METHOD FOR CONTROLLING SCENE RECOGNITION IN REAL TIME TO DETERMINE A CONTEXT BY MEANS OF A WIRELESS DEVICE, AND ASSOCIATED DEVICE

DE602017086070T2Active Publication Date: 2024-11-13STMICROELECTRONICS (ROUSSET) SAS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
DE602017086070
Authority / Receiving Office
DE · DE
Patent Type
Patents
Current Assignee / Owner
Priority Date
2017-03-02
Filing Date
2017-08-18
Publication Date
2024-11-13
Estimated Expiration
2037-08-18

AI Technical Summary

Technical Problem

Current scene detection methods in wireless communication devices, such as smartphones and tablets, lack flexibility and result in excessive battery consumption due to fixed interval detection and reliance on GPS, which is energy-intensive.

Method used

Implement a method for real-time scene detection using a succession of detection steps with adjustable time intervals based on previous scene detection criteria, incorporating environmental sensors like accelerometers and audio sensors, and employing classification algorithms to filter low-confidence scenes and adapt detection frequency according to scene changes.

Benefits of technology

This approach enhances decision-making flexibility while reducing energy consumption by optimizing detection intervals based on previous scene data, thereby prolonging battery life and improving scene change detection accuracy.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Implementations and embodiments of the invention relate to the control of the real-time detection of a scene by a wireless communication device, for example a smart cellular mobile telephone ("smartphone") or a digital tablet, equipped with at least one environmental measurement sensor such as for example an accelerometer.

[0002] Some wireless communication devices, such as certain types of smartphones or tablets, are now capable of performing scene detection, which makes it possible to determine the environment in which the user of the phone or tablet is located. This can thus allow a third party, for example an advertiser or a cultural organization, to send relevant information related to the location of the user of the device.

[0003] So, for example, if the user is in a given tourist location, we can send them the addresses of restaurants near where they are. Similarly, we can also send them information about certain monuments located near where they are.

[0004] It is also possible to have the phone make certain adjustments automatically (for example, an audio sensor detects that the place is quiet and the phone automatically puts itself in vibrate mode).

[0005] Currently, the phone can detect the scene at fixed intervals. However, this does not provide any flexibility, which can lead to excessive power consumption, which is detrimental to the phone's battery life.

[0006] Another solution may be to use the phone's GPS application to be able to locate the user and therefore the detected scene in real time.

[0007] However, again, using the GPS function is detrimental to the phone's consumption.

[0008] The article by HIRAM GALEANA-ZAPIÉN ET AL: "Mobile Phone Middleware Architecture for Energy and Context Awareness in Location-Based Services", SENSORS, vol. 14, no. 12, December 10, 2014 (2014-12-10), pages 23673-23696, XP055386556, describes a middleware architecture tracking the spatio-temporal evolution of a wireless communication device equipped with a GPS sensor, and configured to determine a mobility state of the device and adapt on the fly the sampling rate of the GPS sensor according to the detected mobility state in order to reduce the energy consumption of the device.

[0009] The paper by ANDRE C SANTOS ET AL: "A Domain-Specific Language for the Specification of Adaptable Context Inference", EMBEDDED AND UBIQUITOUS COMPUTING (EUC), 2011 IFIP 9TH INTERNATIONAL CONFERENCE ON, IEEE, October 24, 2011 (2011-10-24), pages 268-273, XP032074987, ISBN: 978-1-4577-1822-9, describes a "domain-specific language" and middleware infrastructure to support specifications, deployments, and maintenance of adaptive scene inference processes

[0010] The article by MEEYEON LEE ET AL: "Analysis of Characteristics of Power Consumption for Context-Aware Mobile Applications", INFORMATION, vol. 5, no. 4, November 14, 2014 (2014-11-14), pages 612-621, XP055386286, describes an analysis of the power consumption characteristics for context-aware mobile applications, which allows defining two main factors that significantly influence power consumption.

[0011] ROBERT GUINNESS's article: "Beyond Where to How: A Machine Learning Approach for Sensing Mobility Contexts Using Smartphone Sensors", SENSORS, vol. 15, no. 5, April 28, 2015 (2015-04-28), pages 9962-9985, XP055386832, presents research results on the use of smartphone sensors (typically GPS and accelerometers), geospatial information (points of interest such as bus stops and train stations), and automated search to measure mobility contexts.

