns on a roundabout, and plot the resultant in- plane field components on an X-Y plot; the resulting geometric figure is usually a non-centred ellipsoid. In a non-magnetic environment, the figure is expected to be a perfect circle with origin at (0,0). The deformation is due the

soft magnetic contribution of the car and the off-centring is caused by the hard magnetic contribution. Correcting the geometrical parameters of the experimentally obtained off-centred ellipsoid, converting it to a centred circle, allows compensation of the dc-magnetic field contributions of the car (cf. for example the procedure detailed on page 4 of EP-B1-1414003).

The hard contribution comes mainly from the engine block and normally represents an equivalent magnetic dipolar moment of between 100 and 500 Am ² . The soft contribution will have a low frequency component due the relative movement between the motor vehicle and the magnetic north. High electrical currents may also provide a significant contribution to the magnetic fields in the vehicle. Lights and signals represent the main low frequency contributions (the signals normally have a frequency of between 0.5 and 1 Hz).

The field measured by a magnetic field sensor inside the car can thus be determined by a plurality of dipolar magnetic sources and by the Earth's magnetic field. The contribution of each source to the total magnetic field normally varies with time. If the contribution of the heart of the driver is separated from the contribution from the other sources, the total magnetic field measured by a magnetic field sensor can be defined as:

B,=B| ^Hβart (t)+Bι ^Undβ8lred (t)= =k(t) ^Heart /(ri-r _H eart) ³ + ∑j k(t) ⁱ /(r _i -r _j ) ³ +Bearth(t)

where the constants k(t) are proportional to the equivalent dipolar magnetic moment of every source, n-rnear _t the distance between the sensor and the heart, and rj-η the distance between the sensor and the y-undesired source. B _earth (t) is the contribution of the Earth's magnetic field, which will be vary with time due the angular displacement of the car with respect to the magnetic north.

As the magnetic field is vectorial, the expression is valid for every magnetic field component. If two-axial or tri-axial magnetic field sensors are used, the expression should be applied to Bx, By and Bz.

If two magnetic field sensors are placed with their sensing directions arranged in parallel, the output signal from one of the sensors can be subtracted from the signal from the other sensor, thus subtracting the contributions to the magnetic field:

BrB ₂ =k(t) ^Heart ((n-rHeart)- ³ - (r ₂ -r _H eart) - ³ ) +∑j (k(t) ⁱ ((r ₁ -r _j )- ³ -(r ₂ -r _j )- ³ ))

If the sensors are placed closer to the monitored person than to the other sources, the first term will be magnified and the second will tend to zero. With a higher number of sensors, similar expressions can be obtained, even further reducing the contribution of the distant magnetic field sources. Another problem is to provide a magnetic field sensor output signal have the lowest possible signal/noise ratio. Depending on the sensors used, the problem can be the low resolution (2.7 nT for a HMC1001 sensor) or the noise (10-30 pT/Hz ^~1/2 for an SDT sensor). In both cases, the sensors should be placed as close as possible to the heart. Better sensors (like fluxgates, improved magneto resistive sensors or spin valves) can allow a larger distance between sensor and heart.

The ideal position for a magnetic field sensor trying to measure magnetic signals from the heart in a controlled ambient is the opening of the fourth intercostal space (the location of ECG lead V2). Now, when trying to measure parameters of the heart of the driver of the vehicle, it can be more difficult to correctly position the sensors with respect to the heart; also, the specific physical characteristics of the driver can vary (height, corpulence...). Placing two sensors separated several centimetres can help to reduce this problem (the heart will be close to one of the sensors, which will thus have a big contribution when subtracting the output signals; if the heart is placed

"between" the sensors, the contribution of the heart to each output signal will

be added when subtracting the output signals (as the sign of the contribution of the heart will be opposite for each sensor), whereas the undesired contributions (external magnetic fields) will probably have the same sign in each output signal). In summary, the position of the sensors is an important aspect when the issue is to get a signal good enough to allow a heart beat rate to be determined.

