Method for recognizing a predetermined gesture

Abstract

A method for recognizing a predetermined gesture, such as a kick (16) or hand movement, intended to control the opening or closing, and / or the locking or unlocking of a motor vehicle opening, the method being implemented by a computing unit (22). The method comprises obtaining an electrical signal from a non-contact motion sensor (18), the electrical signal corresponding to a gesture; determining at least three windows in the electrical signal corresponding to the approach, stop, and withdrawal phases of the gesture; determining the value of at least one characteristic associated with the electrical signal for each of the time windows; and using a boosting-type classification algorithm to determine whether the gesture belongs to one of the following two classes: a positive event indicating that the gesture corresponds to the predetermined gesture, and a negative event indicating that the gesture does not correspond to the predetermined gesture.Figure from the summary: Figure 1.

Description

Title of the invention: Method for recognizing a predetermined gesture technical field

[0001] This disclosure relates to the field of methods for recognizing a predetermined gesture, intended to control the opening or closing, and / or the locking or unlocking of an opening of a motor vehicle, and using motion sensors based on non-contact detection, including body motion sensors such as a kick. Previous technique

[0002] A predetermined gesture sensor, in particular a sensor capable of recognizing a kick-type gesture, 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 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.

[0006] It would therefore be beneficial to have a kick sensor that is more reliable in its motion detection. Summary

[0007] This disclosure improves the situation.

[0008] A method is proposed for recognizing a predetermined gesture, such as a kick or a hand movement, intended to control the opening or closing, and / or the locking or the unlocking of a door of a motor vehicle. The method is implemented by a computing unit and comprises obtaining an electrical signal from a non-contact motion sensor, the electrical signal corresponding to a gesture; determining at least three time windows in the electrical signal, said at least three time windows corresponding to the approach, stop and retraction phases of the gesture; determining the value of at least one characteristic associated with the electrical signal for each of said at least three time windows; forming a vector comprising the values ​​of said at least one characteristic of the electrical signal for each of said at least three time windows;and the use of a boosting-type classification algorithm to determine from the vector whether the gesture belongs to one of the following two classes: a positive event indicating that the gesture corresponds to the predetermined gesture, and a negative event indicating that the gesture does not correspond to the predetermined gesture.

[0009] Gesture sensors, particularly kick sensors, are traditionally developed using capacitive technology. That is, they utilize capacitive touch technology to detect and interpret kicking gestures or foot movements. Capacitive sensors operate on the principle of capacitance, which is the ability of an object to store an electrical charge. In the context of a kick sensor, capacitive technology is used to create an invisible capacitive field in a specific area, typically near the rear bumper of a vehicle. When a person's foot enters or interacts with this field, it disturbs the electrical charge, and the sensor can detect the change in capacitance.

[0010] More recently, a new sensor technology based on radar technology (24 GHz or UWB) has been coupled with a naive algorithm. A kick sensor based on radar technology uses the principles of radar (radio waves) to detect and interpret kicking gestures or foot movements. The sensor emits radio waves and analyzes their reflections to detect the presence and movement of a person's foot. The naive algorithm is designed to extract the characteristics of the signal associated with the reflected radio waves and compares them to a threshold (min / max) that is representative of a positive foot movement gesture.

[0011] The main weakness of the naive algorithm is its lack of robustness against abuse, particularly from involuntary pedestrians (in the case of a foot sensor). Indeed, the misuses and the events to be detected often have the same characteristic values ​​when considered independently.

[0012] The proposed method allows for a more in-depth analysis of the signal by dividing the signal into time windows and using characteristics of each time window with a supervised learning algorithm, which thus classifies the gesture as positive or negative, a movement of a person's foot, based on these characteristics. features.

[0013] The presented method proposes using a combination of feature values ​​to solve the classification problem. Instead of having all the features tested in parallel, as in the case of using a naive algorithm, supervised learning makes it possible to design a sequence of feature tests coupled with a weighting of the results of each test to split the classes to be sorted as much as possible.

