Vehicle life presence detection system

Abstract

Title: Life Presence Detection System in a Motor Vehicle The invention relates to a system (1) for detecting the presence of life in a motor vehicle (V) by means of a radar (10), the system (1) being configured to implement the following steps: provide a radar image by the radar (10), extract a set of raw features from the radar image, these raw features being selected in particular from at least: spatial information, including range, azimuth angle, elevation angle, points of the radar image, and / or velocity information of points of the radar image and / or power information of points of the radar image, this set of raw features being processed by a robust dispersion measurement and / or outlier filtering configured to remove outlier points, so as to obtain processed features,to provide the set of processed features to a classification algorithm, particularly a machine learning type, capable of delivering information on the presence or absence of life in the vehicle. Figure for the abbreviation: Fig. 1,

Description

Title of the invention: System for detecting the presence of life within a motor vehicle

[0001] The invention relates to a system for detecting the presence of life within a motor vehicle.

[0002] The detection of the presence of life using radar-type systems is used, for example, on board motor vehicles. These detections rely, for example, on conventional statistical measures to extract dispersion information from radar data. However, these conventional approaches can be sensitive to noise and data dispersion, which requires significant data collection efforts and can compromise the reliability of the detection. For example, in a child presence detection application (child left in a car), certain use cases can generate a very large and variable point cloud, impacting the processed features intended for classification and thus degrading classification performance.

[0003] For example, when a passenger leaves a bottle of water inside the vehicle and closes the vehicle door, the water inside the bottle, along with the turbulence of the water within the bottle, may behave, for detection purposes, like breathing. This can generate a false alert of life being present in the vehicle (a false positive).

[0004] The invention aims in particular to remedy these difficulties.

[0005] The invention thus relates to a system for detecting the presence of life inside a motor vehicle by means of a radar, the system being configured to implement the following steps: - to provide a radar image using radar, - extract a set of raw features from the radar image, these raw features being chosen from at least: • spatial information, including range, azimuth angle, elevation angle, radar image points, and / or • speed information from points in the radar image and / or • radar image point power information, - this set of raw features being processed by a robust dispersion measure and / or outlier filtering configured to remove outlier points, so as to obtain processed features, - provide the set of processed characteristics to a classification algorithm, in particular of the type based on machine learning (or "Machine Learning" in English), capable of delivering information on the presence or absence of life in the vehicle.

[0006] The extraction of raw features from the radar image includes, in particular, mathematical transformations calculated on the image pixels. The processed features allow for further processing, in particular for classification.

[0007] The invention thus advantageously proposes to use a new set of processed characteristics based on dispersion measurements that are more robust to noise and outliers. In particular, the invention allows for greater robustness than conventional dispersion measurements that use variances or standard deviations, which are very sensitive.

[0008] The classification algorithm makes it possible to reliably provide information on the presence or absence of life in the vehicle based on the raw characteristics of the radar images, in the sense that robust dispersion measurement and / or filtering of outliers makes it possible to exclude false responses (i.e., issuing false positives that are alerts for the presence of life when there is no life on board the vehicle). The invention makes it possible, in particular, to eliminate outliers or false detections, due, for example, to measurement imperfections. An example of a false detection could be a water bottle with moving water or the movement of one or more objects. The classification performed by the algorithm, which is based on machine learning, is therefore more robust and more reliable.Machine learning can thus be performed without incorporating outliers, which improves machine learning and makes classification more reliable. The invention reduces the rate of misclassification.

[0009] The invention also makes it possible to reduce the amount of data to be processed thanks to these robust dispersion and / or filtering measures. This reduces the computing capacity requirements.

[0010] Advantageously, machine learning uses artificial intelligence.

[0011] The robust dispersion measure is based on the interquartile range (IQR).

[0012] In this case, the data used for classification come only from the second and third quartiles.

[0013] The robust dispersion measure is based on the mean absolute deviation (MAD for "mean absolute deviation").

[0014] We have MAD=median(IXi-Xml) where Xm=median(X) and the term “median” refers to the arithmetic mean operation.

[0015] The IQR is defined as the difference between the first quartile and the third quartile and comprises 50% of the data. The MAD is the median of the absolute deviations from the median. These two methods are more robust to outliers and are therefore suitable for data containing noise, such as radar data.