[0012] There is therefore a need to improve the decision-making regarding scene detection by a wireless communication device, for example a cellular mobile phone or a tablet, which allows greater flexibility and has a reduced impact on power consumption.

[0013] And, according to one mode of implementation and embodiment, we will use at least one previously detected scene, and more generally several previous scenes to improve decision-making and better follow scene changes while saving energy in the absence of scene changes.

[0014] Various aspects, embodiments and methods of implementing the invention are defined by the appended claims.

[0015] Thus, according to one aspect, a method is proposed for controlling the real-time detection of at least one scene by a wireless communication device, comprising a succession of scene detection steps mutually spaced by time intervals and an adjustment of the time interval separating a current scene detection step from the previous scene detection step as a function of at least one adjustment criterion linked to at least one previous scene actually detected.

[0016] This previous scene actually detected is generally the scene actually preceding the current scene, but could be another previous scene, for example in the case where the scene actually preceding the current scene would not be usable for various reasons or would present an anomaly for example or would be detected with low confidence.

[0017] Scene detection refers to the discrimination of the scene in which the wireless communication device is located. Several known solutions exist for detecting (discriminating) a scene. These solutions use, for example, one or more dedicated sensors generally associated with a specific algorithm.

[0018] These sensors can be, especially in applications requiring very little current consumption, environmental measurement sensors, for example accelerometers and / or gyroscopes and / or magnetometers and / or microphones. This is a multimodal approach. And in this case, an example of a specific algorithm can be an algorithm implementing a binary decision tree based on descriptors or attributes resulting from specific processing (for example, filtering) on ​​the raw data from the sensors. These descriptors can be, for example, averages, energy values, variances, etc.

[0019] By environmental measurement sensor is meant in particular any type of sensor capable of providing information on the environment in which the wireless communication device is located, including for example spatiotemporal characteristics of the environment of the device, such as for example the temporally fixed or non-temporal nature of the environment, the speed of evolution of the spatiotemporal modification of the environment (from the detection of the movement of the device) and / or the sound and / or spatial and / or visual characteristics of this environment, such as for example the noise level of the environment and / or the altitude and / or the brightness level of the environment (from for example sensors such as barometers, proximity sensors, optical sensors, etc.).

[0020] The sensor(s) used for scene detection may be different from the environmental sensor(s).

[0021] This being particularly true in a context where the device is constantly powered (Always-On) and where battery life is an important criterion, at least one of the environmental sensors may also be a sensor used for scene detection. This may be the case, for example, but not exclusively, for an accelerometer which may be used both as an environmental sensor to give an indication of the intensity of the movement of the device and as a sensor contributing to scene detection.

[0022] Several adjustment criteria are possible to decide whether to modify the time interval separating a current scene detection step from the previous scene detection step. For example, one can detect the presence of a particular signal coming from a particular sensor signifying a particular modification of the scene environment requiring either more frequent scene detection or allowing less frequent scene detection. This particular sensor, for example an accelerometer or a gyroscope, can be an activity detector whose output gives an indication of the speed of change of the environment.

[0023] This being according to one mode of implementation, when each scene actually detected belongs to a set, or corpus, of reference scenes, each reference scene is associated with a reference duration and said adjustment criterion linked to said at least one previous scene actually detected is a duration which is obtained from at least one of the reference durations.

[0024] According to one embodiment, said adjustment may be a function of at least one adjustment criterion linked to K previous scenes actually detected, K being an integer greater than 1.

[0025] Using K previous scenes actually detected is advantageous, especially in the case where the previous scene is detected with low confidence. Indeed, in this case, decision-making on these K scenes allows, in a way, filtering the scene(s) detected with low confidence.

[0026] Thus, for example, we can consider as the scene actually detected the one which is mostly detected among the K previous scenes and adopt as the adjustment criterion that of this majority scene.

[0027] The value of K can also vary depending on the nature of the transition between two successive previous scenes actually detected.

[0028] So if, for example, the transition reveals that we are moving from a quasi-static scene to a rapidly evolving scene, we can increase the value of K.

[0029] According to an embodiment in which the device is equipped with at least one environmental measurement sensor, each scene detection step is advantageously carried out from measurement values ​​delivered by said at least one environmental measurement sensor.

[0030] Said at least one environmental measurement sensor may comprise at least one accelerometer and / or at least one audio sensor.

[0031] According to an embodiment in which the wireless communication device is equipped with several environmental measurement sensors, each scene detection step is advantageously carried out from measurement values ​​delivered by said sensors.

