System for detecting the presence of life inside a motor vehicle

Abstract

The invention relates to a system (1) for detecting the presence of life inside a motor vehicle by means of a radar (10), the system being configured to carry out the following steps: - processing, in particular by means of a presence detection algorithm (ALP), data from the radar in order to provide a series of results (Xi) relating to presence of life in the vehicle; - determining, for each of the moving average windows (MW) at a given time (ti), a confidence level (t); - generating a decision regarding presence of life in the vehicle in the case where the confidence level is greater than a threshold, and; - in the event of instability in the confidence level, ignoring the decision regarding presence of life in the vehicle; - or, in the event of stability in the confidence level, providing the decision regarding presence of life in the vehicle.

Description

[0001] DESCRIPTION

[0002] Title: Vehicle Life Presence Detection System

[0003] [1] The invention relates to a system for detecting the presence of life within a motor vehicle.

[0004] [2] It is known to detect the presence of life, particularly a person inside a motor vehicle, followed by an alert if a person is left inside the vehicle. Such detection and alert systems enhance passenger safety, especially by preventing the danger of leaving a child inside the vehicle. Such detection can, for example, utilize radar located inside the vehicle.

[0005] [3] In particular, there is a need to prevent false alarms. Such a false alarm can occur when a person outside the vehicle causes vibrations, for example by leaning on it. Another situation can occur, such as when the vehicle door is closed; the slamming of the door will create vibrations that, if detected, can trigger false alarms.

[0006] [4] The radar can then erroneously detect the presence of life in the vehicle when no one is there.

[0007] [5] A false alarm can also occur when strong wind creates vibrations inside the vehicle or when someone or something comes against the vehicle, causing the radar to move.

[0008] [6] The invention aims in particular to remedy these difficulties.

[0009] [7] The invention thus relates to a system for detecting the presence of life within a motor vehicle by means of a radar, the system being configured to implement the following steps:

[0010] - to process, in particular by a presence detection algorithm, especially of the machine learning type, radar data to provide a succession of results on the presence of life (indicated for the remainder of the description by the simplified term "result") in the vehicle, each result being associated with a frame, these results being notably of value 1 in the case of a positive result defining the presence of life, or of 0 in the case of a negative result defining the absence of life, to determine, for each of the moving average windows at a given time, a confidence rate associated in particular with the most recent frame entering this moving average window, the confidence rate being the ratio of the number of frames with a positive result to the total number of frames (N frames) in the moving average window considered,

[0011] - to measure the stability of the confidence rate over time, frame by frame,

[0012] - generate a decision regarding the presence of life in the vehicle if the confidence level exceeds a threshold, and

[0013] o In case of instability in the confidence level, ignore the decision regarding the presence of life in the vehicle.

[0014] or, in the case of stable confidence levels, provide the decision regarding the presence of life in the vehicle.

[0015] [8] It has been observed that a confidence level that fluctuates unsteadily is linked to a false alarm regarding the presence of life in the vehicle. The evolution of the confidence level over a series of frames can be represented by a curve as a function of time.

[0016] [9] Thanks to the invention, measuring the stability of the confidence level over time in the life presence detection system represents a significant advance in improving the reliability and accuracy of this type of system. Measuring the stability of the confidence level over time ensures that decisions made based on the confidence level are more reliable, thus reducing the risk of errors (false positives) and improving safety.

[0017]

[0010] For example, the invention allows for improved reliability, or in other words a robustification, of the system for detecting the presence of children and for locating / classifying occupants by reducing false positives.

[0018]

[0011] Preferably, the system operates by analyzing a series of confidence rate values ​​obtained from individual classifiers in each frame. A stability metric (as described below) indicates the stability of the confidence rate. Stability is compared to a threshold, and the confidence rate is concluded to be:

[0019] - not stable if the stability metric is greater than the threshold, or

[0020] - stable if the stability metric is less than or equal to the threshold.

[0021]

[0012] The final decision based on the confidence rate is issued only if the confidence rate is considered stable.

[0013] According to one aspect of the invention, the confidence rate is updated as the moving average window slides over time, so that the confidence rate is calculated on the basis of a series of frames specific to each moving average window.

