DRIVER CONSCIOUSNESS DETECTION

Abstract

Method for detecting a state of consciousness of an individual in an environment, implemented by a detection device, the device being connected to a camera positioned in the environment so as to allow the acquisition of images on which the individual is visible, the method comprising the following steps: - obtaining postural data of the individual, said postural data being extracted from at least one image acquired by the camera and depending on a posture of the individual in the environment, - determining, as a function of said postural data, a state of consciousness of the individual from among at least one conscious state and one unconscious state of the individual. Abstract figure: Figure 5

Description

Title of the invention: DETECTION OF LOSS OF DRIVER CONSCIOUSNESS Technical field

[0001] The present disclosure relates to a device for detecting a state of consciousness of an individual. The invention may in particular be applicable to the detection of the state of consciousness of a vehicle driver. Prior art

[0002] The loss of consciousness of an individual at the controls of a vehicle, for example in the event of discomfort or drowsiness, results in a loss of control of the vehicle which can lead to endangerment of the individual and those around them. The development of intelligent transport systems (or ITS) aims in particular to reduce these risks by developing autonomy of the vehicle with respect to its driver, for example thanks to driving assistance functions. The implementation of these assistance functions can be relevant when it is detected that an individual is no longer able to control his vehicle, for example in the event of loss of consciousness of the individual. Upon detection of such a state, assistance functions such as braking, parking or even emergency call can be triggered.

[0003] Most of the methods allowing such detection of loss of consciousness of an individual at the controls of a vehicle rely on sensors recording the biometric, biological and / or physiological data of the individual (for example, data from encephalograms, heart rate monitors, etc.). Such detection methods then have disadvantages, because they are intrusive and require specific sensors that are difficult to integrate into a vehicle, particularly in the context of an individual at the controls of a vehicle such as a car, a heavy goods vehicle, a tram or even an airplane. Other detection methods rely on facial analysis of the individual (for example, detection of the degree to which the driver's eyes are open). However, these methods lack robustness when the individual changes position or the camera allowing facial analysis changes viewing angle.In particular, in many cases, a loss of consciousness of the individual may be accompanied by a tilting of his head so that facial analysis of the individual is no longer possible.

[0004] More generally, most methods for detecting a state of consciousness of an individual at the controls of a vehicle lack robustness, reliability and / or precision, in particular either because they are unsuitable or too intrusive to be integrated into the environment of the vehicle, or because they have little robustness. in the face of the variability of the vehicle environment (for example, the plurality of possible viewing angles in a vehicle, the plurality of individuals' physiognomies or even the plurality of postures in which the individual may find themselves can be bias factors). Summary

[0005] The present disclosure improves the situation.

[0006] A method for detecting a state of consciousness of an individual in an environment is proposed, implemented by a detection device, the device being connected to a camera positioned in the environment so as to allow the acquisition of images on which the individual is visible, the method comprising the following steps: - obtain data, called postural, relating to the posture of the individual in the environment, said data being extracted from at least one image acquired by the camera, - determine, based on said postural data, a state of consciousness of the individual from among at least one conscious state and one unconscious state of the individual.

[0007] Advantageously, the proposed method makes it possible to detect a state of consciousness of an individual at the controls of a vehicle in a non-intrusive manner by relying on the images of a camera integrated into the environment, for example in the passenger compartment of a vehicle. The method also makes it possible to rely on images acquired by the camera in the context of other functionalities integrated into the vehicle system, for example, for driving assistance functionalities for autonomous or semi-autonomous vehicles.

[0008] According to another aspect, there is provided a device for detecting a state of consciousness of an individual in an environment, said device being connected to a camera positioned in the environment so as to allow the acquisition of images on which the individual is visible, the device comprising at least one computer configured to: - obtain data, called postural data of the individual, relating to a posture of the individual in the environment, said data being extracted from at least one image acquired by the camera, - determine, based on said postural data, a state of consciousness of the individual from among at least one conscious state and one unconscious state of the individual.