[0014] The features described in the following paragraphs may optionally be implemented independently of each other or in combination with each other:

[0015] - said at least one characteristic of the electrical signal for each of said at at least three time windows include at least two of the following characteristics: - the duration of the approach time window; - the duration of the stopping time window; - the duration of the withdrawal time window; - the sum of the durations of the time windows for approach, stop and withdrawal; - the number of points in the time window of approach that have a positive phase shift; - the number of points in the withdrawal time window that have a negative phase shift; - the ratio of the number of points in the approach time window that have a positive phase shift to the duration of the approach time window; - the ratio of the number of points in the withdrawal time window that have a negative phase shift to the duration of the withdrawal time window; - number of absolute variations of the signal determined by incrementing a counter by the absolute value of the difference in amplitude between two consecutive points, each time the absolute value of the difference in amplitude between two consecutive points is greater than a statistically defined threshold and corresponding to an abnormal amplitude rise between two consecutive points; - number of absolute variations of the period of the signal determined by incrementing a counter by the absolute value of the difference in period between two consecutive points, each time the absolute value of the difference in period between two consecutive points is greater than a statistically defined threshold and corresponding to an abnormal amplitude rise between two consecutive points; - the number of absolute phase changes of the signal determined by incrementing a counter by the absolute value of the phase difference between two consecutive points, each time the absolute value of the phase difference between two consecutive points is greater than a statistically defined threshold and corresponding to an abnormal increase in amplitude between two consecutive points; - ratio of the average signal amplitude in the approach time window to the average signal amplitude in the withdrawal time window.

[0016] - the motion sensor is a foot motion sensor, arranged in a util the sensor, located under a car trunk, includes two antennas, a first antenna directed in use towards the rear of the car, and a second antenna directed in use towards the underside of the car, and wherein said at least one characteristic of the electrical signal for each of said at least three time windows comprises: - the ratio of the average signal amplitude in the approach time window for the first antenna to the average signal amplitude in the approach time window for the second antenna; and - the ratio of the average signal amplitude in the withdrawal time window for the first antenna to the average signal amplitude in the withdrawal time window for the second antenna.

[0017] - determining from the vector whether the gesture belongs to one of the two classes includes determining a weight associated with the gesture from the characteristic vector and comparing the weight with a threshold value delimiting whether or not it belongs to a positive impact event.

[0018] - the classification algorithm is of the boosting type, for example AdaBoost (Adaptive Boosting), Gradient Boosting, or XGBoost (Extreme Gradient Boosting).

[0019] - the method further comprises a step of issuing an opening instruction or of closing, and / or locking or unlocking an opening of a motor vehicle, in response to the determination of the positive event indicating that the action corresponds to the predetermined action.

[0020] According to another aspect, a computer program is proposed comprising instructions for the implementation of all or part of a process according to the invention when this program is executed by a processor.

[0021] According to another aspect, a non-transient, computer-readable recording medium is proposed on which such a program is recorded.

[0022] In another aspect, a method for opening a car trunk comprises using the gesture determination method described above, and issuing a command to open or close, and / or lock or unlock the trunk in response to the determination of a positive event indicating that the gesture corresponds to the predetermined control gesture. In one example, the command is issued by a processing unit separate from the processing unit implementing the gesture determination method.

[0023] According to another aspect, a car is proposed comprising the above non-transient recording medium. Brief description of the drawings

[0024] Other features, details and advantages will become apparent from reading the detailed description below and from analyzing the accompanying drawings, in which: Fig. 1

[0025] [Fig.1] shows an illustration of a user opening a car trunk with a gesture of the foot; Fig. 2

[0026] [Fig.2] shows an example of an electrical signal from the non-contact gesture sensor following the detection of a gesture by an antenna, the signal being decomposed into a real part (Q) and an imaginary part (I); Fig. 3

[0027] [Fig.3] shows the phase, amplitude and period graphs of the signal from [Fig.2]; Fig. 4

[0028] [Fig. 4] shows a time window segmentation of a phase shift between the imaginary part (I) and the real part (Q) of the signal in [Fig. 2] according to one embodiment; and Fig. 5

[0029] [Fig. 5] shows an example of distribution into positive and negative events following the use of a boosting-type classification algorithm on the signal from [Fig. 2]. Description of embodiments

[0030] 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 their foot 16. For this purpose, a non-contact sensor 18 is located at the rear of the car below the trunk 12 at the level of the bumper. By passing their foot 16 in front of the sensor 18 in a kicking motion, a trunk opening system 20 detects the kick and triggers the opening of the trunk 12. The system 20 includes the sensor 18 and a processing unit 22 which includes a boosting-type classification algorithm to determine whether the gesture detected by the sensor 18 is indeed a kicking gesture.