[0022]

[0014] Each moving average window is composed of N frames. At each iteration for determining the confidence level of a new moving average window of N frames, the result relating to the oldest frame of the previous moving average window is replaced by the result for the new frame.

[0023]

[0015] As long as a moving average window comprises fewer than N frames, no frame replacement takes place between two successive moving average windows.

[0024]

[0016] According to one aspect of the invention, the method processes, over time, a succession of radar images to produce a succession of positive or negative results. Each radar image defines a frame. Thus, in the invention, several radar images will be used to ignore or provide the decision regarding the presence of life in the vehicle.

[0025]

[0017] According to one aspect of the invention, the measurement of the stability of the confidence rate over time is triggered when the confidence rate reaches or exceeds a threshold which is for example 60% (preferably 70%, even more preferably 80%).

[0026]

[0018] According to one aspect of the invention, the measurement of the stability of the confidence rate over time is carried out by following the evolution of the confidence rate in a sliding stability detection window.

[0027]

[0019] Thus, the moving average window (for example, N frames) allows the confidence level to be calculated within this moving average window, and the stability detection window (on M frames) allows the stability of the confidence level thus calculated (with N frames) to be monitored. The values ​​N and M can be identical or different.

[0028]

[0020] In an example of the invention, for a moving average window of N frames, at time tN when the positive or negative result for the Nth frame is obtained, the confidence level r(t w ) is defined by the following equation:

[0029]

[0021] T(t w ) = Si=^

[0030]

[0022] X(ti) defines the value 1 in the case of a positive result defining the presence of life, or 0 in the case of a negative result defining the absence of life, at time L.

[0023] Similarly generalized for any moving average window at a later time tN or later tN+j, where "j" is an integer greater than or equal to 0, the confidence level T

[0031]

[0032] is defined by the following equation:

[0033]

[0034]

[0025] When the number of frames is less than that defining the dimension of a moving average window, namely ti <T N, each of the missing frames to form a complete moving average window is considered to have a negative result (0) defining an absence of life.

[0035]

[0026] The number of N frames is, for example, 20 frames, preferably 60 frames. For the confidence level, a larger moving average window size, with a larger number N of frames, allows for better smoothing at the cost of a relative delay in the first determination of the confidence level.

[0036]

[0027] The time required to perform the stability measurement corresponds to the duration of the stability detection window, for the acquisition of M frames, for example the number M corresponds to 15 frames).

[0037]

[0028] In the case of stable confidence levels, the decision regarding the presence of life in the vehicle is concluded and provided only after this succession of M successive moving average windows, relative to M successive frames, defining a stability detection window. The number of windows can, for example, be between 5 and 30, particularly between 10 and 20, and more particularly 15.

[0038]

[0029] According to one aspect of the invention, the measurement of the stability of the confidence rate over time, in particular in the stability detection window, is carried out by detecting one or more drops in the confidence rate frame by frame, and in the event of such detections of drops, it is concluded that there is a case of instability of the confidence rate and the decision of presence of life in the vehicle is ignored.

[0039]

[0030] For example, as soon as a downward trend is detected in the use case involving false positives, instability is detected and presence detection will not be provided, thus avoiding false detection.

[0040]

[0031] The downward trend corresponds, for example, to a decrease in the confidence level over several successive frames. Instability is considered to exist, for instance, when a certain number of successive frames are detected with a decrease in the confidence level, this number being greater than a threshold. The threshold can, for example, be between 2 and 10 frames, preferably greater than or equal to 2 and less than or equal to 5, and even more preferably equal to 3 successive frames, with a decrease in the confidence level.

[0032] The method thus focuses on monitoring the confidence level of presence within a mobile stability detection window of M successive frames. Monitoring the confidence level within the stability window begins when the confidence level (of the previous frame) increases above a certain threshold.The criterion for judging stability could, for example, be the sum of the number of declines in the confidence level within the stability detection window or over a given time period. This stems from the fact that most false positives tend to decline after the detection of a very limited number of consecutive positive results.

[0041]

[0033] According to one aspect of the invention, the measurement of the stability of the confidence rate over time is carried out, in particular by artificial intelligence, by determining from the aspect of the curve followed by the succession of consecutive frames, whether this curve is stable or unstable, in particular on the basis of a history of confidence rate frame curves.