[0016] Advantageously the set of treated features obtained by the robust dispersion measure is used as primary inputs in the classification algorithm (also called "classifier").

[0017] Advantageously, robust dispersion measurement is applied to all attributes measured by the radar, including power, range, speed (also called Doppler), azimuth angle and elevation angle.

[0018] When these data processed by the robust dispersion measurement are introduced into a classifier, potentially with other data, the invention makes it possible to obtain performance superior to that of conventional statistical measurements commonly used, by offering better accuracy and greater robustness to noise.

[0019] According to one aspect of the invention, the robust dispersion measurement is applied after the calculation or obtaining of the raw characteristics.

[0020] In another example of an implementation of the invention, outlier filtering is applied before calculating the processed features, and then the processed features are calculated by a classical dispersion measure which uses, for example, mainly variances or standard deviations.

[0021] This improves conventional dispersion measurements by filtering outliers before calculating the processed features. This method demonstrates superior performance compared to that obtained using only processed features derived from a conventional dispersion measurement.

[0022] The classification algorithm allows detection of the presence of children and / or location / classification of occupants in the vehicle.

[0023] Advantageously the invention makes it possible to give a more precise representation of the scene observed by the radar.

[0024] In another example of an implementation of the invention, the robust dispersion measurement can be configured to give differentiated weighting to the points of the radar image.

[0025] In the present invention, a "robust dispersion measure" is defined as a measure that allows for the exclusion of outliers. It has been found that the exclusion of outliers does not significantly alter the reliability of the results, particularly in terms of determining the presence of life on board the vehicle.

[0026] Advantageously the spatial information of the radar image includes in particular the distance (or range) and the orientation between the target and the radar.

[0027] Advantageously, in general, the process treats moving points.

[0028] Advantageously the radar allows Doppler type measurements to measure the speed of targets.

[0029] The system is notably installed on board a motor vehicle.

[0030] The system is specifically configured to detect the presence of a child in the vehicle, based on the classification.

[0031] The invention further relates to a method for detecting the presence of life inside a motor vehicle by means of a radar, the system being configured to implement the following steps: - to provide a radar image using radar, - extract a set of raw features from the radar image, these raw features being chosen from at least: • spatial information, including range, azimuth angle, elevation angle, radar image points, and / or • speed information from points in the radar image and / or • radar image point power information, - this set of raw features being processed by a robust dispersion measure and / or outlier filtering configured to remove outlier points (also called "outliers" in English), so as to obtain processed features, - provide the set of processed characteristics to a classification algorithm, in particular of the type based on machine learning (or "Machine Learning" in English), capable of delivering information on the presence or absence of life in the vehicle.

[0032] Other features, details and advantages of the invention will become clearer upon reading the following description on the one hand, and several exemplary embodiments given by way of illustration and not limitation with reference to the accompanying schematic drawings on the other hand, in which:

[0033] [Fig-1] Fig. 1 is a schematic view of a presence detection system life within a motor vehicle according to an example of an embodiment of the invention;

[0034] [Fig.2] Fig.2 is a block diagram representing the steps in determining a presence of life by the system of the [Fig.l];

[0035] [Fig.3] Fig.3 illustrates a robust suspension method used in the invention.

[0036] The features, variants and different embodiments of the invention can be combined with each other in various ways, provided they are not incompatible or mutually exclusive. others. In particular, variants of the invention may be imagined comprising only a selection of features described subsequently in isolation from the other features described, if this selection of features is sufficient to confer a technical advantage and / or to differentiate the invention from the prior art.

[0037] Figure 1 shows a system 1 for detecting the presence of life within a motor vehicle V according to an example of an embodiment of the invention, to alert on the presence of an occupant 30 (for example a forgotten child) within the vehicle V. The system 1 comprises a presence sensor, here a radar 10, and a computer 20 connected to the radar 10.

[0038] The radar 10 is configured to determine the presence of an occupant 30 within the vehicle V.

[0039] The radar 10 is placed in a central area of ​​the vehicle V, here in the center of the roof 2 of the vehicle V as shown in [Fig.1], the radar 10 can then advantageously cover all the seats of the vehicle V.

