The technical field of the invention is the determination of a stress level of an individual on the basis of at least one physiological-parameter measurement carried out on the individual. The stress level is determined using a membership function, established using fuzzy logic.
It is possible to determine a stress level of an individual on the basis of measurements of one or more physiological parameters of said individual. The physiological parameter may be a cardiac activity, for example measured via an electrocardiogram (ECG) or a simple determination of a cardiac frequency, or a muscular activity, measured via an electromyogram (EMG), or even a measurement of the electrical conductance of the skin. The publication Ollander “A comparison of wearable and stationary sensors for stress detection”, 2016 IEEE international conference on Systems, Man and Cybernetics (SMC), October 2016, describes how certain physiological parameters may be used to establish a stress indicator for an individual.
The emergence of portable connected sensors intended to be worn by an individual has made it possible to take measurements of physiological parameters both simply and inexpensively. It is for example a question of specific sensors able to be fastened to a bracelet or integrated into watches or connected to smart phones. For example, the device Empatica E4 comprises various sensors allowing physiological parameters such as electrodermal activity, cardiac activity or temperature to be easily accessed.
On the basis of the measured parameters, classification algorithms may be implemented, so as to determine whether the individual is in a stressed state or in a rest state. Certain classification algorithms are based on fuzzy logic. This type of algorithm is for example described in the publication Kumar M “Fuzzy techniques for subjective workload-score modeling under uncertainties”, IEEE transactions on systems, man and cybernetics—part B, Vol. 38, No. 6, December 2008. This type of algorithm requires a learning phase to be carried out, in which an individual is placed in a stressful situation, or in various stressful situations. The fact that a learning period, in which the individual is placed in a stressful situation, is required is constraining. In addition, the reliability of such methods may be compromised by physiological variability from one individual to the next.
The inventors propose a method for determining a stress level of an individual that does not require the individual to be placed in a stressed state during calibration. This allows the method to be implemented in a particularly simple manner, using equipment worn by the individual. In addition, the calibration may be repeated periodically.
A first subject of the invention is a method for determining a stress level of an individual, depending on a physiological parameter of the individual, the value of which is liable to vary depending on the stress level of the individual, the method comprising the following steps:
The method may be such that
According to an embodiment, c) comprises taking into account a threshold distance, such that
According to an embodiment, c) comprises calculating an dispersion indicator of the physiological-parameter values measured in the various calibration periods, the threshold distance been determined depending on the dispersion indicator. The dispersion indicator may be or may comprise an extent of the range of variation, corresponding to a deviation between a minimum value and a maximum value of the range of variation. The threshold distance may be obtained by applying a scale factor to the extent of the range of variation.
c) may comprise attributing a value representative of the range of variation, according to which:
According to an embodiment, the physiological parameter is or comprises:
According to an embodiment:
the method further comprising:
e) combining the stress levels determined relative to each physiological parameter, in d), in order to obtain a multi-feature stress-level index.
The combination may be or comprise a sum, or a weighted sum. In step e), the multi-feature stress-level index may be determined by calculating a weighted mean or a median of the stress levels respectively determined relative to each physiological parameter.
In this embodiment, in step c), each membership function may be defined independently of the others.
In this embodiment, with each physiological parameter may be associated a threshold distance and a range of variation of the physiological-parameter values measured during the calibration periods.
In one embodiment, the membership function, or each membership function, is a function defined in an interval comprised between the range of variation and the range of variation increased by the threshold distance. It may be continuous in this interval.
Another subject of the invention is a device for determining a stress level of an individual, comprising:
Other advantages and features will become more clearly apparent from the following description of particular embodiments of the invention, which are given by way of nonlimiting example, and shown in the figures listed below.
The physiological parameter may be a parameter representing the cardiac frequency (or heart rate). If HRj is the heart rate measured at an instant j, the physiological parameter x(t) at t may be:
where Nj is a number of heart reat measurements taken into account, and t−Nj≤j≤t. The number Nj is set so as to include the measurements of the hear rate during a sliding duration of a few seconds or a few tens of seconds, of a few minutes, for example 60 seconds.
The physiological parameter may be a parameter representing the inter-beat interval. If IBIj is the inter-beat interval measured at an instant j, the physiological parameter x(t) at t may be:
where Nj is a number of inter-beat intervals taken into account, and t−Nj≤j≤t. The number Nj is set so as to include the measurements of the inter-beat interval during a sliding duration of a few seconds or a few tens of seconds, of a few minutes, for example 60 seconds.
The physiological parameters described in the two preceding paragraphs are to be considered as preferred parameters.
The objective of the invention is to determine a stress level Sl(t) of the individual in various measurement periods t.
The sensor 2 is connected to a microprocessor 4, the latter being connected to a memory 5 in which are stored instructions for implementing the method described above. The microprocessor 4 receives the measurements of the sensor 2, via a wired link or a wireless link. The microprocessor 4 may be worn/borne by the individual, being arranged with the sensor or being incorporated into an ancillary device carried by the person, for example a portable object such as a smart phone. The microprocessor 4 may also be remote from the individual.