[0009] According to another aspect, there is provided a computer program comprising instructions for implementing the proposed method when this program is executed by a processor.

[0010] According to another aspect, there is provided a non-transitory computer-readable recording medium having recorded thereon a program for implementing of the proposed method when this program is executed by a processor.

[0011] The features set out in the following paragraphs may, optionally, be implemented, independently of one another or in combination with one another:

[0012] In one embodiment, the postural data comprises the positions in the image of a finite set of characteristic points of the individual.

[0013] The characteristic points may correspond to joints of the individual (for example at the shoulders, elbows, wrists of the individual) and / or points of the face of the individual (for example at the nose, ears, chin of the individual).

[0014] In one embodiment, the method further comprises: - the calculation, from the positions of the characteristic points, of the values ​​of postural parameters associated with the posture of the individual in the environment, said postural parameters comprising at least: * lengths of segments formed by two distinct characteristic points among the finite set of characteristic points, * angles formed by two segments having a common characteristic point.

[0015] In one embodiment, the values ​​of the postural parameters are calculated from standardized positions of the characteristic points, said positions being standardized with respect to at least one reference quantity of the individual measured on the acquired image.

[0016] Advantageously, the proposed method then makes it possible to put into perspective the detection of a state of consciousness of an individual in relation to the different postures of the individual as well as in relation to the environment in which such detection is implemented. In particular, the normalization of the postural parameters makes it possible to neglect the impact of the camera and its adjustment parameters (for example, its positioning in the environment) in determining the state of consciousness of the individual. The proposed detection method is then more robust and usable in varied environments and allows results of a certain invariance in relation to changes in positioning of the camera and a fortiori, of the individual, as captured on the acquired images.

[0017] The reference quantity of the individual can then advantageously be chosen so that the values ​​of the postural parameters calculated from the standardized positions have substantially an invariance to the change in positioning of the camera in the environment or to the change in the state of consciousness of the individual. For example, the reference quantity can be chosen so that it varies little when the individual is in a posture associated with a conscious state or in a posture associated with an unconscious state. The reference quantity can also be chosen so that a position normalized with respect to such a reference quantity does not depend significantly or only slightly (relatively, with respect to the other normalized positions) on the location and / or orientation of the camera in the environment (for example in the passenger compartment of the vehicle). The reference quantity may for example be a distance between two chosen reference points (for example, corresponding to an inter-shoulder distance of the individual on the acquired image).

[0018] In one embodiment, reference points respectively associated with reference positions are selected from the finite set of characteristic points, and the postural parameter values ​​are normalized by transforming, for each characteristic point of the finite set of characteristic points, the associated position into a normalized position expressed by the following normalized coordinates:

[0019]

[0020] JT = / = Or : yy» y^ - x* and y* are the normalized coordinates of the normalized position associated with the characteristic point, - x and y are the coordinates of the position associated with the characteristic point, - xR is a first coordinate of a first predefined reference point, - xL is a first coordinate of a second predefined reference point, - yD is a second coordinate of a third predefined reference point, and - yG is a second coordinate of a fourth predefined reference point.

[0021] Advantageously, the values ​​of postural parameters are normalized by a normalization (or transformation) of the positions of the characteristic points forming the postural data, so that the normalized values ​​of postural parameters are calculated from the normalized positions.

[0022] Thus; reference points can be advantageously chosen so that a reference quantity (for example, a distance separating these reference points) can be obtained from the reference positions of said reference points to calculate the normalized positions. In other words, the choice of a reference quantity depends on the choice of the reference points.

[0023] In one embodiment, the state of consciousness of the individual is determined by a classifier from the values ​​of the postural parameters, said classifier being previously trained on training data comprising at least training values ​​of the postural parameters.