The signal processing circuitry can comprise an amplifier such as a low-noise, low offset differential amplifier (also known as instrumentation amplifier) and, in some cases, a derivation circuit.

The signal obtained from the sensor has a very low amplitude, but is amplified by the amplifier. By using a differential amplifier with its inputs fed with the signals from the sensors, the amplification can be made without too much amplification of the noise present in these signals. The signal thus obtained corresponds to a magnetocardiogram (MCG), that is, it shows the magnetic variations caused by the beating of the heart.

The MCG signal is a differential signal, that is, a signal obtained by measuring the difference between the magnetic characteristics at two different positions (when two sensors are used, these two positions correspond to these two sensors). The signal obtained from one of the sensors is used as a reference value for the other signal, and both signals are used by the amplifier. Now, in some cases, there is an excess of fluctuations in the reference signal. In these cases, a derivative circuit can be used to provide a more stable reference signal out of the instable one, whereby this stable reference signal can be applied to the amplifier to improve amplification performance.

A filter circuit can be used to remove the parts of the MCG signal that correspond to information not related to the heart beat rate (heart rate, HR) and also to remove part of the noise that is still present at the output of the amplifier. Butterworth filters provide good results, but when linear responses

(without signal distortion) are not required, Chebyshev filters or other types of filters with high attenuation of undesired signals can give the best results.

Thus, at the output of the filter, an electrical signal is obtained that basically contains the information indicative of the heart beat rate. The filter circuit may not be strictly necessary. However, depending on the sensor used, the output signal from the amplifier can be rather noisy and, in most cases, the R peaks of the MCG wave (that is, its maximum values) will not be clearly visible, wherefore the filter module can be necessary. As explained above, the main function of the filter module is to reduce the noise characteristics and to amplify the MCG characteristics of the signal at the output of the amplifier, in order to make the R peaks clearly detectable (cf., for example, H. Dickhaus, et ai, "CLASSIFICATION OF QRS MORPHOLOGY IN HOLTER MONITORING", Proceedings of The First Joint BMES/EMBS Conference Serving Humanity, Advancing Technology, Oct.13- 16, 1999, Atlanta, GA, USA; page 270; ©1999, IEEE).

Further, the signal processing circuitry can comprise an analogical-to- digital (A/D) converter for digitalizing the filtered signal, and a microprocessor unit arranged to mathematically treat the digitalized signal so as to extract the heart rate from the previously amplified and filtered MCG signal. The signal processing circuitry can comprise fuzzy logic means for extracting said signal or data indicative of a heart beat rate from said resulting signal. These fuzzy logic means can be implemented in the above- mentioned microprocessor unit, and can comprise an algorithm for performing calculus to reject "false MCG peaks" in the (amplified, filtered and) digitalized signal (for example, due to a non-perfect behaviour of the filter).

Even after the filtering and processing mentioned above, the RR- interval obtained (that is, the time distance between the subsequent peaks of the MCG wave) can still have erroneous values if the sensor output is of bad quality (which is likely to be the case inside a motor vehicle). To get a coherent RR-interval, it can be necessary to process the values using

medical rules (cf., for example, C. H. Kumar, ei a!., "A ROBUST R-R INTERVAL ESTIMATOR", Proceedings RC-IEEE-EMBS & 14th BMESI; page 1995; ©1995, IEEE), i.e., monitoring the R-R evolution corresponding to the last heart beats detected and assuring that this evolution correspond with a typical beat-to-beat time trend (this can also be implemented in the above- mentioned fuzzy logic means, by suitably programming the microprocessor unit with the relevant medical rules).

These classification techniques can be used to perform a real time analysis aiming at obtaining reliable heart beat rate data, taking into account information on typical heart rate evolutions.