[0042]

[0034] The invention is thus based on a scheme (or “pattern” in English) of stability / instability of the confidence rate curve, frame by frame.

[0043]

[0035] In the event of detection of an unstable curve, the decision of presence of life in the vehicle is ignored, even if the confidence rate remains above the threshold (for example 60%).

[0044]

[0036] Below the threshold, the measurement of the stability of the confidence rate over time is not triggered, and no decision on the presence of life in the vehicle is issued.

[0045]

[0037] In another embodiment of the invention, the measurement of the stability of the confidence rate over time is based on a statistical measure applied to the confidence rates related to a succession of consecutive frames.

[0046]

[0038] In this case, we speak of a statistical stability metric, which can be applied to both positive and negative results.

[0047]

[0039] In one embodiment of the invention, the statistical measurement uses a statistical stability metric calculated using the standard deviation of the confidence level over the stability detection window defined by the last M frames (STD approach or in English “Standard Deviation”),

[0048]

[0040] In this approach, the standard deviation of the confidence rate values ​​for the previous M frames is calculated. A high standard deviation value will reflect greater instability of the confidence rate, while a low standard deviation value will reflect stability of the confidence rate.

[0041] In an example of an embodiment of the invention, when the standard deviation value increases over a number of consecutive frames within a stability detection window (resulting in M ​​consecutive moving average windows), this number being greater than a threshold (for example, 3), one concludes that there is instability of the confidence rate.

[0049]

[0042] The standard deviation is written as follows (M being the number of frames in the stability detection window, T being the confidence level for frame i, and T~ being the average of the confidence levels over the stability detection window):

[0050]

[0051]

[0043] In another embodiment of the invention, the statistical measurement uses a statistical stability metric calculated using the exponential moving standard deviation of the confidence level with a parameter lambda (EMSTD approach or in English

[0052] “Exponential Moving Standard Deviation”).

[0053]

[0044] This approach is similar to the standard deviation (STD) approach, except that exponential moving standard deviation (EMSTD) gives greater weight to confidence rates from moving average windows associated with the most recent frames for calculating the standard deviation, with a lambda parameter to control the forgetting weight for the history.

[0054]

[0045] The invention further relates to a method for detecting the presence of life within a motor vehicle by means of a radar, the method comprising the following steps: - processing, in particular by a presence detection algorithm, in particular of the machine learning type, radar data to provide a succession of results on the presence of life in the vehicle, each result being associated with a frame, these results being in particular of value 1 in the case of a positive result defining the presence of life, or of 0 in the case of a negative result defining the absence of life,

[0055] determine, for each moving average window at a given time, a confidence level associated with the most recent frame entering that moving average window, the confidence level being the ratio of the number of frames with a positive result to the total number of frames (N frames) in the considered moving average window,

[0056] - to measure the stability of the confidence rate over time, frame by frame,

[0057] - generate a decision regarding the presence of life in the vehicle if the confidence level is above a threshold, and / or ignore the decision regarding the presence of life in the vehicle if the confidence level is unstable.

[0058] or, in the case of stable confidence levels, provide the decision regarding the presence of life in the vehicle.

[0059]

[0046] The stability measure therefore assesses the stability of the confidence level over a set of observations (from the radar) collected over time. If the confidence level is found to be stable, the presence decision is provided.

[0060]

[0047] Other features, details and advantages of the invention will become clearer upon reading the following description on the one hand, and several illustrative and non-limiting examples of embodiments given with reference to the accompanying schematic drawings on the other hand, in which:

[0061]

[0048] [Fig 1] Figure 1 is a schematic view of a life presence detection system within a motor vehicle according to an example of an embodiment of the invention;

[0062]

[0049] [Fig. 2] Figure 2 is a block diagram representing the steps of determining the presence of life by the system of Figure 1;

[0063]

[0050] [Fig. 3] Figure 3 represents two graphs corresponding to a first example of a stability metric used in the invention, for the case of true positives and the case of false positives,

[0064]

[0051] [Fig. 4] Figure 4 represents two graphs corresponding to other examples of stability metrics used in the invention, for the case of true positives and the case of false positives.

