DEVICE AND METHOD FOR CONTROLLING A VEHICLE AIRBAG

Abstract

A device (202) for controlling a vehicle airbag, wherein the device comprises a human injury probability calculation unit (206), a learning unit (208) and an airbag deployment detection unit (210), characterized by the fact that the human injury probability calculation unit (206) is set up to calculate a human injury condition probability and a human injury prediction probability based on vehicle movement information measured by a detection device (250), the learning unit (208) is set up to calculate a post-human injury probability by performing real-time probabilistic machine learning based on the human injury condition probability and the human injury prediction probability, and the airbag deployment determination unit (210) is set up to determine whether the airbag should be deployed based on the post-injury probability of the person.

Description

BACKGROUND 1. Area

[0001] The invention relates to a vehicle, and in particular to controlling the triggering or deployment of an airbag in the vehicle. 2. Description of the related technology

[0002] An airbag is a device designed to protect occupants from impact in a vehicle collision and, along with the seatbelt, is a typical occupant protection feature in a vehicle. When a vehicle impact is detected by a sensor, an actuating gas device (gas generator) is ignited, and the airbag is immediately inflated (deployed) by an explosive gas to protect the occupants. The shorter the time between the vehicle impact and the deployment of the airbag, the better.

[0003] However, it is necessary to determine the airbag deployment based on whether the impact caused by the vehicle collision is severe enough to require airbag deployment, or whether the impact is so severe that deployment is unnecessary. If the airbag fails to deploy in a situation where it is required to protect the occupant, the occupant cannot be protected. Conversely, it is undesirable for the airbag to deploy even in the event of an impact of such magnitude that deployment is unnecessary, as this can lead to costs associated with repositioning the airbag (e.g., replacement, repair).

[0004] In other words, it is necessary to determine precisely whether the airbag should be deployed according to the degree of injury suffered by the occupant in a vehicle collision.

[0005] For example, devices and methods for controlling an airbag with the features according to the preambles of the independent claims are known from DE 10 2012 222 382 A1. Further devices and methods for controlling an airbag are known, for example, from DE 10 2013 113 984 A1 and DE 10 2018 125 638 A1, respectively. BRIEF EXPLANATION

[0006] The object of the invention is to provide an airbag control device and method that are capable of ensuring the robustness of an airbag triggering or deployment logic (hereinafter referred to as: airbag triggering logic) and of protecting occupants more effectively by determining whether an airbag should be deployed based on a post-injury probability of the person, which is calculated using a model for the probability of injury to the person (e.g., occupant such as driver or passenger) and a Bayesian network or network (feedback learning).

[0007] Further aspects of the invention are partly set out in the following description and should partly be evident from the description or can be learned through the practice of the invention.

[0008] According to one aspect of the invention, a device for controlling a vehicle airbag is provided. The device comprises: a human injury probability calculation unit configured to calculate a human injury conditional probability (e.g., the probability of a condition occurring in which, for example, a head, neck, or chest may be injured) and a human injury prediction probability (e.g., a prediction of whether a human injury will occur; hereinafter referred to as human injury prediction probability) based on vehicle motion information measured by a detection device; and a learning unit configured to calculate a post-result human injury probability."post-human injury probability" (e.g., an injury probability as a result of a calculation; hereinafter referred to as post-injury probability), by performing probability-based machine learning with real-time feedback based on the human injury condition probability and the human injury prediction probability, and an airbag deployment determination unit set up to determine, based on the post-human injury probability, whether the airbag should be triggered or deployed.

[0009] The learning unit can be set up to calculate a pre-human injury probability (e.g., an injury probability as a result of an estimate or an intermediate calculation step; hereinafter referred to as pre-injury probability) based on the human injury condition probability and the human injury prediction probability, and to calculate the post-human injury probability by multiplying the human injury condition probability with the pre-human injury probability.