[0032] Said environmental measurement sensors can be chosen from the group formed by an accelerometer, a gyroscope, a magnetometer, an audio sensor, a barometer, a proximity sensor, an optical sensor.

[0033] In other words, one or more accelerometers can be used, or an accelerometer in combination with a gyroscope, possibly in combination with a magnetometer, or several audio sensors, or one or more audio sensors in combination with one or more accelerometers, or a gyroscope, or a magnetometer.

[0034] Each scene detection step may advantageously comprise an implementation of a classification algorithm or classifier, and preferably an implementation of a meta classification algorithm or meta classifier which for example performs a majority vote on the last five measurements in order to avoid aberrant errors, and which may preferably be reset after each modification of a time interval.

[0035] According to another aspect, a wireless communication device is proposed, comprising detection means configured to detect at least one scene in real time, and control means configured to successively activate the detection means so as to implement a succession of scene detection steps mutually spaced by time intervals and to carry out an adjustment of the time interval separating a current scene detection step from the previous scene detection step as a function of at least one adjustment criterion linked to at least one previous scene actually detected.

[0036] According to one embodiment, the apparatus comprises memory means configured to store identifiers respectively associated with a set of reference scenes, each scene actually detected belonging to said set of reference scenes, each reference scene being associated with a reference duration, and said adjustment criterion linked to said at least one previous scene actually detected is a duration obtained from at least one of the reference durations.

[0037] According to one embodiment, the control means are configured to carry out said adjustment as a function of at least one adjustment criterion linked to K previous scenes actually detected, K being an integer greater than 1.

[0038] According to one embodiment, the control means are configured to vary the value of K depending on the nature of the transition between two successive previous scenes actually detected.

[0039] According to an embodiment in which the apparatus is equipped with at least one environmental measurement sensor, the detection means are configured to carry out each scene detection step from measurement values ​​delivered by said at least one environmental measurement sensor.

[0040] According to one embodiment, said at least one environmental measurement sensor comprises at least one accelerometer.

[0041] According to one embodiment, said at least one environmental measurement sensor comprises at least one audio sensor.

[0042] According to an embodiment in which the device is equipped with several environmental measurement sensors, the detection means are configured to carry out each scene detection step from measurement values ​​delivered by said sensors.

[0043] Said environmental measurement sensors are for example chosen from the group formed by an accelerometer, a gyroscope, a magnetometer, an audio sensor, a barometer, a proximity sensor, an optical sensor.

[0044] According to one embodiment, the scene detection means are configured to implement a classification algorithm, for example a meta classification algorithm.

[0045] According to one embodiment, the control means are configured to perform a reset of the classification meta-algorithm after each modification of a time interval.

[0046] The device can be, for example, a cellular mobile phone or a digital tablet.

[0047] Other advantages and characteristics of the invention will appear on examining the detailed description and modes of implementation and embodiment, which are in no way limiting, and the appended drawings in which: THE figures 1 to 3 schematically illustrate different modes of implementation and embodiment of the invention.

[0048] On the figure 1 , the APP reference designates an electronic device, for example a wireless communication device equipped with an ANT antenna. This device can be a cellular mobile phone such as a smartphone or a digital tablet.

[0049] The APP device here includes several environmental measurement sensors CPT1-CPT5, without this number being limiting.

[0050] For information, the sensors CPT1, CPT2, CPT3, CPT4, CPT5 can be chosen from the group formed by an accelerometer, a gyroscope, a magnetometer, an audio sensor such as a microphone, a barometer, a proximity sensor, an optical sensor.

[0051] Of course, the device may be equipped with several accelerometers and / or several gyroscopes and / or several magnetometers and / or several audio sensors and / or a barometer, and / or one or more proximity sensors, and / or one or more optical sensors.

[0052] At least one and generally at least some of these environmental measurement sensors, can, in particular in a multimodal approach, form in combination with a classic ALC discrimination algorithm, for example of the binary decision tree type, intended to work for example on filtered raw data from these sensors, MDET detection means configured to detect a scene. These MDET means can thus for example detect whether the APP device is located in a particular environment (restaurant, moving vehicle, etc.).

[0053] Alternatively, the APP device could be equipped with a specific sensor, for example the CPT1 sensor, different from the environmental sensors mentioned above, for example a micro camera, capable of detecting a scene from the environment of the device.