To avoid confusion at beat detection, predictive fuzzy logic can be used (for example, based on learnings from information obtained from previous beats and/or information on normal heart beat rate trends) to reject "anomalous beats" not eliminated by preceding parts of the system. Thus, substantially correct beat time values (technically, the RR- intervals) can be obtained, and the successive values can be recorded in a memory. Even if no filtering module is used (for example, if the magnetic field sensors provide a sufficiently good and noise-less output), the anomalies (so- called "ectopic beats") can be detected and automatically filtered using a suitable algorithm (cf., for example, George B. Moody, "SPECTRAL

ANALYSIS OF HEART RATE WITHOUT RESAMPLING"; page 715; ©1993, IEEE).

Another aspect of the invention relates to a system for fatigue or drowsiness detection, which incorporates a system as described above and, further, a fatigue or drowsiness detector arranged to process the signal or data indicative of the heart beat rate to detect whether said data are indicative of fatigue or drowsiness of a person and, if said data are indicative of fatigue or drowsiness, to produce a fatigue or drowsiness warning event (for example, a visible and/or audible signal to alert a driver of the vehicle). In this context, we will use the term "fatigue" as a generic term, encompassing drowsiness.

The fuzzy logic means (if such means are incorporated) and the fatigue detector can, for example, be embodied in one single microprocessor unit.

The data processing for fatigue or somnolence detection can be performed in the same microprocessor unit as the one used for extracting the data concerning the heart beat rate, for example, by a special algorithm described below.

The accepted beat times (that is, heart rate indicative data such as "beat-to-beat" times, for the beats taken as "valid" beats in the above described process) can be stored in a memory buffer, typically storing at least 100 values. Once the buffer is full, at every beat, a new beat-to-beat time (or other heart rate indicative parameter) value can be stored into the buffer and the oldest one can be removed (that is, the buffer can operate as a classical FIFO buffer), whereby a new set of values can be obtained every time a new beat time value is recorded, approximately every second.

Thereby, a first set of values can be ready for processing some seconds (for example, 100 seconds) after start of the monitoring.

The recorded heart beat rate sample (that is, for example, the sample comprising 100 subsequently recorded "beat-to-beat" times) can then be analysed to extract somnolence information, for example, for the purpose of detecting that a driver will fall asleep minutes before it happens, to avoid accidents. Different analysis can be performed, for example, time and frequency analysis.

For example, the fatigue detector can comprise software arranged to detect fatigue by establishing, based on the data indicative of the heart beat rate, at least one reference value and at least one current value, said fatigue detector being arranged to trigger a fatigue warning event (such as an alarm signal) when at least one current value deviates more than to a predetermined extent from the corresponding reference value. The current value and the reference value can, for example, be values indicative of the data indicative of the heart beat rate (for example, values

corresponding to an average of the registered heart beat rate data stored in a memory), or of the variability of the data indicative of the heart beat rate, or values corresponding to a spectral analysis of the data indicative of the heart beat rate (such as a ratio between a low frequency component and a high frequency component of a curve corresponding to the heart beat rate spectra).

Actually, said at least one current value and said at least one reference value can comprise a plurality of current values and reference values, selected from the group comprising - a current value and a reference value indicative of the data indicative of the heart beat rate (such as corresponding to an average of said heart beat rate data);

- a current value and a reference value indicative of the variability of the data indicative of the heart beat rate; and - a current value and a reference value corresponding to a spectral analysis of the data indicative of the heart beat rate; whereby said fatigue warning event can be arranged to be triggered when at least two of the current values deviate more than to a predetermined extent from the corresponding reference values. These options will now be described more exhaustively.

A first possibility is temporal: the average beat time (time between subsequent R peaks) of the sample is lower (corresponding to a higher heart rate) when a person is awake than when the person is in a first sleep stage, corresponding to a drowsy state of the person (that is, when the person enters the drowsy state, there is a lower heart rate, and, thus, a longer beat- to-beat time). Monitoring the variation of the average beat time or heart rate, for example, taking the average of the last 50-500 beats, a somnolence parameter can be obtained. Using, for example, 100 samples, a threshold set between 5% and 15% of increase of the average beat time has been found to give rise to a drowsiness warning about 4 to 7 minutes before the driver falls asleep.