[0010] Real-time, probabilistic machine learning can be set up to update the pre-injury probability of a human by relating a current (e.g., from a recently performed computation cycle) post-injury probability of a human to a previous (e.g., from a previously performed computation cycle) pre-injury probability of a human.

[0011] The human injury condition probability can be set up to be calculated by the following equation 1 below. P(xt|ut,xt−1),P(zt|xt)

[0012] The expression P(x t | u t , x t-1 ) represents the human injury prediction probability at a current time (t) according to the measured values ​​of the impact sensors and a previous (t-1) probability of human injury.

[0013] The expression P(z t | x t ) represents the human injury prediction probability, which is predicted based on the injuries of occupants measured / obtained by the simulation.

[0014] The expression x t represents the actual probability of injury to the human being for each of the six areas of the head, neck and chest (e.g. frontal and lateral impact injury for head, neck and chest) at the current time (t).

[0015] The expression u t represents the measured value of the detection device 250 at the current time (t).

[0016] The expression x t-1 represents the actual probability of injury to the person at the previous time (t-1).

[0017] The expression trepresents the human injury prediction probability for each of the six areas of the head, neck, and chest at the current time (t).

[0018] The pre-injury probability of a person can be set up to be calculated using Equation 2 below. (bel(x))¯=∫P(xt|ut,xt−1)bel(xt−1)dxt−1

[0019] The expression P(x t | u t ,x t-1 ) represents the human injury prediction probability according to the measured values ​​of the impact sensors and the current time (t) according to the previous (t-1) human injury probability.

[0020] The expression bel(x t-1 ) represents the previous (t-1) post-injury probability of the human.

[0021] The post-injury probability of a person can be set up to be calculated using Equation 3 below. bel(x)=ηP(zt|xt)bel(x)¯

[0022] The expression η represents a normalization factor.

[0023] The expression P(z t | x t ) represents the human injury prediction probability, which is predicted according to the occupant injuries measured / obtained by the simulation.

[0024] The expression (bel(x)) represents the pre-injury probability of the human.

[0025] The airbag deployment detection unit can be configured to determine airbag deployment based on the post-injury probability of the person exceeding a predetermined reference value.

[0026] The vehicle motion information can include an acceleration value and an angular velocity value of the vehicle, a collision or impact value, a pressure value, a roll value, a pitch value, and a yaw value.

[0027] According to one aspect of the invention, a method for controlling a vehicle airbag is provided. The method comprises: calculating a human injury condition probability and a human injury prediction probability by a human injury probability calculation unit based on vehicle motion information measured by a detection device; calculating a post-human injury probability by a learning unit by performing real-time, probability-based machine learning based on the human injury condition probability and the human injury prediction probability; and determining, by an airbag deployment determination unit, whether an airbag should be deployed based on the post-human injury probability.

[0028] The procedure may further include calculating a pre-injury probability of the human being based on the injury-of-the-human-condition probability and the injury-of-the-human-prediction probability by the learning unit, and may include calculating the post-injury probability of the human being by multiplying the injury-of-the-human-condition probability with the pre-injury probability of the human being by the learning unit.

[0029] Real-time, probability-based machine learning can be set up to update the pre-injury probability of the human by tracing a current post-injury probability of the human back to a previous pre-injury probability of the human.

[0030] The human injury condition probability can be set up to be calculated by Equation 1 below. P(xt|ut,xt−1),P(zt|xt)

[0031] The expression P(x t | u t ,x t-1 ) represents the human injury prediction probability at a current time (t) according to the measured values ​​of the impact sensors and a previous (t-1) probability of human injury.

[0032] The expression P(z t | x t ) represents the human injury prediction probability, which is predicted based on the injuries of occupants measured / obtained through a simulation.

[0033] The expression x t represents the actual probability of injury to the person for each of the six areas of the head, neck and chest at the current time (t).

[0034] The expression u t represents the measured value of the detection device 250 at the current time (t).

[0035] The expression x t-1 represents the actual probability of injury to the person at the previous time (t-1).