[0031] The non-contact sensor 18 can, by way of example, be of two types. It can have one detection antenna (like a UWB sensor) or two detection antennas (for example, a 24 GHz sensor). In the case where the sensor 18 has two detection antennas, the sensor 18 can have a first antenna 18a directed towards the rear of the car 14 (axis A1), and a second antenna 18b directed towards the underside of the car 14 (axis A2). These two antennas combined allow the detection of the Foot movement 16. Sensor 18 receives from each antenna 18a, 18b 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.

[0032] The method 30 for determining a gesture event will now be described. The method makes it possible to determine whether the gesture recorded by the sensor 18 corresponds to a predetermined control gesture. The predetermined gesture is preferably a "kick," that is, a forward swing of the foot followed by a return to its initial position. In variations, the gesture could be more complex, but advantageously it will include an approach phase (of the vehicle 14), a stopping phase, and a withdrawal phase (reversal relative to the vehicle 14). The gesture could also be a hand gesture, having similar characteristics, possibly associated with functionalities of the vehicle 14 other than opening the trunk.

[0033] The method 30 uses the processing unit 22 to analyze the signals received by the non-contact sensor 18 and determine whether a predetermined gesture, i.e., a kick, has occurred. The classification performed by the method 30 avoids false positives or false negatives.

[0034] In a first step, an electrical signal corresponding to a gesture detected by the sensor 18 is obtained. Figure 2 illustrates an example of a signal obtained for an antenna 18a or 18b. This signal comprises a real part (Q) and an imaginary part (I). Figure 3 shows the variations in phase, amplitude, and period of the signal.

[0035] In a second step, the received signal is processed. From the signal, at least three time windows are determined: T1, T2, and T3. These three time windows correspond to the approach, stop, and withdrawal phases of the gesture, that is, the approach, stop, and withdrawal phases of the foot in front of the sensor 18. In order to determine these windows, the phase shift between the imaginary part (I) and the real part (Q) of the signal is analyzed. During a gesture, the phase is initially positive and stable, then undergoes a loss of value, an inflection point appears, and the phase becomes strongly negative, before decreasing and stabilizing. The approach time window T1 will therefore correspond to a time window where the phase is positive and stable.The stopping time window T2 will be the time window adjacent to the approach time window T1, and will therefore correspond to a time window where the phase undergoes a loss of value, an inflection point appears, and the phase becomes strongly negative, then decreases. The withdrawal time window T3 will be the time window adjacent to the stopping time window T2, and will therefore correspond to a time window where the phase is negative and stable. A signal processing algorithm can be used to determine these three time windows.

[0036] There could be more than three time windows. For example, the segmentation could include an additional window before the T4 approach window shown on [Fig.4], and an additional window after the T5 retraction window shown in [Fig.4],

[0037] In a third step, the values ​​of at least one characteristic of the electrical signal are determined for each of the aforementioned at least three time windows. The characteristics may be the following: - the duration of the approach time window; - the duration of the stopping time window; - the duration of the withdrawal time window; - the sum of the durations of the time windows for approach, stop and withdrawal; - the number of points in the time window of approach that have a positive phase shift; - the number of points in the withdrawal time window that have a negative phase shift; - the ratio of the number of points in the approach time window that have a positive phase shift to the duration of the approach time window; - the ratio of the number of points in the withdrawal time window that have a negative phase shift to the duration of the withdrawal time window; - the number of absolute variations of the signal (in its entirety, i.e., including the real and imaginary parts) determined by incrementing a counter by the absolute value of the amplitude difference between two consecutive points, each time the absolute value of the amplitude difference between two consecutive points is greater than a statistically defined threshold corresponding to an abnormal amplitude increase between two consecutive points. In one example, this threshold value is 30; - the number of absolute variations in the signal period determined by incrementing a counter by the absolute value of the period difference between two consecutive points, each time the absolute value of the period difference between two consecutive points is greater than a statistically defined threshold corresponding to an abnormal amplitude increase between two consecutive points. In one example, this threshold value is 2; - the number of absolute phase variations of the signal determined by incrementing a counter by the absolute value of the phase difference between two consecutive points, each time the absolute value of the phase difference between two consecutive points is greater than a statistically defined threshold corresponding to an abnormal amplitude increase between two consecutive points. In one example, this threshold value is 2.