Another possible parameter for monitoring the drowsiness, using a temporal analysis, is based on the variability of the beat time over the sample. When a person is awake, he/she has a larger variability of the beat time interval (or the heart rate) than when he/she is at the initial sleep stage, that is, at the drowsy stage.

Beat time interval or heart rate variability can be calculated using statistical parameters over the sample of recorded data (for example, on the last 50-500 pieces of recorded data). The easiest way to implement this method may be using the standard deviation of the RR interval, or the square root of the mean squared differences of successive RR intervals. Using standard deviation, the variability of the HR or the beat time interval decreases around 40% between the awake and asleep states. Monitoring this parameter and its evolution in subsequent samples each comprising, for example, 100 pieces of data, a decrease of between 10% and 30% can be used to trigger a drowsiness warning 4 to 8 minutes before the driver falls asleep.

A third method is based on a frequency analysis. The spectral power density of the heart rate can be calculated at different bands, for example, at the so-called LF band (0,04-0,15Hz) and HF band (0,15-0,4Hz). The LF band is associated with the sympathic systems and the HF band with the parasympathic (or vagal) systems of the person. The LF/HF ratio, also known as the sympatho-vagal balance, is high when the person is awake (the symphatic systems, LF, prepares the body for activity) and low when the person is asleep (the parasympathic-vagal systems, HF, prepares the body for relax) (cf.: John Trinder, Jan Kleiman, "Autonomic activity during human sleep as a function of time and sleep stage", Journal Sleep Research (2001) 10, pp. 253-264).

The obtained and stored values concerning the RR intervals define a discontinuous tachogram. The (for example) last 50-500 values can be interpolated to obtain a continuous signal, so that it is possible to analyze its spectrum. A typical value for the interpolation can be 2Hz. The spectrum can

be calculated using different approaches like the FFT, Yu!e-Wa!ker, Burg, or Lomb-Scargle methods. Then the spectral power density of the LF band (0,04-0,15Hz) and HF band (0,15-0,4Hz) can be calculated. The values can be recalculated every time a new beat time is entered into the memory, thus providing, for every new beat, an updated information on the variation of LF and HF spectral power density. Using the spectral power density calculated using the last 100 recorded samples, when the LF/HF decreases by for example 50% with respect to the initial awake state, a fatigue warning can be triggered; with the numbers mentioned above, this would typically take place between 4 to 6 minutes before the driver actually falls asleep.

Each one of the three drowsiness indicators may produce (depending, inter alia, on the person who is being monitored) a certain number or false alarms, especially if the thresholds are set to give the warning far in advance of the actual moment of falling asleep (that is, if low thresholds are used to trigger the alarm). To minimise the false alarms, a combination of two or more of the above mentioned parameters can be used. For example, standard variation and LF/HF ratio can be combined using an AND function (whereby the fatigue warning will only be issued when both parameters indicates danger of falling asleep). The above-mentioned methods are only examples of methods that can be used to detect fatigue on the basis of a detected heart rate.

The person referred to above can be a driver of the vehicle, but also a passenger (it can be interesting to monitor also the state of the passengers, for example, so as to hold information on the passengers' physical state in the case of an accident).

Another aspect of the invention relates to a vehicle, including a system according to any of the preceding claims (including, for example, the respective sensors placed in one or more seats and/or seatbelts of the vehicle, for monitoring the heart rate of the driver and/or passengers). A further aspect of the invention relates to a method for detecting the heart beat rate of a person in a vehicle. The method comprises the steps of:

arranging or disposing at least one magnetic field sensor inside the vehicle in a position close to a person's seat in the vehicle; receiving an output signal from said at least one magnetic field sensor; and extracting, from said output signal, data indicative of a heart beat rate.

