Method of determining a gesture

Abstract

Method (30) for determining a predetermined gesture, implemented by a calculation unit (22) connected to a first sensor (18, 19) and a second motion sensor, of the radiofrequency type, respectively producing a first and a second electrical signal and arranged in proximity to each other on a motor vehicle. The method comprises the use of a classification algorithm to obtain first and second scores (SC1, SC2) associated respectively with the first and second electrical signals, as well as, for each of said scores, respective probability densities that the gesture belongs to at least two categories among a first category (Y) and a second category (X), one category corresponding to the determination that the gesture is the predetermined gesture. Abstract figure: Figure 6

Description

Title of the invention: Method for determining a gesture technical field

[0001] This disclosure relates to the field of methods for determining a gesture, such as a kick or a hand strike, using motion sensors, and in particular non-contact motion sensors. Previous technique

[0002] A kick sensor can allow a user to activate or control certain vehicle functions, such as opening a car trunk, by making a movement or gesture with their foot. This is a form of hands-free interaction designed to offer convenience and ease of use.

[0003] Instead of using a physical button on the car or on its key fob, the user can activate the mechanism by simply kicking a specific area of ​​the rear bumper near the trunk, where the sensor is located. The sensor detects the movement and triggers the unlocking and / or opening of the trunk.

[0004] The purpose of a kick sensor is to allow the user to access the vehicle's cargo area without requiring them to put down their belongings or physically interact with the car, such as when their hands are full with grocery bags.

[0005] While kick sensors can be convenient and provide hands-free access, they also have some drawbacks, particularly regarding sensitivity. Kick sensors can sometimes be too sensitive or not sensitive enough, and may then misinterpret other movements or objects, leading to unintentional activations. For example, if the user or a passerby accidentally brushes against the sensor area while walking past the vehicle, this could trigger the opening or closing of the trunk. Summary

[0006] This disclosure improves the situation.

[0007] A method for determining a predetermined gesture, such as a kick or a hand strike, is proposed. The method is implemented by a computing unit connected to a first and a second radio-frequency motion sensor, arranged in close proximity to each other on a motor vehicle. The first and second sensors are adapted to produce, respectively, a first electrical signal and a second electrical signal upon the occurrence of a gesture within a detection zone. The method comprises: the use of a classification algorithm Fication for each of the first and second electrical signals in order to obtain respectively first and second scores associated respectively with the first and second electrical signals, as well as, for each of the first and second scores, respective probability densities that the gesture belongs to at least two categories from a first category and a second category depending on the score value, one from the first category and the second category corresponding to the determination that the gesture is the predetermined gesture, each classification algorithm having been trained individually for each of the first and second motion sensors so as to obtain: a probability density distribution that the gesture belongs to said first category depending on the score value, and a probability density distribution that the gesture belongs to said second category depending on the score value,and a first range of score values, associated with a gesture considered to belong to the first category, a second range of score values, associated with a gesture considered to belong to the second category, and a third range of score values, associated with a gesture considered indeterminate; and the determination that the gesture belongs to the first or second category, depending on the membership of the first and second scores in the first, second, and third ranges and the temporal order in which the first and second scores were determined.

[0008] According to another aspect, a computer program is proposed that includes instructions for implementing all or part of a process as defined herein when this program is executed by a processor. According to another aspect, a non-transient, computer-readable recording medium is proposed on which such a program is recorded.

[0009] The features set out in the following paragraphs can optionally be implemented independently of each other or in combination with each other: - the first score is obtained first, and the determination that the gesture is of the first or second category includes: the determination of the belonging of the first score to the first range, second range or third range; if the first score belongs to the first range, the determination that the gesture is of the first category; and if the first score belongs to the second range, the determination that the gesture is of the second category.

[0010] - once the gesture is determined to be of the first or second category, when a first clock reaches the end of a first time window, a reset of a memory containing information on the determination of the gesture, and a reset of the first clock.