[0024] In one embodiment, the method further comprises: - obtaining facial data of the individual including one of at least less: an orientation of the individual's head, a degree of opening of the individual's eyes, a degree of movement of a face of the individual, and the individual's state of consciousness is further determined based on said facial data.

[0025] In one embodiment, the method further comprises: - the emission, in the event of the detected individual being unconscious, of an alert signal to a control unit.

[0026] Advantageously, such a step allows the proposed detection device to alert the individual or another individual of the detected unconscious state, or to trigger an emergency action by means of an alert signal sent to a control unit.

[0027] Such a control unit may refer to any processing unit or sub-unit allowing the implementation of an action triggered by the result of the detection method (in this case a detected unconscious state). Such a control unit may in particular be remote from the detection device or even from the vehicle system (for example, a mobile telephone or any other remote equipment), the alert signal being for example a communication setup for an emergency call to another individual or to an emergency service. The control unit may also be integrated into the vehicle system, corresponding for example to the autonomous control unit of the vehicle, to a processing unit linked to an assistance function of the vehicle or to a processing unit linked to the visual and / or audio interface of the vehicle.The alert signal can then correspond to an emergency action enabling autonomous driving of the vehicle to be activated, triggering an emergency parking or braking maneuver, or even issuing a visual or audible alert to the individual controlling the vehicle or its passengers. Brief description of the drawings

[0028] Other characteristics, details and advantages will appear on reading the detailed description below, and on analyzing the attached drawings, in which:

[0029] [Fig-1] [Fig.l] shows a detection device according to one embodiment.

[0030] [Fig.2] [Fig.2] shows a detection device according to another mode of rea lization.

[0031] [Fig.3] [Fig.3] shows implemented steps of a detection method according to an embodiment.

[0032] [Fig.4] [Fig.4] shows postural data of an individual according to a mode of realization.

[0033] [Fig.5] [Fig.5] shows postural data of an individual in another mode of realization.

[0034] [Fig.6] [Fig.6] shows postural parameters according to one embodiment.

[0035] [Fig.7] [Fig.7] shows coordinates of postural data according to a mode of realization. Description of the embodiments

[0036] Reference is now made to [Fig.l]. [Fig.l] represents a device 1 for detecting a state of consciousness of an individual. This device 1 may for example be adapted to detect whether the individual is in a conscious or unconscious state in a given environment, which may be a vehicle. For example, the vehicle may be a partially autonomous vehicle and the detection device 1 may be adapted to detect a potential unconscious or drowsy state of the driver at the controls of the vehicle. In the remainder of the description, the environment considered for the implementation of the detection method is a motor vehicle. Alternatively, the environment considered may be any other means of road or air transport controlled (at least partially) by an individual, such as a tram, train or even an airplane.

[0037] The device 1 comprises a camera 10 that can be oriented, or possibly having a fixed orientation, so that the camera can acquire images including at least a portion of the environment, for example a vehicle interior. In particular, the individual at the controls of the vehicle is visible in the acquired images. For example, the camera 10 can be installed in the vehicle interior and oriented so as to acquire an image in which the driver's head and torso are visible, as is the case in the example images of FIGS. 4, 5 and 6. The camera 10 can for example be positioned in or around the central interior rearview mirror of the vehicle, in the central dashboard or behind the steering wheel. [Fig. 4] is an example of an image acquired by a camera 10 positioned at the central interior rearview mirror.Figures 5 and 6 are examples of images acquired by a camera 10 positioned at the steering wheel, facing the driver. The orientation of the camera 10 is considered known, so that the location of the elements of interest (for example, of the individual) on the images acquired by the camera 10 is known. In other words, the pixels occupied by the individual in each image acquired at each given instant are identified. These pixels are defined according to a two-dimensional (x,y) reference frame as shown in [Fig.7], called the image reference frame.