What has been said about the system is also applicable to the method, mutatis mutandis.

For example, said at least one magnetic field sensor can be mounted in a seat belt for the person in the vehicle, and/or in the person's seat. For example, said at least one magnetic field sensor can comprise at least two magnetic field sensors. These sensors can be mounted in the seat belt for the person in the vehicle, or in the person's seat, or one sensor can be mounted in the person's seat and the other one in the seat belt. Said at least two magnetic field sensors can be placed substantially symmetrically with respect to the person's heart when the person is sitting in the vehicle, and/or arranged at different heights.

An output signal from one of the magnetic field sensors can be subtracted from an output signal from another of said magnetic field sensor, so as to obtain a resulting signal less influenced by magnetic fields not originated by the heart of the driver.

Components of output signals from the magnetic field sensors that are related to external magnetic fields not originated by the heart of the driver can be effectively subtracted from each other (for example, by arranging the sensors with their sensing axes in the same direction but opposite sense, and then summing the measured signals), so as to obtain a resulting signal less influenced by magnetic fields not originated by the heart of the driver.

The data indicative of a heart beat rate can be extracted from said resulting signal, for example, by using fuzzy logic means.

The person can be a driver of the vehicle. A further aspect of the invention relates to a method for fatigue detection, for detecting fatigue of a person in a vehicle, comprising the

method described above, and further comprising the steps of processing the data indicative of a heart beat rate to detect whether said data are indicative of fatigue of a person and, if said data are indicate of fatigue, producing a fatigue warning event. The processing of the data indicative of a heart rate can comprise the step of establishing, based on the data indicative of the heart beat rate, at least one reference value and at least one current value. The fatigue warning event can be triggered when at least one current value deviates more than to a predetermined extent from the corresponding reference value, that is, when the deviation between the current value and the reference value exceeds a pre-established threshold, for example, a threshold set to a fixed amount or a threshold expressed as a percentage of the reference value.

For example, at least one current value and reference value can be values indicative of the data indicative of the heart beat rate (for example, indicative of an average of said data), and/or at least one current value and reference value can be values indicative of the variability of the data indicative of the heart beat rate, and/or at least one current value and reference value can be values corresponding to a spectral analysis of the data indicative of the heart beat rate (for example, said current value and reference value can correspond to a ratio between a low frequency component and a high frequency component of a curve corresponding to the heart beat rate spectra).

Said at least one current value and said at least one reference value can comprise a plurality of current values and reference values, selected from the group comprising

- a current value and a reference value indicative of the data indicative of the heart beat rate (for example, indicative of an average of said data);

- a current value and a reference value indicative of the variability of the data indicative of the heart beat rate; and - a current value and a reference value corresponding to a spectral analysis of the data indicative of the heart beat rate. Thus, said fatigue

warning event can be arranged to be triggered when at least two of the current values deviate more than to a predetermined extent from the corresponding reference values.

BRIEF DESCRIPTION OF THE DRAWINGS

To complete the description and in order to provide for a better understanding of the invention, a set of drawings is provided. Said drawings form an integral part of the description and illustrate preferred embodiments of the invention, which should not be interpreted as restricting the scope of the invention, but just as an example of how the invention can be embodied.

The drawings comprise the following figures:

Figure 1 : Block diagram of the main components of a system in accordance with a preferred embodiment of the invention.

Figures 2a and 2b: Schematically illustrate possible positions of the magnetic field sensors. The arrows indicate the sensing axes if uni-axial sensors are used.

Figure 3: Block diagram of the magnetic field sensor arrangement. Figure 4: Block diagram of the signal processing circuitry. Figure 5: Flowchart showing a possible algorithm for obtaining data indicative of the heart beat rate

Figures 6A-6C: Flowcharts showing three appropriate algorithms for fatigue detection.

Figure 7: Block diagram showing how different approaches for detecting fatigue can be combined to reduce the risk for "false alarms".