[0011] - the first score belongs to the third range, the start of a second clock; As long as the second clock has not reached the end of a second window of Time: When the second signal arrives at the calculation unit: obtaining the second score; determining whether the second score belongs to the first range, second range, or third range; if the second score belongs to the first range, determining that the gesture is in the first category; if the second score belongs to the second range, determining that the gesture is in the second category; if the second score belongs to the third range, the gesture is that of the category with the highest probability density between the first and second categories for the first and second scores.

[0012] - the second signal does not reach the processing unit before the end of the second Time window: Comparison of the first score with a first threshold; if the first score is below the first threshold, the determination that the gesture is of the first category; if the first score is above the first threshold, the determination that the gesture is of the second category.

[0013] - when the action is determined and the second clock reaches the end of the second time window, a reset of a memory containing information on determining the gesture, and a reset of the first clock.

[0014] - the predetermined gesture is a kick, and the detection zone is a part rear of a car vertically below a trunk of the car, the method further comprising sending a signal from the computing unit to a control unit, to operate the opening or closing of the trunk when the gesture is of the second category.

[0015] - when the predetermined gesture is not a kick, an absence of sending signal from the computing unit to the control unit. Brief description of the drawings

[0016] Other features, details and advantages will become apparent upon reading the detailed description below, and upon analysis of the accompanying drawings, on which:

[0017] [Fig-1] shows an illustration of a user opening a car trunk of a foot gesture, the car being equipped according to an embodiment of two contactless sensors arranged in a detection zone below the trunk of the car;

[0018] [Fig.2] shows an example of score distribution for a first of the two sensors, each score giving a probability that the gesture is of a first category Y or a second category X;

[0019] [Fig.3] shows an example of score distribution for a second of the two sensors, each score giving a probability that the gesture is of a first category Y or a second category X;

[0020] [Fig.4] shows an example of signal processing for the graph of [Fig.2];

[0021] [Fig.5] shows an example of signal processing for the graph of [Fig.3];

[0022] [Fig.6] shows a flowchart representing a gesture determination method using the scores of the two non-contact sensors. Description of the implementation methods

[0023] Reference is now made to [Fig. 1] which shows a user 10 opening the trunk 12 of a vehicle 14, in this case a car, by means of the action of his foot 16. For this purpose, two non-contact sensors (first sensor 18, second sensor 19) are arranged at the rear of the car below the trunk 12 at the level of the bumper in a detection zone ZD. By passing their foot 16 into the detection zone ZD in front of one or both sensors 18, 19, a trunk opening system 20 detects whether the gesture is a kick in order to control the opening and / or closing of the trunk 12. The system 20 includes a processing unit 22 which determines whether the detected gesture is indeed a kick based on the electrical signals provided by the sensors 18, 19. The detection of this gesture can then be used to operate a vehicle control, such as opening the trunk 12.

[0024] In the example of [Fig. 1], each of the sensors 18, 19 is positioned substantially at one lateral end of the vehicle 14's body, below the bumper. They are positioned so as to detect movement occurring under the bumper, towards the lateral midpoint of the bumper, such as the user's foot gesture illustrated in [Fig. 1]. The sensors 18, 19 could be positioned more or less close to each other laterally, as long as a minimum threshold distance between them is maintained, so that they provide distinct signatures. The vehicle 14 can be equipped with more than two rear sensors 18, 19 that can be used for kick detection according to the method described below. The vehicle 14 could, for example, have three or four.

[0025] The non-contact sensors 18, 19 can, by way of example, be of two types. They can each have one detection antenna (like a UWB sensor) or two detection antennas (for example, a 24 GHz sensor). In the case where sensor 18 has two detection antennas, the first antenna can be directed towards the rear of the vehicle 14, and the second towards the underside of the vehicle 14. Sensor 18 receives from each antenna an electrical signal corresponding to the detected movement. This signal comprises a real part I and an imaginary part Q which have a time phase shift P between them. The sensors could also be CW (continuous wave) type sensors, based on radar technology, and have two antennas.