[0038] The device 1 further comprises at least one computer 11, comprising at least one processor, configured to implement the steps described below to process the image(s) acquired by the camera 10 in order to determine a state of consciousness of the individual. In one embodiment, the device 1 comprises a single computer 11, which can be embedded in the environment (for example the vehicle), which performs all of the steps described below. The computer 11 can be integrated with the camera 10 or be a separate computer, for example a computer of the vehicle's on-board computer, capable of receiving images transmitted by the camera 10.

[0039] Alternatively, and as shown schematically in [Fig.2], the device may comprise at least a first computer 11 on board the vehicle, performing certain steps, and a second remote computer 12 capable of performing other steps. Alternatively, several remote computers may be connected to a computer 11 on board the vehicle and may perform several steps.

[0040] The device 1 further comprises at least one memory 13 storing code instructions executed by the computer or a computer to be able to implement steps described below. Advantageously, the memory 13 can also store data obtained and / or determined by the computer from the images acquired by the camera 10. The memory 13 can also store predefined or preconfigured values ​​used for implementing the method below.

[0041] Reference is now made to [Fig. 3]. [Fig. 3] diagrams a succession of steps implemented by a device 1 as represented in [Fig. 1] for detecting a state of consciousness of an individual at the controls of a vehicle.

[0042] In a step 300, the device 1 collects at least one image acquired by the camera 10. The image is acquired according to a given field of view depending on the positioning of the camera 10. The acquired image may be associated with a timestamp. In particular, the field of view of the acquired image comprises at least a portion of the individual, for example a face, a bust, arms, a trunk of the individual. In a preferred embodiment, an upper part of the body of the individual is visible in the acquired image, such an upper part comprising the head, the neck, the shoulders, the arms and a part of the torso of the individual, as shown in FIGS. 5, 6. In one embodiment, other parts of the body of the individual may be visible, such as the forearms, as shown in FIGS. 4 and 7.

[0043] The characteristics of the acquired image may depend on the adjustment parameters of the camera 10 (for example focal length, recording speed, exposure, etc.) as well as the environment (for example brightness in and outside the vehicle passenger compartment). The image acquired in step 300 may be an image converted to black and white or an infrared image (for example an image in the near infrared spectrum or in English, “Near-Infrared” or nIR). In particular, the image collected by the device 1 may be preprocessed during step 300, for example by the device 1.

[0044] At a step 310, the device 1 obtains postural data of the individual associated with the acquired image. In one embodiment, when several acquired images (for example, successive) are collected at step 300, step 310 is implemented for each acquired image. The postural data obtained may correspond to a re digital presentation of the key parts of the individual's body, called a "skeleton", in the form of a finite set of characteristic points, which can be designated by points of interest, or joints. These characteristic points are notably associated with respective positions in the acquired image and expressed by coordinates (x,y) in the image reference frame. This digital representation can for example be obtained by an algorithm for detecting the pose of the individual on the image or a skeletonization algorithm implemented during step 310. The person skilled in the art may refer to the publication by Zhe Cao, Gines Hidalgo, Tomas Simon, Yaser Sheikh, "OpenPose: Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields", arXiv:1812.08008v2 - May 30, 2019, for an example of implementation of such an algorithm.The postural data can then be obtained in step 310 by implementing an algorithm for detecting the individual on the acquired image and detecting all the characteristic points associated with a posture of the individual on the acquired image, using a neural network. The digital representation of the skeleton of the individual can comprise only characteristic points located at positions on the acquired image, as shown in [Fig. 7], or characteristic points connected to each other by segments, or members connecting two joints, as shown in Figures 4, 5 and 6. With reference to Figures 4, 5, 6 and 7, characteristic points, or joints A, B, L, R, D, U are represented. Each of these characteristic points is associated with a pair of first coordinate x and second coordinate y in the image frame. For example, with reference to [Fig.7], the first coordinate xL of the characteristic point L, the first coordinate xR of the characteristic point R, the second coordinate yD of the characteristic point D and the second coordinate yG of the characteristic point U are represented.