[0065]

[0052] The features, variants, and different embodiments of the invention can be combined in various ways, provided they are not incompatible or mutually exclusive. In particular, variants of the invention may be conceived comprising only a selection of features, described hereafter in isolation from the other described features, if this selection of features is sufficient to confer a technical advantage and / or to differentiate the invention from the prior art.

[0066]

[0053] Figure 1 shows a life presence detection system 1 within a motor vehicle V according to an embodiment of the invention, for alerting to 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.

[0054] The radar 10, which is a radar wave transmitting and receiving device, is configured to determine the presence of an occupant 30 within the vehicle V.

[0067]

[0055] 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 figure 1, the radar 10 can then advantageously cover all the seats of the vehicle V.

[0068]

[0056] Of course, in the invention, several radars can be used by being distributed in different locations in the passenger compartment of vehicle V.

[0069]

[0057] The radar 10 here includes at least one antenna (not shown) designed to emit the electromagnetic wave 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 on a person 30 inside the vehicle V.

[0070]

[0058] The radar 10 uses millimeter electromagnetic waves. The radar is, for example, a Doppler radar using sustained or continuous waves.

[0071]

[0059] The control unit 20 comprises at least one memory and at least one processor. The control unit 20 may be, for example, the electronic control unit (ECU) of vehicle V. The control unit 20 may also be a control unit dedicated to system 1. Instructions enabling the determination of the live status are stored in the memory and implemented by the processor. When implemented, these instructions enable the execution of the process described later.

[0072]

[0060] As illustrated in Figure 2, system 1 is configured to implement the following steps:

[0073] - to process, using a machine learning-type ALP presence detection algorithm, radar data 10 to provide a series of results on the presence of life in the vehicle, each result Xi being associated with a frame, these results being notably of value 1 in the case of a positive result defining the presence of life, or of 0 in the case of a negative result defining the absence of life (step S1),

[0074] To determine, for each moving average window MW at a given time ti, a confidence level T associated with the frame Xi, the most recent frame entering this moving average window MW, the confidence level T being the ratio of the number of frames with a positive result to the total number N of frames in the considered moving average window MW; we therefore calculate a moving average of the confidence level T over this moving average window MW (step S2), - measure the stability of the confidence level T over time, frame after frame (step S3),

[0075] - generate a decision regarding the presence of life in the vehicle if the confidence level T is above a threshold (step S4), and

[0076] o In case of instability of the confidence level T, ignore the decision regarding the presence of life in the vehicle (step S5),

[0077] or, in the case of stability of the confidence rate T, provide the decision of presence of life in the vehicle (step S6).

[0078]

[0061] Each result Xi (1 or 0) is associated with a frame corresponding to a given acquisition time ti, where i is an integer greater than or equal to 1, and each acquisition time ti defines a frame. For example, a result Xi is obtained at an acquisition time ti (corresponding to the first frame) and a result X2 is obtained at a time t2 (corresponding to the first frame).

[0079] Preferably, the time period between two acquisition times ti and ti+i is always identical.

[0080]

[0062] In one embodiment, the radar data 10, used to provide a series of results Xi, comes from radar images from which static objects have been removed using a static object removal technique (also known as "Clutter Removal"), in order to retain moving objects (for example, a breathing child). From these pre-processed radar images, features such as data representing the shape and / or dispersion and / or position and / or velocity of moving objects in the pre-processed radar images are extracted. Then, based on these features, the presence detection algorithm provides a raw result indicating the presence or absence of life in the vehicle.

[0081]

[0063] System 1 operates by analyzing a series of confidence rate values ​​T obtained from individual classifiers in each frame. A stability metric, as described below, indicates the stability of the confidence rate. Stability is compared to a threshold, and the confidence rate is concluded to be:

[0082] - not stable if the stability metric is greater than the threshold, or

[0083] - stable if the stability metric is less than or equal to the threshold.

[0084]

[0064] The final decision based on the confidence rate is issued only if the confidence rate is considered stable (step S6).

[0085]

[0065] The confidence rate T is updated as the moving average window MW slides over time, so that the confidence rate is calculated on the basis of a series of frames specific to each moving average window MW.