[0026] The electrical signal from each sensor can then be used to determine whether what has been detected is a kicking gesture. To this end, a two-category classification algorithm is associated with each sensor to aid in determining the gesture Depending on the signal: a first category Y for a gesture categorized as not being a kick, and a second category X for a gesture categorized as a kick. Although a kick is mentioned, the process and systems described herein apply to other body gestures, such as a hand gesture, and to the control of elements other than the chest. Furthermore, although only two categories are discussed, the classification could potentially involve more than two categories.

[0027] The sensors are autonomous and each has its own individually trained classification algorithm. The classification algorithm can be a naive algorithm based on the SNR (signal-to-noise ratio) value or a supervised machine learning classifier (e.g., boosting). The same classification algorithm can be used for each of the two sensors 18, 19, or different classification algorithms can be used for each of the two sensors 18, 19. The classification algorithm can be executed by the processing unit 22. The classification algorithm could be executed for each sensor by a processing unit integrated into that sensor, and these processing units could be in contact with the processing unit 22, which determines whether the gesture is definitively a kick based on the classification obtained by each sensor, as will be explained below.

[0028] With reference now to [Fig. 2], an example of reference values ​​obtained after training the classification algorithm for the first sensor 18 is shown. The reference values ​​include the distribution of a score (abscissa) and the probability of occurrence (ordinate) of the first gesture category Y (gesture categorized as not a kick) and the second gesture category X (gesture categorized as a kick). [Fig. 3] shows the same graph of reference values ​​after training the classification algorithm but for the second sensor 19. For each sensor, the distribution of scores for each gesture category is Gaussian. The two Gaussians (gesture category X and Y of the same sensor) are shifted relative to each other along the abscissa. In the example in the figures, the scores are distributed over positive and negative values ​​on the abscissa.It is possible that the scores will be distributed only on positive values ​​or only on negative values.

[0029] The Gaussians for the gesture distributions X and Y of the same sensor have vertices SI, S2 that are offset from each other. The Gaussians characterize the individual performance of each sensor. The scoring method is considered sufficiently effective when the separation of the distributions is above a threshold value, for example, less than 25% overlap between the two distributions. Thus, according to one example, the vertex SI of the Gaussian representing the first gesture category Y has a lower score than the vertex S2 of the Gaussian representing the second gesture category X, and this for each sensor. It could be the reverse. If we look at the example in [Fig. 2], vertex S1 of the Gaussian representing gesture category Y for sensor 18 has a score of 0.25, and vertex S2 of the Gaussian representing gesture category X for sensor 18 has a score of 6. If we look at the example in [Fig. 3], vertex S1 of the Gaussian representing gesture category Y for sensor 19 has a score of -0.6, and vertex S2 of the Gaussian representing gesture category X for sensor 18 has a score of 2.

[0030] The score distributions allow us to determine the probability that the detected gesture is considered a kick by each of the sensors 18, 19. However, the individual data from the sensors 18, 19 alone do not allow us to efficiently determine whether the gesture is indeed a kick. For this purpose, and with reference to [Fig. 6], a gesture determination method 30 makes it possible to arbitrate based on the SCI and SC2 scores, the probability of occurrence of the gesture categories X and Y, and the temporal order of score determination (which can also be the temporal order of arrival of the electrical signals from the sensors to the processing unit 22), when two or more sensors are used, as in the example in the figures. When it is determined that the gesture is a kick, the processing unit 22 can then communicate with a control unit to trigger the opening and / or closing of the chest 12.If, on the other hand, it is determined that the gesture is not a kick, no signal is sent to the control unit and safe 12 is not activated.

[0031] In a signal processing (or calibration) phase preliminary to process 30, and with reference to [Fig.4] and [Fig.5], each score distribution of the trained algorithm for each of the two sensors 18, 19 is divided into three ranges of score values ​​Z1, Z2 and Z3.