[0045] In one embodiment, the steps 300 of image acquisition and 310 of obtaining the individual's postural data on the image can be shared for other functionalities of the system or of the vehicle. In other words, the acquisition of images and the obtaining of a skeleton of the individual from the acquired image can be implemented in the context of other functionalities embedded in the vehicle system, for example for gesture recognition, distraction detection (for example a driver holding his mobile phone), fatigue detection, etc., or in English for "In-Cabin Monitoring". Consequently, in embodiments, the computer implementing step 300 and / or step 310 can be distinct from that implementing the following steps.By taking up the case described above in which part of the processing is carried out by a remote computer 12, the steps 300 and 310 can be implemented respectively by the camera 10 and a computer 11 on board a vehicle, and the following steps can be implemented by the remote computer 12. Alternatively, the image acquired by the . camera in step 300 can be transmitted to a remote computer 12 which implements the entire method on the image. The result concerning the detection of a conscious or unconscious state of the individual and possibly the triggering of an action of the driving assistance system depending on the result of such detection can then be returned by the remote computer 12 to the on-board computer 11.

[0046] In a step 320, the device 1 calculates values ​​of postural parameters associated with the posture of the individual from the postural data obtained. Such postural parameters may in particular include: - lengths of segments formed by two distinct characteristic points, - angles formed by two segments sharing the same characteristic point.

[0047] With reference to Figures 4 and 5, segments [UD] and [LR] connecting the characteristic points U, D on the one hand and L, R on the other hand, are represented as an example. With reference to [Fig.6], angles a and [3 formed respectively by the segments [DB] and [BR] on the one hand, and the segments [BR] and [RA] on the other hand, are represented as an example.

[0048] In one embodiment, from the finite set of characteristic points obtained in step 310, a set of segments determined by connecting the successive characteristic points two by two and a set of angles formed between these segments can be obtained in step 320. The set of segments and angles obtained then form the postural parameters associated with the posture of the individual on the acquired image.

[0049] Alternatively, other postural parameters can be calculated from the lengths of the segments and / or the angles formed, for example ratios, the relationships between the lengths of the segments.

[0050] In a step 330, the device 1 calculates normalized values ​​of the postural parameters. For this, a set of characteristic points, called reference points, is selected from the set of characteristic points. For example, with reference to FIGS. 4, 5, 6 and 7, the characteristic points L, R, U and D are selected. The characteristic point L corresponds for example to the pixel at the level of the individual's left shoulder (on the acquired image), the characteristic point R corresponds for example to the pixel at the level of the individual's right shoulder (on the acquired image), the characteristic point D corresponds for example to the pixel at the level of the individual's neck (on the acquired image) and the characteristic point U corresponds for example to the pixel at the level of the individual's nose (on the acquired image).The reference points can be selected so that a reference quantity formed from these reference points allows the normalized values ​​of the postural parameters calculated from such a reference quantity to exhibit substantially invariance with respect to a change in the parameter of the camera 10 or even to a change in . posture of the individual. In other words, the reference points can be selected because a reference quantity obtained from their positions or the segments connecting them vary little during a change in posture of the individual (that is, between a posture associated with a conscious state of the individual and a posture associated with an unconscious state of the individual). For example, with reference to Figures 5 and 6, the length of the segment [LR] corresponding to the distance separating the two shoulders of the individual on acquired images varies significantly little (compared to the other segments of the acquired image for example) between a posture of the individual in a conscious state ([Fig.5]) and a posture in an unconscious state ([Fig.6]), acquired according to the same adjustment parameters of the camera 10.In another example, with reference to Figures 4 and 5, the reference points L, R, D and U are selected as reference points in that ratios between the lengths of the segments formed by a considered pair of characteristic points and the lengths of the segments [LR] or [UD] are substantially constant for distinct camera viewing angles between [Fig.4] (camera 10 at the central interior rearview mirror) and [Fig.5] (camera 10 at the steering wheel).