[0054] This being in a context where the device is constantly powered (Always-On) and where battery life is an important criterion, it is preferable to use as sensors contributing to detecting a scene from the device's environment, one or more environmental type sensors mentioned above.

[0055] It is now assumed as a non-limiting example that all environmental sensors CPT1-CPT5 contribute to the detection of the scene and provide data at measurement times to the ALC discrimination algorithm to enable the scene to be detected.

[0056] The discrimination algorithm implemented in the MDET scene detection means may be a classification algorithm or classifier which is an algorithm well known to those skilled in the art. In this regard, those skilled in the art may refer for all useful purposes, for example to Jason Brownlee's book "Master Learning Algorithms, discover how they work and implement them from scratch", 2016, or to André C. Santos et al.'s article "Context Inference for Mobile Applications in the UPCASE Project" Second International Conference, Mobilware 2009, Berlin, Germany, April 28-29, 2009 Proceedings Pages 352-365, or to Robert E. Guinness' article "Beyond Where to How: A Machine Learning Approach for Sensing Mobility Contexts Using Smartphone Sensors", Sensors 2015, 15, 9962-9985.

[0057] In a particularly simple implementation, a decision tree that has been trained on a database of environmental sensor measurements can be used as a classifier. Such a decision tree is particularly simple to implement and requires only a few bytes of memory and a working frequency of less than 0.01 MHz.

[0058] That being said, it is preferable to use a meta classification algorithm (or meta classifier) ​​also well known to those skilled in the art which will, for example, carry out a majority vote on the last five measurements provided by the sensors, for example, in order to avoid aberrant errors.

[0059] While we can use at least one accelerometer or even several accelerometers, we will have more precision if we also use a gyroscope or even a magnetometer.

[0060] That being said, it can be particularly interesting to use audio sensors which are useful descriptors of the environment. Indeed, if the device is not in motion, then the audio sensor can be interesting to detect the nature of this environment. Of course, depending on the applications, one can use either environmental sensors such as accelerometers or even gyroscopes or magnetometers, or audio sensors or a combination of these two types of sensors. That being said, with a view to the simplicity of the classifier, one can advantageously choose not to retain a multimodal approach, that is to say not to use a combination of these two types of sensors.

[0061] In addition to these sensors, the APP device includes a BLC block capable of cooperating with the CPTi sensors and the MDET detection means to process the detected scene and transmit the information via the ANT antenna of the device.

[0062] The apparatus also comprises MCTRL control means configured to successively activate the MDET detection means so as to implement a succession of scene detection steps mutually spaced by time intervals and to carry out an adjustment of the time interval separating a current scene detection step from the previous scene detection step as a function of at least one adjustment criterion linked to at least the previous scene actually detected.

[0063] In the example described here, the APP device also comprises memory means MM configured to store identifiers respectively associated with a set, or corpus, of reference scenes SR, each reference scene being associated with a reference duration.

[0064] These reference scenes which were obtained by the detection means during a learning phase may be, for example, but not limited to, “BUS”, “OFFICE”, “RESTAURANT”, “TRAIN” scenes representative of the environment in which the device is located.

[0065] Of course the reference durations associated with these scenes vary depending on the nature of the reference scene.

[0066] Thus the reference duration associated with a stable scene, such as the “OFFICE” scene, will be longer than the reference duration of a scene likely to change rapidly, such as the “TRAIN” or “BUS” scene.

[0067] In fact, as we will see below, this reference duration will determine the time interval between two successive scene detections.

[0068] Indeed, in the embodiment described here, each scene actually detected belongs to said set of reference scenes, and said adjustment criterion linked to a previous scene actually detected is the corresponding reference duration.

[0069] These different BLC, ALC and MCTRL means are for example implemented by software modules within the PR processor of the APP device.

[0070] We now refer more particularly to the figure 3 to describe a mode of implementation of the method according to the invention.

[0071] This is implemented for example by a state machine. After an initialization phase 30, a default initial value is set in a step 31 for the time interval between two times of scene detection by the MDET detection means. This default initial value may be of the order of a few seconds, for example 2, 4 or 6 seconds.

[0072] Then, at the end of each time interval (step 32) we acquire (step 33) the measurement values ​​provided by the environmental sensors and we detect the current scene using the classifier or the metaclassifier.

[0073] And, in a step 34, it is determined to which reference scene (SR) the scene which has just been detected corresponds.

[0074] It would then be possible to set the value of the time interval separating the scene detection that has just taken place from the next scene detection to the reference duration associated with the corresponding reference scene.

