WO2015058923A1

WO2015058923A1 - Device and method for estimating the energy expenditure of a person

Info

Publication number: WO2015058923A1
Application number: PCT/EP2014/070599
Authority: WO
Inventors: Sandrine Magali Laure Devot; Koen Theo Johan De Groot; Alberto Giovanni BONOMI; Emile Josephus Carlos Kelkboom; Francesco SARTOR; Antonius Hermanus Maria Akkermans
Original assignee: Koninklijke Philips NV
Current assignee: Koninklijke Philips NV
Priority date: 2013-10-24
Filing date: 2014-09-26
Publication date: 2015-04-30
Anticipated expiration: 2016-04-24

Abstract

The present invention relates to device for estimating the energy expenditure of a person with an increased accuracy. The proposed device comprises a movement sensor (12) for detecting physical activity of the person and for generating a movement signal (13), a heart rate sensor (14) for detecting the heart rate of the person and for generating a heart rate signal (15), an estimation unit (16) for estimating the energy expenditure of the person from said movement signal (13) and/or said heart rate signal (15), and a control unit (18) for controlling said estimation unit (16) to switch between an non-activity mode in which only the movement signal (13) is used for estimating the energy expenditure if no physical activity of the person has been detected and an activity mode in which the movement signal (13) and/or the heart rate signal (15) is used for estimating the energy expenditure if physical activity of the person has been detected.

Description

Device and method for estimating the energy expenditure of a person

FIELD OF THE INVENTION

The present invention relates to a device and method for estimating the energy expenditure of a person.

BACKGROUND OF THE INVENTION

Physical activity can reduce the risk of developing certain diseases, such as obesity or hypertension and improve the overall well-being and quality of life of a person. An accurate and objective quantification of the physical activity is therefore becoming more and more essential, especially to support public health research and intervention programs and clarify the dose-response relation between physical activity and health. The use of accelerometers is very common to predict energy expenditure, as e.g. described in Bonomi AG, W. K. (2011), Advances in physical activity monitoring and lifestyle, International Journal of Obesity, 1-11. They are cheap, simple to use and wearable. However, it is commonly admitted that some physical activities may be misinterpreted by such devices. As an example, it is difficult to accurately estimate energy expenditure during activities such as cycling, in particular for accelerometers worn on the wrist or at the waist. Indeed, very little motion is picked up by the accelerometer, resulting in a significant underestimation of the energy expenditure. Hence, adding a heart rate (HR) sensor leads to a more accurate estimation of the energy expenditure, especially when the HR-based prediction equation is adjusted for subject characteristics (such as age, gender, BMI, etc.) and/or individual fitness level (HR at rest, V0₂max, etc.). If the HR sensor is furthermore integrated in a wearable sensor (e.g. in a belt or in a wrist), the greater accuracy is achieved without negatively impacting the ease-of-use.

Over the past years, many studies have shown that physical activity can reduce the risk of developing certain diseases, such as obesity or hypertension and improve the overall well-being and quality of life. An accurate quantification of the physical activity is therefore becoming more and more essential, especially to support intervention programs. The use of accelerometers is very common to predict energy expenditure, but it is known that some physical activities (especially the ones that demand higher efforts) may be

misinterpreted. JP 10-179560 A discloses a calorimeter made of an acceleration sensor, a pulse sensor and a signal processing card. The CPU of the signal processing card makes a judgment about an exercising state on the basis of an acceleration signal from the

acceleration sensor. Also, the CPU monitors a change in the number of pulses detected with the pulse sensor, converts measured pluses to the number of pulses per constant time and compares the converted number of pulses with the number of pulses at a normal time (nonexercising time) previously stored in a memory. In this case, when the acceleration signal outputted from the acceleration sensor does not show an exercising state, the CPU calculates the consumed calories (i.e. the energy expenditure) on an average time basis, using a proportional relation the number of pulses to the consumed calories. Also, when the acceleration signal from the acceleration sensor shows an exercising state, the CPU judges the exercising state, and calculates a calorie consumed per average time in each exercising state based on the exercising state and speed, and even personal information. In this way the calories consumed during exercise time and a non-exercise time shall be accurately determined.

US 20120083705 Al discloses an activity monitoring system and method which is adapted to estimate energy expenditure using either acceleration information only, heart rate data only, or a combination of these data. The prior art method is adapted to select the appropriate energy estimation equations when the heart rate sensor is positioned in a body location for which heart rate data is unreliable (e.g. torso).

SUMMARY OF THE INVENTION

It is an object of the present invention to provide a device, processor and processing method for more accurately estimating the energy expenditure of a person without substantial additional efforts and means.