[0036] The expression z t represents the human injury prediction probability for each of the six areas of the head, neck, and chest at the current time (t).

[0037] The pre-injury probability of a person can be set up to be calculated using Equation 2 below. (bel(x))¯=∫P(xt|ut,xt−1)bel(xt−1)dxt−1

[0038] The expression P(x t | u t ,x t-1) represents the human injury prediction probability according to the measurements of the impact sensors and the current time (t) according to the previous (t-1) human injury probability.

[0039] The expression bel(x t-1 ) represents the previous (t-1) post-injury probability of the human.

[0040] The post-injury probability of a person can be set up to be calculated using Equation 3 below. bel(x)=ηP(zt|xt)bel(x)¯

[0041] The term η represents a normalization factor.

[0042] The expression P(z t | x t ) represents the human injury prediction probability, which is predicted according to the occupant injuries measured / obtained by the simulation.

[0043] The expression (bel(x)) represents the pre-injury probability of the human.

[0044] The airbag deployment detection unit can be set up to determine airbag deployment based on the post-injury probability of the person exceeding a predetermined reference value.

[0045] The vehicle motion information can include an acceleration value and an angular velocity value of the vehicle, a collision or impact value, a pressure value, a roll value, a pitch value, and a yaw value.

[0046] According to a further aspect of the invention, a device for controlling a vehicle airbag is provided. The device comprises: a human injury probability calculation unit configured to calculate a human injury condition probability and a human injury prediction probability based on vehicle motion information measured by a sensing device; a learning unit configured to calculate a pre-injury probability based on the human injury condition probability and the human injury prediction probability; to calculate the post-injury probability by multiplying the human injury condition probability by the pre-injury probability; and to update the pre-injury probability.by feeding back a current post-injury probability of the human to a previous pre-injury probability of the human through probability-based machine learning with real-time feedback, and an airbag deployment determination unit set up to determine whether the airbag should be deployed based on the post-injury probability of the human.

[0047] According to a further aspect of the invention, a method for controlling a vehicle airbag is provided. The method comprises: calculating a human injury condition probability and a human injury prediction probability by a human injury probability calculation unit based on vehicle motion information measured by a detection device; calculating a pre-human injury probability based on the human injury condition probability and the human injury prediction probability by a learning unit; and calculating the post-human injury probability by the learning unit by multiplying the human injury condition probability by the pre-human injury probability.Updating the pre-injury probability of the human by the learning unit through feedback of a current post-injury probability of the human to a previous pre-injury probability of the human through probability-based machine learning with real-time feedback, and determining, by an airbag deployment determination unit, whether an airbag should be deployed based on the post-injury probability of the human. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] These and / or other aspects of the invention will become apparent from the following description of the embodiments in conjunction with the accompanying drawings and will be easier to understand, in which: Fig. 1 is a view showing a vehicle according to one embodiment. Fig. 2 is a view showing a device for controlling an airbag according to one embodiment. Fig. Figure 3 is a view showing a method for controlling the triggering / deployment of an airbag in a vehicle according to one embodiment. Fig. 4 is a view that uses a probabilistic Bayesian network for machine learning of the in Fig. The 3 illustrated methods for controlling the triggering / deployment of an airbag are shown. Fig. 5 is a view in which the airbag deployment result according to one embodiment is compared with the result of the prior art. DETAILED DESCRIPTION

[0049] Fig. Figure 1 is a view showing a vehicle according to one embodiment. Where a component, device, element, or the like of this disclosure is described as having a purpose or performing an operation, function, or the like, the component, device, or element should here be considered to be "configured" to fulfill that purpose or perform that operation or function.

[0050] With reference to Fig. 1 Impact sensors 102 and 106, pressure sensors 104, angular velocity sensors 110 and airbags 108 and 112 are installed in a vehicle.