DESCRIPTION OF PREFERRED EMBODIMENTS OF THE INVENTION

In accordance with a preferred embodiment of the invention shown in figure 1 , the system comprises a magnetic field sensor module 1 comprising suitably arranged magnetic field sensors, and an electronic signal processing circuitry 2. Also, if the system is a system for fatigue detection, a fatigue detector 3 or somnolence processor can be included.

The magnetic field sensor module 1 comprises, in this embodiment, two uniaxial-fluxgates sensors, (such as FGM-3, produced by Speake&Co), a double regulation power supply, a frequency to voltage converter and a summing (or subtracting) circuit. The magnetic field sensor module detects a signal component 1a related to the magnetic field of the heart, caused by the electrical pulses of the heart, and also a signal component 1 b originated by other sources, not related to the heart beat. An output signal 1c from the magnetic field sensor module is supplied to the electronic signal processing circuitry 2, which obtains, from said signal, data indicative of the heart beat rate (for example, data indicating the relative time position of subsequent detected heart beats, or the time between subsequent beats). These data 2a can be used as an input to the fatigue detector. Fatigue detector and signal processing circuitry can obviously be implemented in one single processor module. As shown in figures 2a and 2b, the two uni-axial magnetic field sensors 11 and 12 can be placed in the seatbelt 100 (figure 2a) or in the seat 101 (figure 2b) of a vehicle, with their sensing axes (illustrated by arrows in figures 2a and 2b) parallel to the chest or back, respectively, of the monitored person. If the sensors are placed in the seatbelt, their sensing axes can be arranged perpendicularly to the longitudinal direction of the seatbelt. The sensing axes of the sensors are opposed in figures 2a and 2b (this allows effective subtraction of the sensed signals by using a summing circuit).

This configuration allows good detection of the heart beat related magnetic signal component because the sensing axes are parallel to the main heart magnetic field component.

The sensors can be powered from the battery of the vehicle. In order to reduce supply voltage variations, a double regulation can be used (see FGM-series Magnetic Field Sensors Application Notes, http://www.fatquarterssoftware.com/downloads/fqmapp.pdf), decreasing the voltage from 12-15 V to 9 V first, and then to 5 V.

The fluxgate outputs are rectangular pulses whose frequency varies inversely proportional to the magnetic field. The frequency output of every sensor is converted to voltage using a frequency to voltage converter such as LM2907 or equivalent. The two voltages are then put into a summing circuit 13 (which, from a system point of view, can be considered to be included in the electronic signal processing circuitry), as schematically illustrated in figure 3 (elements illustrated in figures 1 and 2 are illustrated using the same reference numerals in figure 3) (in figure 2, the sensing axes of the magnetic field sensors are parallel and directed in opposed senses, whereby a summing circuit 13 can be used for effective subtraction of noise components; if the sensing axis were aligned in the same direction and sense, a subtracting circuit could obviously be used for effective subtraction of the same noise components, that is, of corresponding components in the output signals from the sensors that are due to external magnetic fields not related to the beating of the heart, cf. what has been stated above concerning elimination of non-desired signal components). A variable resistor on one of the inputs makes it possible to adjust the weight of the contribution of each magnetic field sensor, for zeroing the summing (or subtraction) circuit output during calibration. To calibrate the sensors, the arrangement can be placed inside a pair of Helmholtz coils, with the sensing axes direction and the axes of the coils oriented E-W. When a small current passes through the coils, the output of the sum circuit should be zero if both sensors have exactly the same calibration constant. If not, adjusting the variable resistor a zero output can be obtained. The signal processing circuitry is illustrated in figure 4. The output signal 1c from the magnetic field sensor module 1 is supplied to the input of an instrumentation amplifier 21 (such as INA138, from Burr-Brown), with enough gain to obtain a voltage signal with a maximum dynamic range defined by the supply voltage (for example, from 0 to 5V). If the environment where the system is used has a high-power magnetic fluctuation, a derivative circuit 22 can be used, based on an inverting operational amplifier (any

standard operational amplifier can be used) in derivative configuration. This derivative circuit can be used to create a virtual reference signal for the instrumentation amplifier in order to compensate this fluctuation.