In a first aspect of the present invention a device for estimating the energy expenditure of a person is presented comprising:

a movement sensor for generating a movement signal,

a heart rate sensor for detecting the heart rate of the person and for generating a heart rate signal,

an estimation unit for estimating the energy expenditure of the person from said movement signal and/or said heart rate signal,

wherein the movement sensor is adapted to detect a physical activity of the person depending on a value of the movement signal, the device further comprising: a control unit for switching on the heart rate sensor during a predetermined amount of time corresponding to the detection of physical activity of the person and such that the movement signal and/or the heart rate signal is used for estimating the energy expenditure, and for switching off the heart rate sensor otherwise such that only the movement signal is used for estimating the energy expenditure.

The predetermined amount of time may correspond to the physical activity period of the person as detected by the movement sensor. Alternatively, the predetermined amount of time is a recovery period starting at the end of the physical activity period of the person and ending at when the heart rate of the person becomes stable.

In a further aspect of the present invention a method for estimating the energy expenditure of a person is presented, the method comprising:

estimating the energy expenditure of the person from a movement signal provided by a movement sensor and/or a heart rate signal provided by a heart rate sensor for detecting the heart rate of the person,

wherein the method further comprising

detecting physical activity of the person depending on a value of the movement signal,

switching on the heart rate sensor during a predetermined amount of time

corresponding to the detection of physical activity of the person and such that the movement signal and/or the heart rate signal is used for estimating the energy expenditure, and

switching off the heart rate sensor otherwise such that only the movement signal is used for estimating the energy expenditure.

In yet a further aspect of the present invention, there is provided a corresponding computer program which comprises program code means for causing a computer to perform the steps of the method disclosed herein when said computer program is carried out on a computer as well as a non-transitory computer-readable recording medium that stores therein a computer program product, which, when executed by a processor, causes the method disclosed herein to be performed.

Preferred embodiments of the invention are defined in the dependent claims. It shall be understood that the claimed method, processor, computer program and medium have similar and/or identical preferred embodiments as the claimed system and as defined in the dependent claims.

The present invention is based on the recognition that combining a movement sensor, such as an accelerometer, with a HR sensor leads to a better estimation of the energy expenditure, but HR sensors can be more power demanding than accelerometers. The addition of a HR sensor leads therefore to a (significant) reduction of the battery life of the whole device. Further, the HR sensor can be prone to motion artefacts, leading to inaccurate HR measurements during the physical activity. Still further, the linear relationship between HR and energy expenditure that is used for prediction mostly holds for aerobic activities. Thus, switching off the heart rate sensor when appropriate makes it possible to extend the battery life of the HR sensor or even of the complete device, if the battery of the device supplies also the HR sensor.

One element of the present invention is therefore to appropriately switch from a movement-based estimation to a HR-based estimation (and vice- versa) so as to take the most out of two methods. In particular, the movement-based energy expenditure estimation is used whenever there is a non-physical activity, which is also referred to as the non-activity mode of the estimation unit. This will include low intensity activities, such as sedentary activities, desk work, housework, etc. The HR-based energy expenditure estimation is used whenever there is a high physical activity, which is also referred to as the activity mode of the estimation unit. This will include light to moderate and vigorous physical activity. In this way the accuracy of the energy expenditure estimation is improved.

The present invention is further based on the recognition to know when the heart rate signal is less useful in the context of energy expenditure estimation. It is neither generally known nor expected nor obvious that a movement signal of the person is more useful than a HR signal to predict energy (i.e. calories) expenditure during inactivity. During inactivity there is quite a lack of motion so that the movement signal would rather be considered to provide a poor estimation of the energy expenditure. On the other hand, during inactivity the HR signal may suffer from influences related to posture, stress, emotions and others and may thus not accurately reflect the expended energy.

Thus, in contrast to what is known or expected, the present invention proposes to use the movement signal during inactivity because it generally represents a less power consuming measurement modality than HR signal measurements, at least with many practical embodiments of movement sensors and HR sensors. Further, the selection of the movement signal for use in the estimation of the energy expenditure during inactivity provides a reproducible and stable outcome to the user while measured in absence of motion. On the other hand, the HR signal would fluctuate during inactivity causing a poorly realistic variation in the energy expenditure during rest. Preferably, a reliability check is introduced, according to which said control unit is configured to control said estimation unit to check the reliability of the heart rate signal and, if physical activity of the person has been detected, to use only the heart rate signal for estimating the energy expenditure if the heart rate signal is considered reliable and to use only the movement signal for estimating the energy expenditure if the heart rate signal is considered unreliable or not available. This reliability check ensures that no noise HR signal measurements are used which would degrade the quality of the energy expenditure estimation.