[0075] However, in the case where we have many reference scenes, it may be simpler to define subsets, for example three subsets, of reference scenes corresponding for example to classes of environmental variations.

[0076] Thus, for example, we can define a class C1 corresponding to a scene environment considered to be stable, containing for example the scenes "OFFICE" and "RESTAURANT", a class C2 corresponding to a scene environment considered to be evolving rapidly, containing for example the scene "TRAIN", and a class C3 corresponding to a scene environment considered to be intermediate, that is to say evolving less rapidly, containing for example the scene "BUS".

[0077] We can then take as the reference duration associated with a class, for example the average of the reference durations of each scene in the class or the minimum reference duration or even the maximum reference duration, without these examples being limiting.

[0078] Depending on the class obtained at the end of step 34, we will define three different values ​​for the time interval separating the instant of detection of the current scene from the instant of detection of the next scene.

[0079] Thus, in step 35 we can set a time interval value of the order of 10 seconds while we can set this value to 6 seconds in step 36 and to two seconds in step 37.

[0080] In other words, the faster the movement, the more we will reduce the time interval between two successive scene detections.

[0081] In step 38, it is checked whether this time interval value has been changed from the previously stored value.

[0082] If so, then in step 39, we reset the classifier before returning to step 32.

[0083] Otherwise, no classifier reset is performed.

[0084] If we consider in the newly executed step 33 that the detected scene is the current scene, then we can clearly see that the time interval taken into account in step 32 has been adjusted according to an adjustment criterion linked to the previous scene detection carried out in the previous step 33.

[0085] While in the implementation mode just described the time interval between two successive scene detections is adjusted from the previously detected scene, it would be possible, as illustrated in the figure 3 , to determine in steps 34 to 37, the value of the time interval which will separate the detection of scene I-1 carried out in step 33 from the detection of the following scene I which will be carried out in the following step 33, from K previous detected scenes, here the K scenes IK-2 to 1-2.

[0086] In this regard, several possibilities are offered.

[0087] Two non-exhaustive examples are mentioned below.

[0088] We can calculate the average of the reference durations associated with the K corresponding reference scenes.

[0089] We can also take from among the K previous scenes the one which is mostly present and then determine the corresponding class C1, C2 or C3.

[0090] Of course, the previous K scenes may or may not be consecutive for at least some of them.

[0091] Furthermore, the value of K can vary depending on the nature of the transition between two successive previous scenes actually detected.

[0092] So if for example the transition reveals that we are moving from a quasi-static scene to a rapidly evolving scene, we can increase the value of K to adjust the time interval which will be used for the detection of subsequent scenes.

Claims

1. A method for controlling the real-time detection of at least one scene by a wireless communication device, comprising a succession of scene detection steps (33) mutually spaced by time intervals and an adjustment (34-37) of the time interval separating a current scene detection step (33) from the preceding scene detection step (33) depending on at least one adjustment criterion linked to at least one preceding actually detected scene, each actually detected scene belonging to a set of reference scenes (SR), each reference scene being associated with a reference duration, and said adjustment criterion linked to said at least one preceding actually detected scene is the reference duration corresponding to the preceding actually detected scene.

2. The method for controlling the real-time detection of at least one scene by a wireless communication device, comprising a succession of scene detection steps (33) mutually spaced by time intervals and an adjustment (34-37) of the time interval separating a current scene detection step (33) from the preceding scene detection step (33) depending on at least one adjustment criterion linked to K preceding actually detected scenes (I-K-2 to 1-2), K being an integer greater than 1, each actually detected scene belonging to a set of reference scenes (SR), each reference scene being associated with a reference duration, and said adjustment criterion linked to said at least one preceding actually detected scene is a duration obtained from the at least one K reference durations corresponding to the K preceding actually detected scenes.

3. The method according to claim 2, wherein the adjustment criterion linked to said at least one preceding actually detected scene is the average of the K reference durations corresponding to the K preceding actually detected scenes.

4. The method according to claim 2, wherein the adjustment criterion linked to said at least one preceding actually detected scene is the reference duration corresponding to the mostly detected scene from the K preceding actually detected scenes.

5. The method according to one of claims 2 to 4, wherein the value of K varies according to the nature of the transition between two successive preceding actually detected scenes.

6. The method according to one of the preceding claims, wherein the device is equipped with at least one environmental measurement sensor (CPTi) and each scene detection step (33) is carried out on the basis of measurement values delivered by said at least one environmental measurement sensor.