[0051] Thus, the reference points are selected from the finite set of characteristic points of the postural data, so that a normalization of the postural parameters on the basis of such reference points makes it possible to neglect the impact of the positioning (and therefore of the viewing angle) of the camera 10, as well as of the variation in posture of the individual according to these viewing angles.

[0052] Coordinates xR, xL, yD and yG of such reference points, shown in [Fig.7], are then considered to transform (or normalize) the coordinates (x,y) of all other characteristic points into transformed (or normalized) coordinates (x*,y*). Such normalized coordinates (x*,y*) can be expressed as follows: [°053] A—

[0054] v-_ 22k Or : - x* and y* are the normalized coordinates of the normalized position associated with a characteristic point considered, - x and y are the coordinates of the position associated with the characteristic point considered, - xR is the first coordinate of a first predefined reference point R, - xL is a first coordinate of a second predefined reference point L, - yD is a second coordinate of a third predefined reference point D, and - yG is a second coordinate of a fourth predefined reference point U.

[0055] Alternatively, the normalized coordinates (x*,y*) may be expressed as next:

[0056]

[0057]

[0058]

[0059] Or : - x* and y* are the normalized coordinates of the normalized position associated with a characteristic point considered, - x and y are the coordinates of the position associated with the characteristic point considered, - is the first average coordinate of a first predefined reference point R over a predefined number of successively acquired images, - is a first coordinate of a second predefined reference point L on a predefined number of successively acquired images, - yD is a second coordinate of a third predefined reference point D on a predefined number of successively acquired images, and - y; is a second coordinate of a fourth predefined reference point U on a predefined number of successively acquired images. In particular, the reference points L and R are considered as a reference according to the first coordinate (or horizontal reference) in the image frame, while the reference points U and D are considered as a reference according to the second coordinate (or vertical reference) in the image frame. In other words, the first x coordinate of any characteristic point will be transformed on the basis of the reference points L and R (and more particularly of the first coordinates of these points), while the second y coordinate of any characteristic point will be transformed on the basis of the reference points U and D (and more particularly of the second coordinates of these points). The selection of the reference points L, R, U, D can be associated with adjustment parameters of the camera 10 (in particular its positioning in the environment), so that the reference points L, R, U, D can be redetermined as soon as the camera 10 is moved or changes its viewing angle. Furthermore, the selection of the reference points L, R, U, D can be implemented before step 330, for example from step 300 of acquiring the images according to given adjustment parameters of the camera 10. In particular, the calculation of the average coordinates of the reference points can be implemented on a predefined number of images acquired according to such adjustment parameters of the camera 10.For example, for a given positioning of the camera 10, the first ten or twenty images acquired by the camera can be considered by the device 1 to determine the pixels corresponding to the shoulders, nose and neck of the individual and to determine the average positions associated with these. pixels. The selection of reference points and the calculation of their associated positions (or average positions) can then be implemented as an initialization step of the process.

[0060] From such normalized coordinates (x*,y*) for the set of characteristic points forming the postural data, normalized values ​​of the postural parameters can then be calculated during step 330. The lengths (or distances) calculated for each pair of characteristic points (transformed) can for example be Euclidean and the calculated angles can be expressed in radians, being included in the interval [0; 2jt].

[0061] In a step 340, the device 1 transmits the normalized values ​​of the postural parameters associated with the acquired image to a classifier. Such a classifier may for example be implemented by a functional unit of the device 1. Such a classifier may also be implemented by equipment different from the device 1 (for example, another computer 12 or another functional unit remote from the device 1). Such a classifier may for example be a predictive learning model. Such a classifier may in particular be pre-trained beforehand on the basis of a sequential training algorithm, from images representing different states of consciousness of individuals in a given environment. For example, given values ​​of postural parameters associated with the images of FIGS. 4 and 5 may serve as training data for the classifier for the detection of a “conscious state of the individual” class.Given values ​​of postural parameters associated with the image of [Fig.6] may serve as training data for the classifier for detecting an “unconscious state of the individual” class. In one embodiment, the classifier may be a gradient boosting or amplification model. The classifier is pre-trained to predict, from values ​​of postural parameters, a state of consciousness of an individual among at least a conscious state and an unconscious state.