Afterwards, the signal is fed to a bandpass filter module 23, based on a quad-operational amplifier (such as LM2902). The filter can comprise two stages, with the following characteristics:

- Stage 1 : high pass, 2 ^nd order, Butterworth active filter with a cutting frequency of 5Hz and +5dB of gain.

- Stage 2: low pass, 4 ^th order, Butterworth active filter, using two operational amplifiers, with a cutting frequency of 20Hz and +15dB of gain.

After amplification and filtering a signal indicative of the heart beat rate is obtained, and can be digitalized with an analogue-to-digital (AJD) converter 24 with, for example, at least 8 bits of resolution. This converter can obviously be integrated in a microprocessor or digital signal processor (DSP).

In any case, the digitalized signal is introduced into a microprocessor 25 (or DSP) which processes the signal in order to detect when every beat occurs, and thus produces data directly indicative of the heart beat rate (such as a series of numbers indicating the beat-to-beat time of subsequent beats). Figure 5 schematically illustrates how the output signal of the analogue-to-digital converter 25 is sampled (501 ) by signal processing means associated with the microprocessor. The processing means continue to sample the signal until a (local) maximum is detected (502), which is interpreted as the detection of a new beat (503), whereby the time position of the beat and the magnitude or amplitude of the signal at that moment are registered (503). Next, it is checked (504) whether the magnitude of the "new beat" is much higher than that of the previous beat. If so, it is considered (505) that the previous beat was an invalid beat (due to noise, for example), and the value (magnitude and time position) of the new beat replaces the one of the previous beat. If not, it is checked (506) whether the magnitude of the new beat is similar to the magnitude of the previous beat. If it is not similar, it

is considered (507) that the new beat is a "false positive", that is, that it does not correspond to a beat, and a new sample (501 ) is obtained. Also, the

"false positives" are counted (508) and if they are considered to be too many, the system interprets that it has a bad reference to compare with the new detected beats and resets itself by deleting (509) the information stored as

"previous beat", which is used as a reference for the "false positive" decision.

Now, if the magnitude of the "new beat" is similar to the magnitude of the last detected beat (506), it is checked (510) whether the chronological separation between the new beat and the previous beat is similar to the separation in time between the previous beat and the beat preceding that one. If not, this is once again taken as a "false positive" (507). If yes, the beat is taken as valid beat (511 ), and the value(s) (such as time position, or delay in time versus the previous beat) replaces the corresponding value(s) of the previous beat, in a FIFO memory buffer (the values corresponding to previous beats are moved towards a "discharge" end of the buffer, and when the buffer is full, every time a new beat is registered, the oldest registered beat is removed). The detection of a valid "new beat" can also trigger the fatigue detector, if the system includes such a detector.

Thus, as can be understood from what has been discussed above, a filtering of "anomalous beats" or "false positives" can be performed both on the basis of the magnitude/amplitude of the detected signal, and of the position in time of the detected "beats", comparing with data obtained from previous beats and/or with data prestored in the system (relating, for example, to pre-established maximum and minimum beat-to-beat times). For example, if the last "beat-to-beat" distance is less than 80% or more than

120% of the previous "beat-to-beat" distance, this last beat can be considered anomalous and therefore filtered out from the sample (that is, considered to be a "false positive").

The fatigue detector can be arranged to operate every time a new "valid" beat has been detected and added to the memory buffer or similar, which can be of the FIFO ("First In First Out") type.

Basically, once a set of data relating to the heart beat rate (such as the beat-to-beat time) has been obtained (for example, once a set of 128 beat-to- beat times has been detected and recorded in the memory buffer), a reference value can be obtained. Next, every time a new piece of data is entered into the memory buffer (whereby the oldest piece of data is removed, if the FIFO type buffer is used), the corresponding current value is counted on the basis of the new set of data. The current value is compared to a predetermined threshold, and if it exceeds said threshold, a fatigue warning event can be triggered (for example, an audible and/or a visible signal can be generated).