7. The method according to claim 6, wherein said at least one environmental measurement sensor (CPTi) comprises at least one accelerometer.

8. The method according to one of claims 6 or 7, wherein said at least one environmental measurement sensor (CPTi) comprises at least one audio sensor.

9. The method according to one of the preceding claims, wherein the wireless communication device is equipped with a plurality of environmental measurement sensors (CPTi), and each scene detection step is carried out on the basis of measurement values delivered by said sensors.

10. The method according to claim 9, wherein said environmental measurement sensors (CPTi) are chosen from the group formed by an accelerometer, a gyroscope, a magnetometer, an audio sensor, a barometer, a proximity sensor and an optical sensor.

11. The method according to one of the preceding claims, wherein each scene detection step comprises an implementation of a classification algorithm (ALC).

12. The method according to claim 11, wherein each scene detection step comprises an implementation of a classification meta - algorithm.

13. The method according to claim 12, comprising a re-initialisation (39) of the classification meta-algorithm after each modification of a time interval.

14. A wireless communication device, comprising detection means (MDET) configured to detect in real time at least one scene, and control means (MCTRL) configured to successively activate the detection means in such a way as to implement a succession of scene detection steps mutually spaced at time intervals and for carrying out an adjustment of the time interval separating a current scene detection step from the preceding scene detection step, depending on at least one adjustment criterion linked to at least one preceding actually detected scene, the device further comprising memory means (MM) configured to store identifiers that are respectively associated with a set of reference scenes (SR), each actually detected scene belonging to said set of reference scenes, each reference scene being associated with a reference duration, and said adjustment criterion linked to said at least one preceding actually detected scene is the reference duration corresponding to the preceding actually detected scene.

15. The wireless communication device, comprising detection means (MDET) configured to detect in real time at least one scene, and control means (MCTRL) configured to successively activate the detection means in such a way as to implement a succession of scene detection steps mutually spaced at time intervals and for carrying out an adjustment of the time interval separating a current scene detection step from the preceding scene detection step depending on at least one adjustment criterion linked to K preceding actually detected scenes (I-K-2 to 1-2), K being an integer greater than 1, the device further comprising memory means (MM) configured to store identifiers that are respectively associated with a set of reference scenes (SR), each actually detected scene belonging to said set of reference scenes, each reference scene being associated with a reference duration, and said adjustment criterion linked to said at least one preceding actually detected scene is a duration obtained from at least one of the K reference durations corresponding to the K preceding actually detected scenes.

16. The device according to claim 15, wherein the adjustment criterion linked to said at least one preceding actually detected scene is the average of the K reference durations corresponding to the K preceding actually detected scenes.

17. The device according to claim 15, wherein the adjustment criterion linked to said at least one preceding actually detected scene is the reference duration corresponding to the mostly detected scene from the K preceding actually detected scenes.

18. The device according to one of claims 15 to 17, wherein the control means (MCTRL) are configured to vary the value of K as a function of the nature of the transition between two successive preceding actually detected scenes.

19. The device according to one of claims 14 to 18, equipped with at least one environmental measurement sensor (CPTi) and wherein the detection means are configured to carry out each scene detection step on the basis of measurement values delivered by said at least one environmental measurement sensor.

20. The device according to claim 19, wherein said at least one environmental measurement sensor (CPTi) comprises at least one accelerometer.

21. The device according to one of claims 19 or 20, wherein said at least one environmental measurement sensor (CPTi) comprises at least one audio sensor.

22. The device according to one of claims 14 to 21, equipped with a plurality of environmental measurement sensors (CPTi), and wherein the detection means (MDET) are configured to carry out each scene detection step on the basis of measurement values delivered by said sensors.

23. The device according to claim 22, wherein said environmental measurement sensors (CPTi) are chosen from the group formed by an accelerometer, a gyroscope, a magnetometer, an audio sensor, a barometer, a proximity sensor and an optical sensor.

24. The device according to one of claims 14 to 23, wherein the scene detection means (MDET) are configured to implement a classification algorithm (ALC).

25. The device according to claim 24, wherein the scene detection means (MDET) are configured to implement a classification meta - algorithm.

26. The device according to claim 25, wherein the control means (MCTRL) are configured to carry out a re-initialisation of the classification meta-algorithm after each modification of a time interval.

27. The device according to one of claims 14 to 26, forming a mobile cellular telephone or a digital tablet.