[0038] In the case where the sensor 18 has two antennas, the following characteristics can also be considered: - ratio of the average signal amplitude (in its entirety, i.e., including real part and imaginary part) in the time window of approach by the average of the signal amplitude in the time window of withdrawal, the signal having, according to an example, been modified by a high-pass filter to center the signal at 0; - the ratio of the average signal amplitude in the approach time window for the first antenna to the average signal amplitude in the approach time window for the second antenna; and - the ratio of the average amplitude of the complete signal (i.e., including the real and imaginary parts) in the time window of withdrawal for the first antenna to the average amplitude of the signal in the time window of withdrawal for the second antenna, the signal having, in an example, been modified by a high-pass filter to center the signal at 0.

[0039] In a fourth step, a vector is formed from the values ​​of the electrical signal characteristics determined for each of the time windows. The vector will have the form of an N x 1 vector of dimension 1, with the N characteristics of this signal.

[0040] In a fifth step, a boosting type classification algorithm determines from the vector whether the predetermined gesture belongs to one of the following two classes: positive event of predetermined gesture (the gesture is indeed a kick), and negative event not corresponding to the predetermined gesture (i.e. in this case, a kick).

[0041] The classification algorithm is a two-class boosting set-theoretic algorithm.

[0042] A boosting algorithm is a machine learning technique that combines the predictions of several weak or basic models to create a stronger and more accurate predictive model. Boosting belongs to the ensemble learning methods, where several models are trained and combined to make better predictions than the individual models.

[0043] In boosting, the basic models are generally simple and weak, such as decision trees with limited depth or shallow neural networks. Each basic model is trained on a subset of the training data or with different weights assigned to the data points. The models are trained iteratively, with each subsequent model focusing on the examples that were misclassified by the previous models.

[0044] During the training process, reinforcement assigns higher weights to the misclassified examples, effectively giving them more importance. Subsequent models are then trained to rank the misclassified examples, attempting to correct the errors made by the previous models. This iterative process continues until a predefined stopping criterion is met, such as a number maximum number of models or a desired level of precision.

[0045] The final prediction is made by combining the predictions of all the basic models, generally through a weighted voting system. The weights are determined based on the performance of each basic model during training. The idea behind boosting is that by combining several weak models, each focusing on different aspects of the data, the overall model can capture complex relationships and improve predictive accuracy.

[0046] Popular boosting algorithms include AdaBoost (Adaptive Boosting), Gradient Boosting and XGBoost (Extreme Gradient Boosting).

[0047] With reference to [Fig. 5], the classification algorithm will output a weight. If this weight is greater than a threshold value (illustrated by the dashed line), it will be concluded that a positive kick event has occurred. If this weight is less than the threshold value, it will be concluded that a negative event not corresponding to a kick has occurred.

[0048] The threshold value is determined upstream of process 30 during the training and calibration of the classification algorithm.

[0049] The classification algorithm may or may not use the N features provided to it to determine the weight of the event.

[0050] The use of a boosting-type classification algorithm reduces the dimensionality of the gesture determination problem. If the features had been considered in isolation, the problem could not have been solved. Thanks to boosting, sequential tests are performed, which allows a weight to be applied, resulting in a single dimension, and finally, classification.

[0051] At the end of the process, if it is determined that the action corresponds to the predetermined control action (positive event), a command to open or close, and / or to lock or unlock the safe may be issued. If, on the other hand, it is determined that the action does not correspond to the predetermined control action (negative event), no command is issued.

[0052] The sending of the instruction can be controlled by the computing unit 22 or another computing unit connected to the computing unit 22.