In a preferred embodiment the device further comprises a classification unit for extracting one or more features of the movement signal and to determine if there is physical activity or no physical activity of the person by use of a classification based on the extracted one or more features of the movement signal. Those extracted features may be one or more of the peaks, the intervals between peaks, the rise times, the fall times, the number of peaks, the gradient, the intensity, the duration of peaks, the similitude or correlation between subsequent intervals, the periodicity, and the spectral energy of the movement signal. For the feature extraction various algorithms or methods may be used, such as the generally known logistic regression, a neural network, a support vector machine, Naive Bayes, a linear or quadratic discriminant. These embodiments provide for further improvements of the accuracy of the energy expenditure estimation.

Said classification unit may further be configured to detect the type of physical activity, in particular to detect a cycling, walking, running or cross-training activity, in which embodiment it is further preferred that said classification unit comprises two or more classifiers, each using a predetermined set of features for detecting a predetermined type of activity. In this way the accuracy of the detection of the activity state of the person can be further improved. For instance, for detecting a cycling activity may be easier by use of a different set of features that are preferably used for detecting a running activity. The knowledge of the type of activity may be used to improve the estimation of the energy expenditure. For instance, a different estimation algorithm may be used. Indeed, different activities involving specific muscle groups and certain type of muscle action (isometric or isotonic) influence specific heart rate responses which are uniquely linked to changes in energy expenditure. This shows how activity type information can be used to generate activity-specific equations to predict energy expenditure from heart rate data.

Still further, in an embodiment said control unit is configured to switch said estimation unit between a non-activity mode and an activity mode only if the same physical activity status of the person has been detected for a predetermined time period or if for a predetermined percentage of the last number of time windows the same physical activity status of the person has been detected. In other words, a smoothing is applied to prevent the estimation unit from switching between the two modes too often and too quickly, which may lead to inaccurate estimations, may even make the estimation impossible or may even damage the estimation unit or the control unit.

Preferred embodiments of said movement sensor are an acceleration sensor or a gyroscope or GPS or any tool capable of sensing human purposeful body motion. Preferred embodiments of said heart rate sensor are a photoplethysmograph PPG sensor or an ECG sensor or a heart sound sensor capable of detecting heart beats.

Said movement sensor, said heart rate sensor, said estimation unit and/or said control unit, preferably all those units, are built into a body worn device, in particular a wristband, wristwatch, smartphone or body belt. In other embodiments, only the sensors are worn on the patient's body, while the estimation unit and the control unit, which are preferably implemented by a processor and/or by software (such as an app), may be arranged far from the patient, e.g. in a computer, laptop, tablet, smartphone, etc. In this case a connection is established between all the units, e.g. as wireless connection (such as using WLAN or Bluetooth or the like) or as wired connection. The device may also be

implemented in a fitness device or exercise machine as used at home or in gyms.

BRIEF DESCRIPTION OF THE DRAWINGS

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiment(s) described hereinafter. In the following drawings:

Fig. 1 shows a diagram illustrating heart rate signal measurements,

Fig. 2 shows a diagram illustrating results of energy expenditure estimations based on movement signals and based on heart rate signals,

Fig. 3 shows a schematic diagram of a first embodiment of a device according to the present invention,

Fig. 4 shows a schematic diagram of a second embodiment of a device according to the present invention,

Fig. 5 shows a flow chart illustrating a first embodiment of a method according to the present invention,

Fig. 6 shows a flow chart illustrating a second embodiment of a method according to the present invention, Fig. 7 shows a flow chart illustrating how to control activation and deactivation of HR sensor in accordance with an embodiment of the present invention,

Fig. 8 shows a flow chart illustrating a first part of third embodiment of a method according to the present invention, and

Fig. 9 shows a flow chart illustrating a second part of third embodiment of a method according to the present invention.

DETAILED DESCRIPTION OF THE INVENTION

As mentioned above, the use of a HR (heart rate) sensor has many advantages, but it has been found that it has also some limitations. In particular, the linear relationship between energy expenditure and HR mostly holds during (steady) aerobic activities. This leads to two main issues:

i) At rest, or at low intensity of physical activity, HR-based estimation methods may be inaccurate. Most of them assume a constant value for the energy expenditure, namely using a rough estimate of the Basal Metabolic Rate. As an alternative, an individual calibration in a laboratory may be used, but it is a cumbersome procedure, which is not desired for a large-scale consumer product.

ii) In addition, external factors (such as mental fatigue or stress) may lead to a HR change without a real link with the metabolism. Such factors are more likely to negatively impact the energy expenditure procedure during daily life activities than during exercise. During physical exercise, the HR change due to metabolism demand is expected to be largely dominant compared to other potential causes of HR change.