[0062] Advantageously, by transmitting normalized values ​​of postural parameters to the classifier in step 340, the robustness of the classifier (and therefore a fortiori, of the detection method) is not altered in the event of differences in positioning of the camera 10 between the images used to train the classifier and the acquired images (for which a prediction by the classifier is expected). Another advantage of the step of normalizing the postural parameters is that the classifier does not have to be retrained when the adjustment parameters of the camera 10 acquiring the images change. The normalization of the values ​​of the postural parameters then advantageously makes it possible to improve the robustness of the detection method, in particular by taking into account the diversity of the environmental and image acquisition parameters.

[0063] Based at least on the standardized values ​​of the postural parameters transmitted to step 340, a state of consciousness of the individual is determined.

[0064] In an optional step 341, the device 1 may transmit additional data to the classifier, on the basis of which the prediction of the classifier may be determined or refined. The additional data transmitted to the classifier may, for example, comprise biological, biometric, facial data of the individual, collected from the acquired images and from additional sensors included in the environment. The additional data may also come from data collected as part of other functionalities of the vehicle assistance system. For example, the additional data may comprise an orientation of the individual's head, a degree of opening of the individual's eyes or a degree of movement of a face of the individual.

[0065] At the end of step 340 (or optionally, step 341), the device 1 can determine, according to the prediction of the classifier, a state of consciousness of the individual associated with the acquired image on the basis of the normalized values ​​of the parameters (and optionally, additional data). On the basis of such a determined state of consciousness, the device 1 can trigger an action at a step 351. For example, such an action can be a simple iteration of the method described on another acquired image, if a conscious state of the individual is determined at the end of step 340 (and optionally, step 341). On the contrary, if an unconscious state of the individual is determined at the end of step 340 (and optionally, step 341), such an action can be triggered by an alert signal emitted by the devices to a control unit integrated into the vehicle or remote from the vehicle, such an action being able to be: - an alert action, for example the emission of a visual and / or audible signal via a vehicle interface in order to alert the individual or any other individual in or around the environment (for example another passenger or other individuals in other neighboring vehicles), and / or - an emergency action, for example activation of automatic vehicle control, triggering of an autonomous emergency maneuver such as parking or braking or triggering of an emergency call (for example to a medical service).

[0066] Optionally, the action triggered in step 351 as previously described may be conditioned by the verification of a triggering criterion in an optional step 350. Such a triggering criterion verified in the optional step 350 may for example be a number of occurrences of an unconscious state detected by the classifier on a successive number of acquired images exceeding a predefined threshold (for example five or ten images). Thus, the alert or emergency action may be triggered in step 351 if an unconscious state of the individual is detected on a successive number of images exceeding a threshold. Such a triggering criterion then advantageously makes it possible to overcome specific detection errors of the classifier or specific positions of the individual associated with an unconscious state even when the individual is conscious (for example, the individual tries to grab an object and approaches a posture learned by the classifier as being associated with an unconscious state).

[0067] Furthermore, the verification of a triggering criterion in the optional step 350 advantageously makes it possible to refine the possibility of actions taken by the device 1 in step 351. For example, depending on the number of successive occurrences of the determined unconscious state, different alert levels or actions can be triggered according to different predefined thresholds (or ranges). For example, if an unconscious state of the individual is detected on two to five successive images, an alert action can be issued in step 351. If an unconscious state of the individual is detected on more than five successive images, an emergency action can be triggered in step 351.