[0040] Of course, in the invention, several radars can be used by being distributed in different locations in the passenger compartment of the vehicle V.

[0041] The presence sensor 10 is here a radar wave transmission-reception device.

[0042] The radar 10 here comprises at least one antenna (not shown) designed to emit electromagnetic waves and at least one sensor (not shown) designed to receive a reflected electromagnetic wave after reflection of the electromagnetic wave, and in particular after reflection from a person 30 inside the vehicle V.

[0043] The presence sensor 10 is here more specifically a radar (from the common English acronym "radio detection and ranging") using millimeter electromagnetic waves. The radar is, for example, a Doppler radar using sustained or continuous waves.

[0044] The calculator 20 includes at least one memory and at least one processor. The control unit 20 can be, for example, the electronic control unit (ECU) of vehicle V. The control unit 20 can also be a control unit dedicated to system 1. Instructions for determining the status of life are stored in memory and implemented by the processor. When implemented, these instructions enable the execution of the process described later.

[0045] The computer 20 is connected to the radar 10, in particular to control the emission of the electromagnetic wave.

[0046] The life presence detection system 1 is configured to implement the following steps: - provide a radar image by radar 10 (step 101 in [Fig.2]), - extract a set of raw Fr features from the radar image (step 102), these raw characteristics Fr containing all this information: • spatial information, namely range, azimuth angle, elevation angle, radar image points, • Speed ​​information from points in the radar image, • radar image point power information, - this set of raw features Fr being processed by a robust dispersion measure and / or outlier filtering configured to remove outlier points (also called "outliers" in English), so as to obtain processed features Ft,

[0047] - provide the set of processed features Ft to an algorithm of classification, of the type based on machine learning (or "Machine Learning" in English) (step 103), capable of delivering information on the presence or absence of life in the vehicle (step 104).

[0048] The treated characteristics Ft are called “Features” in English, in the context of the invention.

[0049] The extraction of raw features Fr from the radar image includes, in particular, mathematical transformations calculated on the image pixels. The processed features Ft allow for further processing, notably for classification. The raw features Fr are processed by a robust dispersion measurement and / or outlier filtering configured to remove points with outliers, so as to obtain processed features Ft intended for classification.

[0050] The classification algorithm makes it possible to reliably provide information on the presence or absence of life in the vehicle based on the raw Fr characteristics derived from radar images, in the sense that robust dispersion measurement and / or filtering of outliers makes it possible to exclude false responses (i.e., to avoid false positives that indicate the presence of life when there is no life on board the vehicle). Thus, the invention improves performance in detecting a person and improves the determination of false responses.

[0051] The invention makes it possible, in particular, to eliminate outliers or false detections, due for example to measurement imperfections, by removing them. An example of a false detection could be a water bottle with moving water or the movement of one or more objects in the passenger compartment. The classification performed by the algorithm, which is based on machine learning, is therefore more robust and reliable. Machine learning can thus be carried out without integrating the values aberrants, which improves machine learning and makes classification more reliable. The invention reduces the rate of misclassification.

[0052] The invention also makes it possible to reduce the amount of data to be processed thanks to these robust dispersion and / or filtering measures. This reduces the computing capacity requirements.

[0053] Machine learning uses artificial intelligence.

[0054] The robust dispersion measure is based on the interquartile range (IQR). In this case, the data used for classification come only from the second and third quartiles (see [Fig. 3]). In the example described, outliers below the minimum value (Q1 - 1.5 IQR) are excluded, and outliers above the maximum value (Q3 + 1.5 IQR) are excluded.

[0055] Alternatively, the robust dispersion measure is based on the mean absolute deviation (MAD for "mean absolute deviation").

[0056] We have MAD=median(IXi-Xml) where Xm=median(X) and the term “median” refers to the arithmetic mean operation.

[0057] The set of treated features obtained by the robust dispersion measure is used as primary inputs in the classification algorithm (also called "classifier").

[0058] Robust dispersion measurement is applied to all attributes measured by the radar, including power, range, speed (also called Doppler), azimuth angle and elevation angle.

[0059] When these data processed by the robust dispersion measurement are introduced into a classifier, potentially with other data, the invention makes it possible to obtain performance superior to that of conventional statistical measurements commonly used, by offering better accuracy and greater robustness to noise.