Further, the HR measurements can be prone to motion artefacts, especially during activities with non-repetitive movements, i.e. mostly during daily life activities (sedentary activities, housework, desk work, etc.). Inaccurate HR measurements will lead to inaccurate estimates of the energy expenditure. Conversely, during exercise, periodic movements are expected; they can be more easily filtered out, resulting therefore into more accurate HR estimates as can be seen from Fig. 1 showing a diagram illustrating heart rate signal measurements. In said diagram heart rate signals are shown for various activities taken with a conventional chest belt HR sensor (signal SI), low quality (Q) (S2), medium quality (S3) and high quality (S4). The signals S2, S3 and S4 in Fig. 1 indicate the heart rate signal collected using a HR sensor based on photo-plethysmography. The difference between SI, S2, and S3 is that they represent the measured signal in presence of low noise, medium noise or high noise level, which impact the signal quality. Still further, a HR sensor is more power demanding than an accelerometer. Therefore, the addition of the HR sensor (significantly) increases the power consumption of the device and leads to a (significant) reduction of its battery life.

Hence, it is an element of the present invention to switch between an acceleration-based estimation and a HR-based estimation (and vice- versa) so as to take the most out of two methods. Basically, it means use the acceleration-based energy expenditure estimation method whenever there is a non-physical activity and to use the HR-based energy expenditure estimation method whenever there is a physical activity. This improves the accuracy of the energy expenditure estimation, i.e. most of the above mentioned limitations are mitigated, as can be seen from Fig. 2 showing a diagram illustrating results of energy expenditure estimations based on movement signals ("acceleration-based") and based on heart rate signals ("HR-based"). The term AEE stands for activity energy expenditure or activity-induced calories expenditure. The AEE error represents the difference between measured and estimated AEE.

Fig. 3 shows a schematic diagram of a first embodiment of a device 10a for estimating the energy expenditure of a person according to the present invention. It comprises a movement sensor 12 for detecting physical activity of the person and for generating a movement signal 13, a heart rate sensor 14 for detecting the heart rate of the person and for generating a heart rate signal 15, an estimation unit 16 for estimating the energy expenditure

17 of the person from said movement signal and/or said heart rate signal, and a control unit

18 for controlling said estimation unit (by use of a control signal 19) to switch between an non-activity mode in which only the movement signal is used for estimating the energy expenditure if no physical activity of the person has been detected and an activity mode in which the movement signal and/or the heart rate signal is used for estimating the energy expenditure if physical activity of the person has been detected.

Fig. 4 shows a schematic diagram of a second embodiment of a device 10b for estimating the energy expenditure of a person according to the present invention. In addition to the device 10a it comprises a classification unit 20 for extracting one or more features of the movement signal 13 and to determine if there is physical activity or no physical activity of the person by use of a classification based on the extracted one or more features of the movement signal 13. The control unit 18 and the estimation unit 16 (and the classification unit 20, if available) may be implemented by a processor or computer, e.g. as a software program or an app. Said processor may be combined with the movement sensor 12 and the HR sensor 14 into a common device worn by the person, e.g. as wristband, wristwatch or the like. In another embodiment the processor may be arranged away from the person who only carries the movement sensor 12 and the HR sensor 14, which are connected to the processor in any suitable way (wireless or wired).

Details and further embodiments of the invention will be explained in the following by reference to Figs. 5 to 9 showing flowcharts of preferred embodiments of the steps of a method according to the present invention.

A first embodiment of a method 100 is depicted in Fig. 5. The process takes as input the movement signal 13, in particular an acceleration signal (ACC) recorded during one epoch. In a first step S10, carried out by the classification unit 20 or the control unit 18, one or more features of the movement signal 13 are extracted, e.g. based on standard pattern recognition techniques. Such features may include one or more the peaks, the intervals between peaks, the rise times, the fall times, the number of peaks, the gradient, the intensity, the duration of peaks, the similitude or correlation between subsequent temporal intervals, the periodicity, and the spectral energy of the movement signal 13. Example of classification methods are logistic regression, neural networks, support vector machine, Na^'ive Bayes, Linear/Quadratic discriminant, etc. These classifiers can be based on a single feature or on multiple features.