[0086]

[0066] The method processes, over time, a succession of radar images to produce a succession of positive or negative results. Thus, in the invention, several radar images will be used to ignore or provide the decision regarding the presence of life in the vehicle.

[0087]

[0067] In a first embodiment of the invention, the measurement of the stability of the confidence rate over time is triggered when the confidence rate T reaches or exceeds a threshold which is for example 60%, preferably 70%, even more preferably 80%.

[0088]

[0068] The stability of the confidence level T over time is measured by monitoring the evolution of the confidence level within a sliding stability detection window SW. Thus, the moving average window MW (for example, N frames) allows the confidence level to be calculated within this moving average window, and the stability detection window SW (over M frames) allows the stability of the confidence level T thus calculated (with N frames) to be monitored. The values ​​N and M can be identical or different.

[0089]

[0069] In an example of the invention, for a moving average window of N frames, at time tN when the positive or negative result for the Nth frame is obtained, the confidence level r(t w ) is defined by the following equation:

[0090]

[0070] T(t w ) = Si=^

[0091]

[0071] X(ti) defines the value 1 in the case of a positive result defining the presence of life, or 0 in the case of a negative result defining the absence of life, at time L

[0092]

[0072] Similarly, in a generalized manner for any moving average window at a later time tN or later tN+j, where "j" is an integer greater than or equal to 0, the confidence level T

[0093]

[0094] is defined by the following equation:

[0095]

[0096]

[0074] When the number of frames is less than that defining the dimension of a moving average window, namely ti <T N , each of the missing frames to form a complete moving average window is considered to have a negative result (equal to 0) defining an absence of life.

[0097]

[0075] The number of N frames is, for example, 20 frames, preferably 60 frames. For the confidence level, a larger moving average window size, with a larger number N of frames, allows for better smoothing at the cost of a relative delay in the first confidence level.

[0076] The time required to perform the stability measurement corresponds to the duration of the stability detection window; for the acquisition of the M frames, for example, the number M corresponds to 15 frames.

[0098]

[0077] In the case of stable confidence levels, the decision regarding the presence of life in the vehicle is concluded and provided only after this succession of M successive moving average windows, relative to M successive frames, defining a stability detection window. The number of windows can, for example, be between 5 and 30, particularly between 10 and 20, and more particularly 15.

[0099]

[0078] The measurement of the stability of the confidence rate T over time, in the SW stability detection window, is carried out by detecting one or more drops in the confidence rate T frame by frame, and in the event of such detections of drops, it is concluded that there is a case of instability of the confidence rate T and the decision of presence of life in the vehicle is ignored (step S5).

[0100]

[0079] For example, as can be seen in the graph called Graph 1B in Figure 3, as soon as the downward trend is detected in the case involving false positives, the instability is detected and the presence detection will not be provided, thus avoiding a false detection. Graphs 1A and 1B in Figure 3 illustrate the evolution of the confidence level (on the y-axis) over time (on the x-axis).

[0101]

[0080] A downward trend corresponds, for example, to a decrease in the confidence level T over several successive frames, giving rise, for example, to oscillations. Instability is considered to exist, for example, when a certain number of successive frames with a decrease in the confidence level is detected, this number being greater than a threshold. The threshold may, for example, be between 2 and 10 frames, preferably greater than or equal to 2 and less than or equal to 5, and even more preferably equal to 3 successive frames, with a decrease in the confidence level.

[0102]

[0081] The method thus focuses on monitoring the confidence level of presence within a mobile stability detection window SW of M successive frames. Monitoring the confidence level within the stability window SW begins when the confidence level (of the previous frame) rises above a certain threshold. The criterion for judging stability can, for example, be the sum of the number of drops in the confidence level within the stability detection window SW or over a given time range. This stems from the fact that most false positives tend to decrease after the detection of a very limited number of consecutive positive results.

[0082] The measurement of the stability of the confidence rate T over time is carried out, in particular by artificial intelligence, by determining from the aspect of the curve followed by the succession of consecutive frames, whether this curve is stable or unstable, in particular on the basis of a history of confidence rate frame curves.

[0103]

[0083] The invention is thus based on a scheme (or “pattern” in English) of stability / instability of the confidence rate curve, frame by frame.