[0032] The first range of values, Z1, is a window of scores for which it is almost certain that the detected gesture is of the first category Y (i.e., not a kick). This is a range of values ​​described as having "high confidence." The second range of values, Z2, is a window of scores for which it is almost certain that the detected gesture is of the second category X (i.e., a kick). This is also a range of values ​​described as having "high confidence." The third range of values, Z3, is a window of scores for which there is no conclusion as to whether the detected gesture is of the first X or second category Y. This is a range of values ​​described as having "low confidence."

[0033] To delimit the value ranges Z1, Z2, and Z3, thresholds A-TH-MIN and A-TH-MAX are chosen. The threshold A-TH-MIN delimits the first from the third value range, and the threshold A-TH-MAX delimits the third from the second value range. The first Z1 and second Z2 value ranges are therefore discontinuous, and the third The Z3 value range lies between the Z1 and Z2 value ranges and is defined as the area between A-TH-MIN and A-TH-MAX. The Z3 value range is a low-confidence range because the relative probabilities of the gesture being category X and Y within this range do not allow us to directly deduce whether it is a category X or Y gesture.

[0034] The A-TH-MIN threshold reflects a threshold below which there is a near-zero probability of the gesture being category X compared to the probability of the gesture being category Y. For example, A-TH-MIN is defined as the 2% probability threshold for the Gaussian distribution with the highest score peak among the two Gaussians of the same sensor (i.e., the Gaussian distribution of gesture X for examples of sensors 18 and 19 in [Fig. 2] and [Fig. 3]). Indeed, below the 2% probability that the gesture is category X, the probability that the detected gesture is indeed gesture X is near zero, but within this score range, the probability that the gesture is category Y is non-negligible. Setting the threshold at 2% is an example of a threshold value. The threshold could be 5%, 4%, 3%, 1% or less than 1% as long as it reflects a threshold below which there is an almost zero probability that the gesture is category X as opposed to category Y.In the example in [Fig.4], the A-TH-MIN threshold is defined for a score of -7. In the example in [Fig.5], the A-TH-MIN threshold is defined for a score of -4.

[0035] The single A-TH-MAX reflects a threshold above which there is a near-zero probability of the gesture being of category Y compared to the probability of the gesture being of category X. For example, A-TH-MAX is defined as the 98% probability threshold for the Gaussian with the lowest score peak among the two Gaussians of the same sensor (i.e., the Gaussian of gesture Y for examples of sensors 18 and 19 in [Fig. 2] and [Fig. 3]). Indeed, above the 98% probability that the gesture is of category Y, the probability that the detected gesture is of category Y is near zero, but within this score range, the probability that the gesture is of category X is non-negligible. Establishing the threshold at 98% is an example of a threshold value. The threshold could be 95%, 96%, 97%, 99% or more than 99% as long as it reflects a threshold above which there is a near-zero probability of occurrence of event Y relative to event X.In the example in [Fig.4], the A-TH-MAX threshold is defined for a score of 12.5. In the example in [Fig.5], the A-TH-MAX threshold is defined for a score of 6.5.

[0036] Finally, another threshold value is determined for each score distribution. The A-TH-CENTER threshold value corresponds to the intersection of the score distributions of categories X and Y. For the example in [Fig. 4], the A-TH-CENTER threshold is defined for a score of 0.2. For the example in [Fig. 5], the A-TH-CENTER threshold is defined for a score of -0.2.

[0037] The determination of the thresholds A-TH-MIN, A-TH-CENTER and A-TH-MAX for each of the scores of each of the sensors 18, 19 can be done in any order.

[0038] Once the signal processing has been carried out, the gesture determination method 30 can be implemented by the computing unit 22 each time an electrical signal is provided by at least one of the two sensors 18, 19. The determination is a function of the scores and their temporal order of determination (or the temporal order of arrival of the electrical signals). The method 30 will now be described in relation to [Fig. 6]

[0039] In a first step 32, the signal from one of the two sensors 18,19 is detected first.

[0040] In the next step 33, the signal was entered into the classification algorithm discussed above in order to obtain an SCI score associated with the sensor 18 or 19 whose signal was detected.