In a second step S 12 the epoch is classified as C_HR if it is recognized as a physical (e.g. aerobic) activity or if not. If the epoch is assigned to the class C_HR, the HR- based energy expenditure (EE) prediction shall be used. Otherwise, the acceleration-based energy expenditure prediction shall be used. The process is then repeated for the next epoch.

In particular, if the epoch is assigned to the class C_HR (i.e. in activity mode) it is checked in step S14 if the HR sensor is switched on. If the HR sensor is switched on the reliability of the HR signal measurement is checked in step SI 6. Checking the reliability of the HR signal prevents from using noisy HR measurements for energy expenditure prediction, increasing therefore the overall accuracy. Different methods can be used to assess the reliability of the HR measurements. In particular, a reliability index can be defined to compare with a pre-defined threshold and ensure a minimum quality. Finally, if the HR signal is found reliable, the energy expenditure is estimated by use of the HR signal in step S I 8. Energy expenditure estimates can be derived from HR according to a model including subjects characteristics providing data on fitness level, gender, basal metabolic rate and resting heart rate. An example is presented in a published report which shows how the heart rate can estimate energy expenditure according to the HR-FLEX method (Rennie et al., Estimating energy expenditure by heart-rate monitoring without individual calibration, Medicine and Science in Sports and Exercise, 2000). Otherwise, the energy expenditure is estimated by use of the acceleration signal in step S26. The acceleration signal can be summarized in epochs of fixed lengths and the resulting features often called "activity counts" can be used in linear equations to estimate energy expenditure together with subjects characteristic data such as body weight and height (Bonomi et al., Estimation of Free-Living Energy Expenditure Using a Novel Activity Monitor Designed to Minimize Obtrusiveness Obesity, 2010).

If it has been found in step S 14 that the HR sensor is not switched on, the HR sensor is now switched on in step S20. It should be noted that, depending on the

implementation, when the HR sensors is switched on in step S20, no HR data are available for a certain about of time given the activation time of the HR sensor and the need to compute and average HR data from a sequence of heart beats. Thus, even when the HR sensor is switched on in step S20, the EE estimation can still be based on the acceleration signal in step S26 in a first instance. When the HR data is available or a predetermined time after starting the HR measurement, the method continues with step S I 6.

If the epoch is assigned to the class C_ACC in step S 12 (i.e. in non-activity mode), it is checked in step S22 it is checked if the HR sensor is switched on. If this is the case the HR sensor is switched off in step S24, where after the energy expenditure is estimated by use of the acceleration signal in step S26. If the HR sensor is already switched off, step S26 is directly carried out.

As explained above this method provides the advantage of a more accurate estimation of the energy expenditure. Moreover, the on-time of the HR sensor is limited to the periods where the HR-based method is used for energy expenditure estimation, reducing therefore the power consumption and increasing the battery life.

It should be noted that all types of movement-based (in particular acceleration- based) and HR-based methods for estimations can be used according to the present invention. In particular, the HR-based estimation might use a general equation for all users, or might use 'static' subject characteristics, such as weight, age, sex, BMR, BMI, height or more dynamic characteristics, related to fitness level, such as resting HR, or V0₂max. The HR-based estimation might also be adjusted by correcting for the individual fitness level, e.g. by monitoring the HR during specific activities and calibrating the equation used for the estimation according to these individual values.

Another embodiment of a method 200 is depicted in Fig. 6. In addition to the steps of the method 100, some further steps are added in this method 200.

In practice, the HR estimates may not always be available. Following a period with the HR sensor switched off, some delay is expected to get again accurate HR values. An extra step is therefore added to handle the actual on and off switches of the HR sensor. In particular, the availability and reliability of the HR signal measurement during the epoch under consideration is checked. After it has been found in step S14 that the HR sensor is switched on, the HR signal measurement is continued in step S28.

Moreover, it is appropriate to add a mechanism to avoid potential repetitive on and off switches of the HR sensor from epoch to epoch. Frequent switches of the hardware can be detrimental, especially if the HR signal measurement needs some time to adjust and delivers HR signal values with some delay. This is expected to happen for sensors that are unobtrusive and therefore more prone to artefacts. Depending on the sensor, the on and off switch mechanism might therefore need some smoothing to prevent from falling back systematically to the acceleration-based method because the HR measurements are not available and/or not reliable. Hence, in steps S30 and S32 appropriate stability conditions are checked to prevent a hysteresis effect due to switching.