[0060] The robust dispersion measure is applied after the calculation or obtaining of the raw characteristics Fr.

[0061] In another example of an implementation of the invention, outlier filtering is applied before calculating the treated characteristics Ft, and then the treated characteristics Ft are calculated by a conventional dispersion measurement.

[0062] This makes it possible to improve conventional dispersion measurements by applying a filtering of outliers before calculating the processed characteristics Ft. This method shows superior performance to that obtained using only processed characteristics from a conventional dispersion measurement.

[0063] The classification algorithm allows detection of the presence of children and / or location / classification of occupants in the vehicle.

[0064] The invention makes it possible to give a more precise representation of the scene observed by the radar 10.

[0065] The IQR is defined as the difference between the first quartile and the third quartile and comprises 50% of the data. The MAD is the median of the absolute deviations from the median. These two methods are more robust to outliers and are therefore suitable for data containing noise, such as radar data.

[0066] In addition, an outlier filter can be applied to the range radar / Doppler data.

[0067] In another embodiment of the invention, the robust dispersion measurement can be configured to give differentiated weighting to the points of the radar image.

[0068] A "robust dispersion measure" is a measure that allows the exclusion of outliers. It has been found that the exclusion of outliers does not significantly alter the reliability of the results, particularly in terms of determining whether there is life on board the vehicle.

[0069] The spatial information of the radar image includes, in particular, the distance (or range) and orientation between the target and the radar 10.

[0070] In general, the process treats moving points.

[0071] The radar 10 allows Doppler type measurements to measure the speed of targets.

Claims

Demands

1. A system (1) for detecting the presence of life within a motor vehicle (V) by means of a radar (10), the system (1) being configured to implement the following steps: - provide a radar image by the radar (10), - extract a set of raw features (Fr) from the radar image, these raw features (Fr) being selected in particular from at least: • spatial information, including range, azimuth angle, elevation angle, points of the radar image, and / or • velocity information of points of the radar image and / or • power information of points of the radar image, - this set of raw features (Fr) being processed by a robust dispersion measurement and / or outlier filtering configured to remove outlier points, so as to obtain processed features (Ft),- to provide the set of processed features (Ft) to a classification algorithm, particularly one based on machine learning, capable of delivering information on the presence or absence of life in the vehicle.

2. System (1) according to the preceding claim, wherein the raw feature set (Fr) is processed by the robust dispersion measure which is based on the interquartile range (IQR).

3. System (1) according to claim 1, wherein the raw feature set (Fr) is processed by the robust dispersion measurement which is based on the mean absolute deviation (MAD).

4. System (1) according to any one of the preceding claims, wherein the raw feature set (Fr) is processed by the robust dispersion measure and wherein the processed feature set (Ft) obtained by the robust dispersion measure is used as primary inputs in the classification algorithm.

5. System (1) according to any one of the preceding claims, wherein the raw feature set (Fr) is processed by measuring robust dispersion that is applied to all attributes measured by the radar, including power, range, speed, azimuth angle and elevation angle.

6. System (1) according to any one of the preceding claims, wherein the raw feature set (Fr) is processed by the robust dispersion measure which is applied after the calculation of the raw features (Fr).

7. System (1) according to claim 1, wherein outlier filtering is applied before calculating the treated characteristics (Ft), and then calculating the characteristics by a conventional dispersion measurement.

8. System (1) according to any one of the preceding claims, mounted on board a motor vehicle.

9. System (1) according to any one of the preceding claims, configured to detect the presence of a child in the vehicle, based on the classification.

10. A method for detecting the presence of life within a motor vehicle (V) by means of a radar, the system being configured to implement the following steps: - providing a radar image by means of a radar (10), - extracting a set of raw features (Fr) from the radar image, these raw features (Fr) being selected in particular from at least: • spatial information, in particular range, azimuth angle, elevation angle, points of the radar image, and / or • velocity information from points of the radar image and / or • power information from points of the radar image, - this set of raw features (Fr) being processed by a robust dispersion measurement and / or outlier filtering configured to remove outlier points, so as to obtain processed features (Ft),- provide the set of processed features (Ft) to a classification algorithm, particularly one based on a, machine learning, capable of delivering information on the presence or absence of life in the vehicle.

Images