Different approaches are schematically illustrated in figure 6. According to a first possible approach, when a buffer (such as a buffer having 128 memory positions for storing 128 subsequently registered beat- to-beat times, in a FIFO manner) is filled for the first time, a "reference value" is calculated (611 ), this reference value being the average of the beat-to-beat times registered in the buffer at that time. Subsequently, every time a new beat-to-beat time is entered into the buffer (and the "oldest" previous beat-to- beat time is deleted from the buffer content), a "current value" is calculated (612), the current value being the average beat-to-beat time of the new buffer content. Next, it is checked (613) if the current value is more than X% of the reference value, X being typically 110-120. If the current value exceeds this threshold, a fatigue warning event is triggered (614). If the current value is not above said threshold, a new beat-to-beat time value is obtained and stored in the buffer (and the oldest beat-to-beat time is removed from the buffer), and the process is repeated (steps 612-613).

According to a second possible approach, when the buffer is filled for the first time, a reference value is calculated (621 ), the reference value being the standard deviation of the beat-to-beat times registered in the buffer. Subsequently, every time a new beat-to-beat time is registered in the buffer (and the "oldest" previous beat-to-beat time is deleted from the buffer content), a current value is calculated (622), the current value being the

standard deviation of the new buffer content. !t is checked (623) if the current value is more than Y% below the reference value, Y being typically in the order of 40. If the current value is more than Y% below the reference value, a fatigue warning event is triggered (624). If not, a new beat-to-beat time value is obtained (and the "oldest" one is removed from the buffer), and the process is repeated (steps 622-623).

According to a third possible approach, when the buffer is filled for the first time, a reference value is calculated (631 ). This is done by interpolating the buffer content (for example, applying a 2Hz interpolation), so as to obtain a corresponding continuous signal. To this resulting signal, the Burg algorithm is applied, so as to obtain the spectrum of the signal. Next, the spectral power density is calculated for the LF band (0,04-0,15Hz) and for the HF band (0,15-0,4Hz), and by division the LF/HF ratio is obtained. This LF/HF ratio based on the first 128 valid samples is the reference value. Subsequently, each time a new valid beat is detected and the corresponding beat-to-beat time is introduced in the buffer (and the "oldest" previous beat- to-beat time is deleted), a new interpolation is performed so as to obtain a corresponding continuous signal (632), and subsequently the spectral power densities for the LF and HF bands are calculated and the LF/HF ratio is obtained (633); this new LF/HF ration is the current value. Subsequently, it is checked (634) whether the current value is more than Z% below the reference value, Z being typically in the order of 50. If the current value is more than Z% below the reference value, a fatigue warning event is triggered (635). If the current value is not below said threshold, a new beat-to-beat time value is obtained and stored in the buffer (whereby the "oldest" one is removed form the FIFO buffer), and the process is repeated (steps 632-634).

"AND" logic 700 can be used to "combine" two or more of the approaches mentioned above, so as to produce an "effective fatigue warning event" 701 when two or more of said approaches has produced their corresponding "individual" fatigue warning events (614, 624, 635), as

schematically illustrated in figure 7. If so, no warning signal is sent to the user until said "effective fatigue warning event" is produced.

In this text, the term "comprises" and its derivations (such as "comprising", etc.) should not be understood in an excluding sense, that is, these terms should not be interpreted as excluding the possibility that what is described and defined may include further elements, steps, etc.

On the other hand, the invention is obviously not limited to the specific embodiment(s) described herein, but also encompasses any variations that may be considered by any person skilled in the art (for example, as regards the choice of materials, dimensions, components, configuration, algorithms, etc.), within the general scope of the invention as defined in the claims.