The calibration phase comprises steps 100 to 120. An important aspect of the invention is that in this phase, the individual is at rest, or, more precisely, considers himself as being at rest. Thus, the calibration phase comprises calibration periods tr, in which periods the individual is considered as being solely in a rest state. He is therefore not in a stressed state. In the calibration phase, the sensor 2 measures a physiological-parameter value xr(tr), in various periods tr, the index r designating the fact that the individual is considered as being at rest. If the calibration is carried out while the individual is in a certain stressed state, this degrades the reliability of the determination of the stressed state of the individual at measurement times subsequent to the calibration.
By rest state of an individual, what is for example meant is a state in which the individual is awake, but his physical and mental activity is minimal. For example, the individual is alone, sat or lying down, and performing no particular activity. In the rest of the description, the rest state corresponds to the state in which the individual is during the calibration.
In the measuring step 100, the sensor measures the physiological-parameter value xr(tr) corresponding to the calibration period tr.
In step 110, the calibration period tr is incremented, then step 100 is reiterated or the iteration loop formed by steps 100 and 110 is exited. In the step 110, the iteration loop may be exited at the end of a preset number of iterations, or depending on the values xr(tr) measured in the various calibration periods tr. For example, it is possible to calculate a statistical quantity from the values xr(tr) measured in the calibration periods t, and to stop the iterations depending on a variation in this statistical quantity, and in particular when the variation in the statistical quantity is negligible. The statistical quantity may be the mean, or the median, or a dispersion indicator such as variance, standard deviation, or a deviation between a maximum value xr,max and a minimum value xr,min of the physiological parameter x.
In step 120, the values xr(tr) measured during the various calibration periods tr are used to define the membership function ƒ. For example, a range of variation Xr is defined in which the values xr(tr) measured during the various calibration periods tr, which are called calibration values, lie. The range of variation Xr is bounded by a minimum value xr,min and a maximum value xr,max. Thus, Xr=[xr,min,xr,max]. The range of variation may be characterized by its extent Δxr. The extent is such that Δxr=xr,max−xr,min (1).
Step 120 may comprise applying a statistical test in order to eliminate aberrant values xr(tr). A Dixon test, known to those skilled in the art, may for example be performed. The elimination of aberrant values allows the reliability of the method to be improved.
It is also possible to determine a threshold distance dS. The threshold distance dS may be such that:
dS=α×Δxr, (2), α being a real positive number, designated by the term scale factor. The value of the scale factor α depends on the physiological parameter in question. It is typically comprised between 0.1 and 0.5. The scale factor allows the threshold distance dS to be determined from the scope Δxr, as described below.
The membership function ƒ is intended to define a stress level Sl on the basis of a physiological-parameter value x(t) measured in a measurement period t, subsequent to the calibration phase. The stress level Sl may for example vary between 0 and 1, 0 corresponding to a rest state and 1 corresponding to a stressed state of the individual. According to the principles of fuzzy logic, the membership function ƒ may define intermediate levels, comprised between 0 and 1, and corresponding to an intermediate stressed state. The membership function ƒ is preferably continuous in an start space E defined by the values that the measured physiological parameter is capable of taking. The start space E may for example be the set of real positive numbers. Thus, ƒ: E=+→[0,1] and ƒ(x(t))=Sl(t).
An example of a membership function ƒ is illustrated in the
According to one variant, shown in
The example given above is valid when the membership function ƒ is an increasing function, i.e. when the stress level increases as the measured value of the physiological parameter increases. In certain particular cases, for example when the parameter in question is skin resistance, the membership function ƒ is a decreasing function: as the stress level increases, the measured parameter value decreases. In such a configuration:
The normalization by dS allows intermediate state levels comprised between O (x(t)→xr,min) and 1 (x(t)→xr,min−dS) to be obtained. The arrow → means “tends toward”.
Generally, it is possible to attribute a representative value to the range of variation Xr. Depending on the distance d between the parameter value x(t) and the representative value, the membership function ƒ determines a stress level Sl(t). In the examples given above, the representative value was respectively set equal to the maximum value xr,max and to the minimum value xr,min of the range of variation. In other examples, the representative value may be a statistical indicator applied to the calibration values xr(tr) measured during the calibration. It may for example be a question of the mean Xr or of the median med(Xr) of the calibration values. It may also be a question of a fractile, for example a quartile (the first quartile, when the membership function is a decreasing function or the fourth quartile when the membership function is an increasing function) or a decile (for example the first decile when the membership function is a decreasing function or the tenth decile when the membership function is an increasing function). The stress level, corresponding to a measured parameter value x(t), may then be calculated depending on the distance between the value of the parameter and the value representative of the range of variation. The distance d may be normalised by an indicator of the dispersion of the calibration values, for example the extent Δxr of the range of variation or the standard deviation of the calibration values xr(tr).
The scale factor α may be determined depending on an indicator of the dispersion of the values measured during the calibration. The dispersion indicator may be the extent Δxr of the range of variation Xr. It may also be a question of a variance or a standard deviation of the calibration values xr(tr).