[0041] In step 33, it is determined whether the SCI score is in the range of values ​​Z1, Z2, or Z3, the ranges of values ​​being those determined for the score distributions of the algorithm trained by the sensor whose signal was detected.

[0042] If the SCI score is in the first range of values ​​Zl (step 34), it is determined that the gesture is of category Y, and therefore that there is no kick (step 34-1). This determination can be stored in a cache (not shown) containing a value associated with the gesture (example: gesture = Y).

[0043] If the SCI score is in the second range of values ​​Z2 (step 35), it is determined that the gesture is of category X, and therefore that there is a kick (step 35-1). This determination can be stored in the cache containing a value associated with the gesture (example: gesture = X), and then used by the processing unit 22 to subsequently trigger the opening or closing of the chest 12.

[0044] Once the gesture is determined in steps 34-1 or 35-1, it is reset in step 50, as soon as the first clock H1 reaches the end of the first time window T1 (step 49). During the reset, the first clock H1 is restarted and a cache memory containing the value of the determined gesture is cleared.

[0045] If the SCI score is in the third range Z3 (step 48), there is no direct determination of the gesture category at this stage. In this case, at step 36, a second clock H2 is activated for a predetermined time window T2. This clock H2 will wait for any signal from the other sensor. According to one embodiment, the values ​​of the time windows T1 and T2 are such that an overlap of the two time windows is not possible. For example, the T1 window turns off before any start of the T2 window.

[0046] At step 37, if during this time window T2, no signal from the other of the two If no signal is received from sensors 18 or 19, then the gesture associated with the SCI score will be determined relative to the A-TH-CENTER threshold value established previously during calibration on the score distribution for the detected sensor 18 or 19 (step 38). If SCI < A-TH-CENTER, the gesture is determined to be category Y, and therefore there is no kick (step 39). If SCI > A-TH-CENTER, the gesture is determined to be category X, and therefore there is a kick (step 40). If SCI = A-TH-CENTER, the gesture is determined to be category Y or X depending on the settings configured beforehand for this specific case.

[0047] If during the second time window T2, the signal from the other of the two sensors is received, it will then be analyzed to determine its SC2 score (step 41), and then in which range of values ​​Z1, Z2, or Z3 of the score distribution for the other of the detected sensors 18 or 19 belongs the SC2 score.

[0048] If the SC2 score is in the first range Zl (step 42), it is determined that the gesture is of category Y, and therefore that there is no kick (step 42-1). This determination can be stored in the cache.

[0049] If the SC2 score is in the second range Z2 (step 43), it is determined that the gesture is of category X, and therefore that there is a kick (step 43-1). This determination can be stored in a cache containing a value associated with the gesture (example: gesture = X), and then used by the processing unit 22 to subsequently trigger the opening or closing of the chest 12.

[0050] If the SC2 score is in the third range Z3 (step 44), there is no direct determination. In this case, in step 45, the probability values ​​of each gesture are compared for the obtained SCI and SC2 scores (4 values). The selected gesture will then be the one with the highest probability. For example, if the SCI score for sensor 18 is 5 (the SCI score is indeed in zone Z3), the probability density for the gesture to be in category X is 0.095 and that for the gesture to be in category Y is 0.03. If the SC2 score for sensor 19 is 0 (the SC2 score is indeed in the range Z3), the probability density for the gesture to be in category X is 0.12 and that for the gesture to be in category Y is 0.04. Thus, of the four probability densities, the highest is 0.12, which corresponds to a determination that the gesture is of category X (therefore the gesture is categorized as a kick).The determination can then be stored in the cache containing a value associated with the gesture, and used by the computing unit 22 to subsequently trigger the opening and / or closing of the chest 12 if the gesture is of category X.

[0051] Once the action is determined in steps 39, 40, 42-1, 43-1, 45, there is a reset in step 50, as soon as the second clock H2 reaches the end of the time window T2 (step 46). The reset consists of restarting the first clock H1 (which, according to one embodiment, if it is not already at zero, had been paused). by starting the second clock H2), a reset (without starting) of clock H2, and a reset of the value associated with the gesture stored in memory.