Claims

Claims

1. Method for detecting a state of consciousness of an individual in an environment, implemented by a detection device (1), the device being connected to a camera (10) positioned in the environment so as to allow the acquisition (300) of images on which the individual is visible, the method comprising the following steps: - obtaining (310) data, called postural, relating to a posture of the individual in the environment, said data being extracted from at least one image acquired by the camera (10) and comprising the positions in the image of a finite set of characteristic points (A, B, L, R, D, U) of the individual, - calculate (320), from the positions of the characteristic points (A, B, L, R, D, U), values ​​of postural parameters associated with the posture of the individual in the environment, said postural parameters comprising at least: * lengths of segments ([UD], [LR], [DB], [BR], [RA]) formed by two distinct characteristic points among the finite set of characteristic points (A, B, L, R, D, U), * angles (a, |3) formed by two segments ([UD], [LR], [DB], [BR], [RA]) having a common characteristic point, the values ​​of the postural parameters being calculated (330) from standardized positions of the characteristic points (A, B, L, R, D, U), said positions being standardized with respect to at least one reference quantity of the individual measured on the acquired image, - determining (340), based on said postural data, a state of consciousness of the individual from among at least one conscious state and one unconscious state of the individual.

2. Method according to claim 1, reference points (L, R, U, D) respectively associated with reference positions being selected from the finite set of characteristic points (A, B, L, R, D, U), and in which the postural parameter values ​​are normalized by transforming, for each characteristic point of the finite set of characteristic points (A, B, L, R, D, U)„ the associated position into a normalized position expressed by the following normalized coordinates: - x* and y* are the normalized coordinates of the normalized position associated with the characteristic point, - x and y are the coordinates of the position associated with the characteristic point, - xR is a first coordinate of a first predefined reference point, - xL is a first coordinate of a second predefined reference point, - yD is a second coordinate of a third predefined reference point, and - yG is a second coordinate of a fourth predefined reference point.

3. Method according to any one of claims 1 or 2, in which the state of consciousness of the individual is determined (340) by a classifier from the values ​​of the postural parameters, said classifier being previously trained on training data comprising at least training values ​​of the postural parameters.

4. A method according to any one of the preceding claims further comprising: - obtaining (341) facial data of the individual comprising one of at least: an orientation of the head of the individual, a degree of opening of the eyes of the individual, a degree of movement of a face of the individual, and wherein the state of consciousness of the individual is further determined as a function of said facial data.

5. Method according to any one of the preceding claims, further comprising: - the emission (351), in the event of an unconscious state of the detected individual, of an alert signal to a control unit.

6. Device (1) for detecting a state of consciousness of an individual in an environment, said device being connected to a camera (10) positioned in the environment so as to allow the acquisition of images on which the individual is visible, the device comprising at least one calculator (11, 12) configured to: - obtain (310) data, called postural data of the individual, relating to a posture of the individual in the environment, said data being extracted from at least one image acquired by the camera and comprising the positions in the image of a finite set of characteristic points (A, B, L, R, D, U) of the individual, - calculate (320), from the positions of the characteristic points (A, B, L, R, D, U), values ​​of postural parameters associated with the posture of the individual in the environment, said postural parameters comprising at least: * lengths of segments ([UD], [LR], [DB], [BR], [RA]) formed by two distinct characteristic points among the finite set of characteristic points (A, B, L, R, D, U), * angles (a, |3) formed by two segments ([UD], [LR], [DB], [BR], [RA]) having a common characteristic point, the values ​​of the postural parameters being calculated (330) from standardized positions of the characteristic points (A, B, L, R, D, U), said positions being standardized with respect to at least one reference quantity of the individual measured on the acquired image, - determining (340), based on said postural data, a state of consciousness of the individual from among at least one conscious state and one unconscious state of the individual.

7. Computer program comprising instructions for implementing the method according to one of claims 1 to 5 when this program is executed by a processor.

8. Non-transitory recording medium readable by a computer on which is recorded a program for implementing the method according to one of claims 1 to 5 when this program is executed by a processor.

Images