An exemplary stability condition 1 checked in step S30 may be that it is checked if there is at least pi % of epochs classified as C_HR in the past nl epochs. An exemplary stability condition 2 checked in step S32 may be that it is checked if there is at least p2 % of epochs classified as C_ACC in the past n2 epochs. In embodiments moving windows of length nl epochs and n2 epochs, respectively, is applied and it is looked for a majority of consistently classified epochs. The parameters pi, p2, nl, n2 are tuned depending on the trade-off that is to be achieved.

It shall be noted that the present invention is not limited to these stability conditions. Any mechanism that performs a smoothing on the outcome of the classification unit may as well be used. In the above methods, estimation of energy expenditure EE for a given sample is either based on acceleration data or heart rate HR data. It is assumed that acceleration data is available throughout the entire period the system is activated, i.e. the acceleration sensor is always switched-on and exhibits body acceleration data properly. The HR sensor is only activated when needed, mainly to save battery life, and is steered by acceleration data, possibly extended with supplementary data. The decision to deactivate the HR sensor is steered by acceleration and/or HR data, possibly extended with supplementary subject specific data. When only acceleration data is available (since the HR sensor is switched-off or is in a start-up/initialization mode), EE estimation is based on acceleration data, possibly extended with supplementary data. When both acceleration data and HR data are available, HR data is being analyzed to select only HR information in the EE process at appropriate time instances. Switching between acceleration based EE estimation and HR based EE estimation when the HR sensor is active, is steered by HR data and acceleration data. The mechanism to properly select HR data as modality to estimate EE during physical inactive periods, is mainly introduced to estimate accurately burned calories during the excess post- exercise oxygen consumption period and periods with intense physical activities. During the excess post-exercise oxygen consumption period, the body tries to restore itself to its pre- exercise state by consuming oxygen.

The logic to control activation and deactivation of HR sensor, as well as modality selection in the EE estimation unit, is explained by means of a state flow chart illustrated in Fig. 7. The figure contains several thresholds indicated by T and an

accompanying indicator. The control unit outputs variable b, which is Boolean indicating activation of the HR sensor (b=l) or deactivation (b=0). The control unit also produces a Boolean variable s that indicates which modality to select in the EE estimation unit: either ACC data or HR data will be selected when s=0 or s=l, respectively. The classification unit outputs C and the HR sensor outputs HR (if available as HR is not permanently activated) which are delivered to the control unit.

In a default state S30, the HR sensor is deactivated (i.e. b=0) and EE estimation is based on ACC (i.e. s=0). As long as C < Ta, the system remains in the default state S30. Whenever an epoch classification estimate C exceeds a threshold Ta (test S31), the system moves into a state S32 wherein a timer is reset and starts counting. Next the system goes immediately to state S33, meaning that the evaluated epoch was assessed as an active period. The effect is activation of the HR sensor, signaled by b=l . In S33, EE is based on ACC (i.e. s=0). As long as the monitored HR does not exceed threshold Ti (test 34), the system stays in state S33, i.e. EE estimation of subsequent epochs will be based on ACC, unless the counter exceeds threshold Tt (test S35). If the predetermined time Tt defined by the timer is elapsed, because HR has not exceeded threshold Ti, the system goes to the default state S30. This additional check and timer functionality is introduced to prevent that the HR sensor remains active while HR does not exceed the criterion S34: HR > Ti for a prolonged duration of time. This would be undesirable as leaving out the timeout would negatively impact the battery life time of the HR sensor.

If the current state is S33 and the HR does satisfy during an epoch the criterion 34 HR > Ti, the system goes to state S36. In this state the HR remains activated (i.e. b=l). Estimation of EE of subsequent epochs remains on the basis of HR, until criterion S37 HR < Te is met. If in state S36, HR becomes lower Te, the system falls back to the default state S30 again, meaning that the HR sensor will be deactivated and EE estimation of subsequent epochs will be based on ACC. Experiments showed that settings for the thresholds are for example Ta=0.5, Tt=5, Ti=k*Hrest and Te=k*Hrest-d. Here, the resting heart rate Hrest refers to an attribute of subject specific data. Parameter Hrest, may be determined during a dedicated protocol upon estimation of caloric expenditure through acceleration and heartbeat information. Moreover, the resting heart rate could be updated by requesting the user to conduct a similar protocol at a later stage. Parameter d, could be regarded as a deadband setting to prevent oscillation or repetitive activation-deactivation cycles of the HR sensor when the system is in state S36. Parameter k is a multiplying factor. The deadband parameter is for example set to d=5 and the multiplying factor k=1.5.

In the above description, it is assumed that a HR value is immediately available once the HR sensor is activated. This is not always feasible in practice. To solve this issue, an additional check in state S33 is introduced to evaluate the availability of HR after activation of the HR sensor. If the system goes from S32 to state S33 and no HR is available after entering S33, the system remains in S32 and EE estimation is based on ACC.