Claims

Claims

1. Method (30) for determining a predetermined gesture, such as a kick (16) or a hand, the method being implemented by a computing unit (22) connected to a first sensor (18, 19) and a second radiofrequency type motion sensor, arranged in use in proximity to each other on a motor vehicle, the first and a second sensor being adapted to produce respectively a first electrical signal and a second electrical signal upon the occurrence of a gesture in a detection zone (ZD), the method comprising: a. Using a classification algorithm for each of the first and second electrical signals to obtain first and second scores (SCI, SC2) respectively associated with the first and second electrical signals, and, for each of the first and second scores, respective probability densities that the gesture belongs to at least two categories from a first category (Y) and a second category (X) based on the value of the score, one of the first category and the second category corresponding to the determination that the gesture is the predetermined gesture, each classification algorithm having been trained individually for each of the first and second motion sensors so as to obtain: i. a probability density distribution that the gesture belongs to said first category (Y) as a function of the value of the score, and a probability density distribution that the gesture belongs to said second category (X) as a function of the value of the score, and ii. a first range (Zl) of score values, associated with a gesture considered to belong to the first category, a second range (Z2) of score values, associated with a gesture considered to belong to the second category, and a third range (Z3) of score values, associated with a gesture considered to be indeterminate; and b. Determining whether the gesture belongs to the first or second category, depending on the membership of the first and second scores at the first, second and third ranges and the time order of determining the first and second scores.

2. The method of claim 1, wherein the first score is obtained first, and determining that the gesture is of the first or second category comprises: a. Determining (34, 35) whether the first score belongs to the first range (Z1), second range (Z2) or third range (Z3); b. If the first score belongs to the first range, the determination (34-1) that the gesture is of the first category; and c. If the first score belongs to the second range, the determination (35-1) that the gesture is of the second category.

3.

4. Method according to the preceding claim, in which once the gesture is determined to be of the first or second category, when a first clock (Hl) reaches the end of a first time window (Tl) (49), a reset (50) of a memory containing information on the determination of the gesture, and a reset of the first clock. Method according to one of claims 1 to 3, in which: a. If the first score belongs to the third range (48), the start (36) of a second clock (H2); b. Until the second clock (H2) has reached the end of a second time window (T2) (46): i. When the second signal arrives (37) at the computing unit:

1. Obtaining (41) the second score; 2. Determining (42, 43) whether the second score belongs to the first range, second range or third range; 3. If the second score belongs to the first range (42), the determination (42-1) that the gesture is of the first category; 4. If the second score belongs to the second beach (43), determination (43-1) that the gesture is of the second category; 5. If the second score belongs to the third range (44), the gesture is that of the category having the highest probability density between the first and second categories for the first and second scores (45).

5. Method according to the preceding claim, in which: i. If the second signal does not arrive at the calculation unit before the end of the second time window: ii. Comparing (38) the first score with a first threshold (AT-TH-CENTER), 1. If the first score is lower than the first threshold, determining (39) that the gesture is of the first category; 2. If the first score is higher than the first threshold, determining (40) that the gesture is of the second category.

6. Method according to any one of claims 4 and 5, wherein: when the gesture is determined and the second clock reaches the end of the second time window (46), a reset (50) of a memory containing information on the determination of the gesture, and a reset of the first clock.

7. A method according to any preceding claim, wherein the predetermined gesture is a kick, and the detection area is a rear portion of a car vertically below a trunk of the car, the method further comprising sending a signal from the computing unit to a control unit, to actuate the opening or closing of the trunk when the gesture is of the second category.

8. Method according to the preceding claim, in which when the predetermined gesture is not a kick, an absence of sending of signal from the calculation unit to the control unit.

9. Computer program comprising instructions for implementing implementation of the method according to one of claims 1 to 8 when this program is executed by a processor.

10. 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 8 when this program is executed by a processor.