Steps 130 and 140, which are described below, correspond to a phase of use of the sensor 2 to estimate a stressed state of the individual for whom the membership function ƒ was defined. In step 130, a measurement of the physiological parameter x(t) is carried out in a measurement period t. The physiological parameter x(t) measured in each measurement period is the same as that measured in the calibration periods.
In step 140, the membership function ƒ defined beforehand is applied to the value x(t) of the physiological parameter in the measurement period, so as to determine a stress level Sl(t)=ƒ(x(t)). At the end of step 140, the period may be incremented and another iteration of steps 130 to 140 carried out.
Thus, in the method described above, the calibration is carried out solely with parameter values measured in the calibration, while the individual is considered as being in a rest state. The calibration does not require parameter measurements to be carried out while the individual is in a stressed state. One advantage of the method is that the calibration is faster and simpler to carry out. Another advantage is that the calibration may be repeated periodically, in order to take into account a possible physiological variability of the user. In this case, when a repetition is desired, following step 140, the method implement steps 100 to 120. Since the calibration is particularly simple to carry out, it is possible to frequently repeat the calibration.
One example application is shown in
According to one embodiment, the method described above may be applied while simultaneously measuring various parameters xi, the index i identifying the parameter in question, with 1<i≤I, I designating the number of physiological parameters in question.
For each physiological parameter xi, a calibration is carried out in steps 100i, 110i and 120i. These steps are respectively similar to steps 100, 110 and 120 described above. They are respectively implemented on the basis of values xr,1(tr) . . . xr,i(tr) . . . xr,I(tr) of the physiological parameters in question, in different calibration periods tr.
Each step 120i is carried out considering a range of variation Xr,i, of the parameter xi during the calibration. The range of variation Xr,i has an extent Δxr,i. To each physiological parameter xi is assigned a scale factor αi. It will be noted that the scale factor αi may be different from one physiological parameter to the next. The step 120i allows a membership function ƒi relative to the physiological parameter xi to be defined. The membership functions ƒi, ƒi+1, respectively associated with two different parameters xi,xi+1, are established independently of each other. It is however preferable that, for each membership function, the rest state and the stressed state correspond respectively to the same levels, for example 0 for the rest state and 1 for the stressed state. Thus, a definition of I membership functions ƒ1 . . . ƒI, respectively associated with the I measured physiological parameters x1 . . . xI, is achieved. To each physiological parameter in question may correspond one range of variation, determined in the calibration, and one threshold distance. Each membership function ƒi is established depending on the range of variation and on the threshold distance that are associated with each physiological parameter.
After each membership function ƒi has been defined, the method comprises a step 130i, implemented for each parameter xi(t) measured in a measurement period t, so as to determine a stress level Sli(t) associated with each parameter xi(t), according to the expression Sli(t)=ƒi(xi(t)). Thus, a definition of I stress levels Sl1(t) . . . SlI(t), respectively associated with the I physiological parameters x1 . . . xI in question, is achieved.
In a step 150, the various stress levels Sl1(t) . . . SlI(t), respectively associated with each parameter xi(t), are combined, so as to determine an overall, or multi-feature, stress level Sl(t), according to the principles of fuzzy logic. The combination may be a calculation of a mean value or of a median value. It may also be a question of a weighted mean, in which each stress level Slit) is assigned a weighting factor λi dependent on the importance that it is desired to attribute to the physiological parameter xi relative to the other parameters in question. The various stress levels Sl1(t) . . . SlI(t) may be combined by applying predetermined inference rules.
On the basis of the multi-feature stress level Sl(t), it is possible to determine whether the individual is in a stressed state. Initial experimental trials have shown that when the multi-feature stress level Sl(t) is higher than 0.3 or 0.4, the individual may be considered to be in a stressed state in the measurement period.
Moreover, on the basis of the multi-feature stress level Sl(t), it is possible to define an activity state, corresponding to an intermediate state between the rest state and the stressed state. The activity state corresponds to an individual performing a normal mental or physical activity, without being in a stressed state. The activity state corresponds to a multi-feature stress level lying between:
Experimental trials have been carried out on a cohort of 20 subjects aged from 19 to 30 years, who were successively subjected to four different stressful situations:
In each test, three types of physiological signals were measured:
The signals were acquired as an acquisition frequency of 1000 Hz.
In a preliminary phase, the parameters x, or features, that were the most suitable for the detection of a stressful situation were selected. It was a question of:
The value of each parameter was calculated for a measurement period comprised between 3 and 5 minutes.
For each parameter, a membership function was established, as described above.
Next, each parameter, and its membership function, were combined in order to form a multi-feature stress indicator. This index was obtained by calculating a mean of the value of the membership function for each parameter.
It may be seen that the highest values of the stress indicator are obtained for the TSST test.
The invention will possibly be employed to track the stress level of individuals. It may for example be a question of tracking stress level in a professional environment, or of tracking the stress level of individuals who are subject to anxiousness in particular situations, for example in a means of transportation. It may also be applied to track the stress level of an athlete.
Number | Date | Country | Kind |
---|---|---|---|
18 56504 | Jul 2018 | FR | national |