[0104]

[0084] In the event of detection of an unstable curve, the decision of presence of life in the vehicle is ignored, even if the confidence rate remains above the threshold (for example 60%).

[0105]

[0085] Below the threshold, the measurement of the stability of the confidence rate over time is not triggered, and no decision on the presence of life in the vehicle is issued.

[0106]

[0086] In an example embodiment of the invention, in a case where we consider a number N of frames equal to 20, for example, if for 12 of the frames we obtain a positive result on the 20 frames, we therefore have a 60% confidence rate in this sliding moving average window over the 20 frames.

[0107]

[0087] If a stable curve is detected (see Graph 1A in Figure 3), particularly if the confidence level does not decrease, the decision regarding the presence of life in the vehicle is provided. Graph 1A shows a confidence level T that increases before reaching a constant plateau. The curve in Graph 1A represents the case of a true positive result, and the curve in Graph 1B represents the case of a true negative result.

[0108]

[0088] In two other embodiments of the invention with reference to the curves in Figure 4, the measurement of the stability of the confidence level T over time is based on a statistical measure applied to the confidence value frames. The evolution of the confidence level over time defines a succession of frames.

[0109]

[0089] In this case, we speak of a statistical stability metric, which can be applied to both positive and negative results.

[0110]

[0090] In an example of an embodiment of the invention relating to curves C2A and C2B of graphs GRAPH 2A and GRAPH 2B in Figure 4, the statistical measurement uses a statistical stability metric calculated using the standard deviation of the confidence level over the last M frames (STD approach, or "Standard Deviation"), where M corresponds, for example, to 15 frames on these curves. The curves of graph GRAPH 2A represent the case of a true positive ("True Positive" or TP), and the curves of graph GRAPH 2B represent the case of a true negative ("True Negative" or TN). Curves C1A and C1B are the curves shown in Figure 3.

[0111]

[0091] In this approach, the standard deviation of the confidence rate values ​​within the preceding M frames is calculated. A high standard deviation value will reflect greater instability of the confidence rate, while a low standard deviation value will reflect greater stability of the confidence rate.

[0112]

[0092] In an example of an embodiment of the invention, when the value of the standard deviation increases over a number of consecutive frames in a stability detection window, this number being greater than a threshold, we conclude that there is an instability of the confidence rate (C2B curve of the GRAPH 2B graph), which gives a "true negative" or TN.

[0113]

[0093] We can see in Figure 4 that the stability metric remains high in the TN case when the confidence level is unstable, thus preventing an erroneous decision from being produced. The stability metric decreases as soon as the confidence level reaches a stable state, ready to provide the final decision.

[0114]

[0094] The standard deviation is written as follows (M being the number of frames in the stability detection window, T being the confidence rate for frame i, and T~ being the average of the confidence rates over the stability detection window):

[0115]

[0116] In the implementation of the invention in relation to curves C3A and C3B of Graph 2A and Graph 2B, the statistical measurement uses a statistical stability metric calculated using the exponential moving standard deviation of the confidence level with a parameter lambda (EMSTD approach or "Exponential Moving Standard Deviation"). The value of lambda is taken to be 0.2 in the example of the curves.

[0117]

[0097] This approach is similar to the standard deviation (STD) approach, except that the moving exponential standard deviation (EMSTD) gives a higher weight to confidence rates from the most recent frames for calculating the standard deviation, with a lambda parameter to control the forgetting weight for the history.

[0118]

[0098] On curves C3A and C3B of Graph 2A and Graph 2B given in Figure 4, the parameter lambda to control the forgetting weight for the history is taken equal to 0.2.

[0119]

[0099] The EMSTD approach is based on the following relationships:

[0120]

[0121]

[0100]

[0122]

[0123]

[0101]

[0124]

[0102] lambda or A is a forgetting factor,

[0125]

[0103] pt is the average confidence rate at time t,

[0126]

[0104] ot is the standard deviation of the confidence rate at time t,

[0127]

[0105] rt is the residual confidence rate which corresponds to rt= xt- p(t- 1 )

[0128]

[0106] xt is the confidence rate at time t,

[0129]

[0107] p(t-1 ) is the average confidence rate at time t-1.