Claims

Demands

1. A method for recognizing a predetermined gesture performed by a user, such as a kick (16) or hand movement, intended to control the opening or closing, and / or the locking or unlocking of a motor vehicle opening, when said predetermined gesture is recognized, the method being implemented by a computing unit (22) and comprising: a. Obtaining an electrical signal from a non-contact motion sensor (18), the electrical signal corresponding to a gesture; b. The determination of at least three time windows (T1, T2, T3) in the electrical signal, said at least three time windows corresponding to phases of approach, stop and withdrawal of the gesture; c. The determination of the value of at least one characteristic associated with the electrical signal for each of the said at least three time windows; d. The formation of a vector comprising the values ​​of said at least one characteristic of the electrical signal for each of said at least three time windows; and e. The use of a boosting type classification algorithm to determine from the vector whether the gesture belongs to one of the following two classes: positive event indicating that the gesture corresponds to the predetermined gesture, and negative event indicating that the gesture does not correspond to the predetermined gesture.

2. A method according to claim 1, wherein said at least one characteristic of the electrical signal for each of said at least three time windows comprises at least two of the following characteristics: - the duration of the approach time window; - the duration of the stopping time window; - the duration of the withdrawal time window; - the sum of the durations of the time windows for approach, stop and withdrawal;

3. - the number of points in the time window of approach that have a positive phase shift; - the number of points in the withdrawal time window that have a negative phase shift; - the ratio of the number of points in the approach time window that have a positive phase shift to the duration of the approach time window; - the ratio of the number of points in the withdrawal time window that have a negative phase shift to the duration of the withdrawal time window; - number of absolute variations of the signal determined by incrementing a counter by the absolute value of the difference in amplitude between two consecutive points, each time the absolute value of the difference in amplitude between two consecutive points is greater than a statistically defined threshold and corresponding to an abnormal amplitude rise between two consecutive points; - number of absolute variations of the signal period determined by incrementing a counter by the absolute value of the period difference between two consecutive points, each time the absolute value of the period difference between two consecutive points is greater than a statistically defined threshold and corresponding to an abnormal amplitude rise between two consecutive points; - number of absolute variations of the signal phase determined by incrementing a counter by the absolute value of the phase difference between two consecutive points, each time the absolute value of the phase difference between two consecutive points is greater than a statistically defined threshold and corresponding to an abnormal amplitude rise between two consecutive points; - ratio of the average signal amplitude in the approach time window to the average signal amplitude in the withdrawal time window. A method according to claim 1 or 2, wherein the motion sensor is a foot motion sensor, disposed in use under a car trunk, the sensor including two antennas, a first antenna directed in use towards the rear of the car, and a second antenna directed in use towards the underside of the car, and wherein said at least one electrical signal characteristic for each of said at least three time windows includes: - the ratio of the average signal amplitude in the approach time window for the first antenna to the average signal amplitude in the approach time window for the second antenna; and - the ratio of the average signal amplitude in the withdrawal time window for the first antenna to the average signal amplitude in the withdrawal time window for the second antenna.

4. A method according to any one of the preceding claims, wherein the determination from the vector whether the gesture belongs to one of the two classes includes the determination of a weight associated with the gesture from the characteristic vector and the comparison of the weight with a threshold value delimiting the membership or not of a positive blow event.

5. A method according to any one of the preceding claims, wherein the classification algorithm is of the boosting type, for example Adaptive Boosting, Gradient Boosting, or Extreme Gradient Boosting.

6. A computer program comprising instructions for carrying out the method according to any one of claims 1 to 5 when this program is executed by a processor.

7. A non-transient, computer-readable recording medium on which a program is recorded for the implementation of the method according to any one of claims 1 to 5 when this program is executed by a processor.

8. Method for opening a car trunk (12) (14) comprising: using the method for determining a gesture according to any one of claims 1 to 5, and issuing an instruction to open or close, and / or lock or unlock the trunk in response to the determination of the positive event indicating that the gesture corresponds to the predetermined control gesture.

9. An opening method according to the preceding claim, wherein the setting is issued by a computing unit separate from the computing unit implementing the gesture determination method.

10. Car (14) comprising the non-transient recording medium of claim 7.

Images