Still another embodiment of a method of estimating energy expenditure is shown in Figs 8 and 9. This embodiment corresponds to a mode where the HR sensor is activated during a predetermined period just after the end of a physical activity. As shown in Fig. 8, the start of the physical activity is firstly detected in a step S71 by analyzing epochs of the acceleration signal ACC (13). When the start of the physical activity is detected, the start time Tstart (75) is issued and the method then seeks for the detection of the end of the physical activity in a step S72. Once the end of the physical activity is detected, the end time Tend (76) is issued and the HR sensor is turned on in a step S73 and starts measuring the heart rate (HR) during a recovery period to determine HRrecovery (74), the recovery period ending when the heart rate becomes stable.

After HRrecovery is measured, the method can start predicting the expended energy during the activity as shown in Fig. 9. Firstly, the heart rate value at the end of the activity HRend (82) is estimated in a step S81 based on the HRrecovery measurement (74). This estimation may be required in practice, as it will take a few seconds before the HR- sensor is turned on and starts measuring reliable HR values. Thus, the initial drop of HR during the recovery period is therefore likely to be missed. Secondly, HR during the activity HR(Tstar Tend) (85) is estimated in a step S84 using as input HRend (82) and the acceleration data during the activity ACC(Tstart:Tend) (83). The expended energy (88) is finally estimated in a step S86 using a function f based on HR(Tstart:Tend) (85) and the necessary auxiliary data (87), such as personal information, EE = f(HR(Tstart:Tend), auxiliary data). In this way the problem of motion artefact correction during specific activity types may be bypassed by using only HR recovery data to extrapolate the energy expenditure during the previous physical activity. A further advantage is related to power savings derived from the frugal activation of the HR sensor which occurs during recovery only.

In a preferred embodiment short HR signal probes can also be used as additional features for the classification, which relies then not only on the movement signal. Thereby, the classification accuracy is increased without consuming much more power (depending on the probing frequency).

As it turns out that cycling is a particularly difficult activity to detect, especially when the movement sensor, e.g. the accelerometer, is worn on the wrist, another embodiment comprises two classifiers in a row, the first one focusing on detecting cycling, while the second one performs the classification (as defined before) for the detected non- cycling periods. This allows defining and using movement-based features that are more discriminative to that cycling activity.

As yet another embodiment a classification unit is used that distinguishes between several types of physical activities (e.g. walking, running, cross-training, etc.) for which a HR-based energy expenditure prediction is of advantage.

Generally, any HR sensor (sometimes also referred to as HR monitor) can be used in combination with a movement sensor (sometimes also referred to as activity monitor). An example of an integrated device which can be used for this purpose is an optical HR sensor containing a photo-plethysmography sensor (for HR detection) and a 3-axial accelerometer (for activity detection) as e.g. described in WO 2013/038296 Al. Such an integrated device may be implemented as a wrist worn device, which can be worn by a user at his arm. This technology allows monitoring HR unobtrusively and comfortably at any location on the body (for example at the wrist). This embodiment is ideal since users will have their fitness evaluation done just by wearing a watch or bracelet throughout the day and possibly night.

Generally, other kinds of HR sensors and movement sensors can be used in a device and method according to the present invention as well. For instance, as HR sensor an electrocardiograph ECG sensor may be used, and as movement sensor one or more motion sensors may be used. Other examples of HR sensors are photoplethysmograh PPG sensors, and camera PPG sensors and Global Positioning System GPS can be also used as movement sensor.

Activity monitoring solutions in the personal health space are oriented towards providing accurate information to the user regarding activity level and energy expenditure with respect to daily targets. The DirectLife service highly relies on the accuracy of the wearable activity monitor to generate user satisfaction during the activity intervention. The device is often used as a carrier of information on the daily achievements. Poor accuracy of the method to assess activity intensity and calories expenditure leads to user dissatisfaction and causes a dramatic decrease in the net promoter score (NPS) for the solution. New activity monitors, which may use or include the device or method according to the present invention have the potential to provide highly accurate energy expenditure estimates given the ability to record heart rate data together with body movement, in particular body acceleration.

However, given the (current) limited battery life and the (current) uncertainty related to the reliability/accuracy of heart rate information during certain activities of daily living, the ability of switching from one modality to the other at adequate times would allow the device to provide a continuously accurate and rewarding feedback to the activity monitor user.