[0130]

[0108] We can see from curves C2B and C3B of Graph 2B that the stability metric remains high for the use case involving false positives when the confidence rate is unstable with oscillations, which makes it possible not to provide a false decision.

[0131]

[0109] For the use case involving true positives, the stability metric decreases as soon as the confidence level reaches stability, in order to issue the final decision. This can be seen in curves C2A and C2B of Graph 2A. In particular, when the standard deviation value decreases over a number of consecutive frames within a moving average window, this number being greater than a threshold, the confidence level is considered stable.

Claims

DEMANDS

1. System (1) for detecting the presence of life inside a motor vehicle by means of a radar (10), the system being configured to implement the following steps: - to process, in particular by a presence detection algorithm (ALP), especially of the machine learning type, radar data to provide a succession of results (Xi) on the presence of life in the vehicle, each result being associated with a frame, these results being notably of value 1 in the case of a positive result defining the presence of life, or of 0 in the case of a negative result defining the absence of life, determine, for each moving average (MW) window at a given time (ti), a confidence level (T) associated in particular with the most recent frame entering this moving average window, the confidence level being the ratio of the number of frames with a positive result to the total number of frames (N) in the moving average (MW) window considered, - measure the stability of the confidence level over time, frame by frame, - generate a decision regarding the presence of life in the vehicle if the confidence level exceeds a threshold, and o In case of instability in the confidence level, ignore the decision regarding the presence of life in the vehicle. or, in the case of stable confidence levels, provide the decision regarding the presence of life in the vehicle.

2. A system according to the preceding claim, configured to implement a step in which the confidence rate (T) is updated as the moving average window (MW) slides over time, so that the confidence rate is calculated on the basis of a series of frames specific to each moving average window.

3. System according to the preceding claim, configured to implement a step in which the measurement of the stability of the confidence rate (T) over time is triggered when the confidence rate reaches or exceeds a threshold which is for example 60%, preferably 70%, even more preferably 80%.

4. System according to any one of the preceding claims, configured to implement a step in which the measurement of the stability of the confidence rate (T) over time is carried out by tracking the evolution of the confidence rate in a stability detection window (SW) which is sliding.

5. A system according to any one of the preceding claims, configured to implement a step in which the measurement of the stability of the confidence rate over time, in particular within the stability detection window (SW), is carried out by detecting one or more drops in the confidence rate frame by frame, and in the event of such detections of drops, a case of confidence rate instability is concluded and the decision of presence of life in the vehicle is ignored.

6. System according to the preceding claim, configured to implement a step in which the measurement of the stability of the confidence rate (T) over time is carried out, in particular by artificial intelligence, by determining from the aspect of the curve followed by the succession of consecutive frames, whether this curve is stable or unstable, in particular on the basis of a history of confidence rate frame curves.

7. A system according to any one of claims 4 to 6, configured to implement a step in which the measure of the stability of the confidence rate (T) over time is based on a statistical measure applied to the confidence rates related to a succession of consecutive frames.

8. System according to the preceding claim, configured to implement a step in which the statistical measurement uses a statistical stability metric calculated using the standard deviation of the confidence rate over the last M frames.

9. System according to claim 7, configured to implement a step in which the statistical measurement uses a statistical stability metric calculated using the exponential moving standard deviation of the confidence rate with a parameter lambda.

10. A method for detecting the presence of life inside a motor vehicle using radar (10), the method comprising the following steps: - to process, in particular by means of a presence detection algorithm, in particular of the machine learning type, radar data (10) to provide a succession of results on the presence of life in the vehicle, each result being associated with a frame, these results (Xi) being in particular of value 1 in the case of a positive result defining a presence of life, or of 0 in the case of a negative result defining an absence of life, determine, for each moving average window at a given time (ti), a confidence rate (T) associated in particular with the most recent frame entering this moving average window, the confidence rate being the ratio of the number of frames with a positive result to the total number of frames (N frames) in the moving average window considered, - measure the stability of the confidence level (T) over time, frame by frame, - generate a decision regarding the presence of life in the vehicle if the confidence level is above a threshold, and o In case of instability in the confidence level, ignore the decision regarding the presence of life in the vehicle. or, in the case of stable confidence levels, provide the decision regarding the presence of life in the vehicle.

Images