In summary, the present invention proposes a new method and device to estimate the energy expenditure in daily life and while exercising, using either a movement- based method or an HR-based method and switching from one to the other at appropriate moments in time. The method includes several steps performed at an epoch level. A typical epoch length is one minute. The on-time of the HR sensor is preferably limited to the periods where the HR-based estimation is used, reducing therefore the power consumption and increasing the battery life. To conclude, for which activity energy expenditure can be estimated using the movement signal instead of the HR signal is not a generally accepted concept. The proposed device and method selects which body motion would benefit from HR data for the estimation of energy expenditure by looking at body movement characteristics related to motion rate and intensity.

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 word "comprising" does 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. A device for estimating the energy expenditure of a person, comprising:

a movement sensor (12) for generating a movement signal (13),

a heart rate sensor (14) for detecting the heart rate of the person and for generating a heart rate signal (15),

an estimation unit (16) for estimating the energy expenditure of the person from said movement signal (13) and/or said heart rate signal (15),

wherein the movement sensor is adapted to detect a physical activity of the person depending on a value of the movement signal, the device further comprising:

a control unit (18):

for switching on the heart rate sensor during a predetermined amount of time corresponding to the detection of physical activity of the person and such that the movement signal (13) and/or the heart rate signal (15) is used for estimating the energy expenditure, and

for switching off the heart rate sensor otherwise such that only the movement signal (13) is used for estimating the energy expenditure.

2. The device as claimed in claim 1, further comprising a classification unit (20) for extracting one or more features of the movement signal (13) and to determine if there is physical activity or no physical activity of the person by use of a classification based on the extracted one or more features of the movement signal (13).

3. The device as claimed in claim 2, wherein said classification unit (20) is configured to extract as features one or more of the peaks, the intervals between peaks, the rise times, the fall times, the number of peaks, the gradient, the intensity, the duration of peaks, the similitude or correlation between subsequent intervals, the periodicity, the spectral energy of the movement signal (13) and/or to use one or more of a logistic regression, a neural network, a support vector machine, Naive Bayes, a linear or quadratic discriminant.

4. The device as claimed in claim 2, wherein said classification unit (20) is configured to detect the type of physical activity, in particular to detect a cycling, walking, running or cross-training activity.

5. The device as claimed in claim 2, wherein said classification unit (20) comprises two or more classifiers, each using a predetermined set of features for detecting a predetermined type of activity.

6. The device as claimed in claim 1 or 2, wherein said control unit (18) is configured to switch said estimation unit (16) between a non-activity mode and an activity mode only if the same physical activity status of the person has been detected for a predetermined time period or if for a predetermined percentage of the last number of time windows the same physical activity status of the person has been detected.

7. The device as claimed in claim 1 or 2, wherein said control unit (18) is configured to control said estimation unit (16) to check the reliability of the heart rate signal (15) and, if physical activity has been detected, to use only the heart rate signal (15) for estimating the energy expenditure if the heart rate signal (15) is considered reliable and to use only the movement signal (13) for estimating the energy expenditure if the heart rate signal (15) is considered unreliable.

8. The device as claimed in claim 1 or 2, wherein the predetermined amount of time corresponds to the physical activity period of the person as detected by the movement sensor.

9. The device as claimed in claim 1 or 2, wherein the predetermined amount of time is a recovery period starting at the end of the physical activity period of the person and ending at when the heart rate of the person becomes stable.

10. The device as claimed in claim 1 or 2, wherein said movement sensor (12) comprises an acceleration sensor or a gyroscope or GPS and/or wherein said heart rate sensor (14) comprises a photoplethysmograph sensor or an ECG sensor or a heart sound sensor capable of detecting heart beats.

11. The device as claimed in claim 1 or 2, wherein said movement sensor (12), said heart rate sensor (14), said estimation unit (16) and/or said control unit (18) are built into a body worn device, in particular a wristband, wristwatch, smartphone or body belt.

12. A method for estimating the energy expenditure of a person, comprising: estimating the energy expenditure of the person from a movement signal (13) provided by a movement sensor (12) and/or a heart rate signal (15) provided by a heart rate sensor (14) for detecting the heart rate of the person,

wherein the method further comprising

- detecting physical activity of the person depending on a value of the movement signal, and

switching on the heart rate sensor during a predetermined amount of time

corresponding to the detection of physical activity of the person and such that the movement signal (13) and/or the heart rate signal (15) is used for estimating the energy expenditure, and - switching off the heart rate sensor otherwise such that only the movement signal (13) is used for estimating the energy expenditure.

13. Computer program comprising program code means for causing a computer to carry out the steps of the method as claimed in claim 12 when said computer program is carried out on the computer.