[0051] At least one pair of frontal impact sensors 102 is installed at the front of the vehicle to detect whether the vehicle has a frontal impact and how severe the impact is. At least one pair of side pressure sensors 104 is installed on both sides of the vehicle to detect the pressure exerted on both sides of the vehicle. At least one pair of side impact sensors 106 is installed on both sides of the vehicle to detect whether the vehicle has a side impact and how severe the impact is.

[0052] In the vehicle according to the embodiment, in addition to the impact sensors 102 and 106 and the pressure sensors 104, at least one pair of angular velocity sensors 110 can be installed to detect an impact direction.

[0053] Airbags 108 and 112 can include front airbags 108 and side airbags 112. At least one pair of front airbags 108 is installed on the front of the driver's seat and on the front of the passenger's seat. The side airbags 112 are installed on the left side of the driver's seat and on the right side of the passenger's seat, respectively.

[0054] To ensure information about vehicle movement, the vehicle can, in addition to the ones in Fig. The sensors shown in Figure 1 include other types of sensors for measuring acceleration, angular velocity, impact, pressure, roll, pitch, yaw, and the like.

[0055] Fig. Figure 2 is a view showing an airbag control device according to one embodiment.

[0056] In Fig. 2 is called a detection device 250, which includes all in Fig. 1 described sensors 102, 104, 106 and 110 as well as sensors for recording vehicle movement information.

[0057] An airbag control unit (ACU) 202 of the vehicle according to the in Fig. In the embodiment shown in Figure 2, the device can determine, based on a measurement result from the detection device 250, whether the airbags 108 and 112 are to be deployed and can generate an airbag deployment command. The airbag control device 202 can comprise application software (ASW) 204, a probability model for human injury or a probability model of human injury(ies) 206, Bayesian network logic 208, logic for determining airbag deployment 210, and basic software (BSW) 212.

[0058] The ASW 204 can control application software installed in the airbag control device 202.

[0059] The human injury probability model 206 (human injury probability calculation unit) can be provided to reflect the measurement result of the detection device 250 when the vehicle collides with another vehicle or obstacle in order to calculate a human injury prediction probability. The human injury prediction probability calculated by the human injury probability model 206 can refer to a probability of a human injury predicted by inputting the measurement result of the detection device 250 during a vehicle impact into a human injury model obtained through various experiments, such as simulations.In the probability model for human injury 206, a relatively simplified model of the human body structure can be used to keep the computational effort in calculating the probability of human injury at a reasonable level.

[0060] The Bayesian network logic (learning unit) 208 can calculate the post-injury probability of a person through probabilistic machine learning from the human injury prediction probability, which is a computational result of the probability model for a person injury 206. The Bayesian network logic 208 can also determine whether the airbags 108 and 112 should be deployed based on the computational result of the post-injury probability of a person. However, the Bayesian network logic 208 can obtain a more reliable basis for determining whether the airbag should be deployed by correcting a bug by re-updating a pre-injury probability of a person from a previous cycle through feedback from post-injury probability values ​​of a person.Here, the post-injury probability of the human is a result of machine learning in the Bayesian network logic 208 according to one embodiment and is a basis for determining whether the airbags 108 and 112 are triggered.

[0061] The airbag deployment determination logic (airbag deployment determination unit) 210 can determine, based on the value for the post-injury probability of the person calculated by the Bayesian network logic 208, whether the airbags 108 and 112 should be deployed. The airbag deployment determination logic 210 can generate the airbag deployment or deployment command if the post-injury probability value of the person exceeds a predetermined reference value, so that the airbags 108 and 112 can / should be deployed or deployed.

[0062] The BSW 212 can control basic software installed in the airbag control unit 202. The airbag deployment command generated by the airbag triggering logic 210 can be transmitted to the airbags 108 and 112 via the BSW 212.

[0063] Fig. Figure 3 is a view showing a method for controlling the airbag deployment of a vehicle according to one embodiment.

[0064] Referring to Fig. 3. The airbag control device 202 can receive the vehicle motion information, which includes the acceleration value, angular velocity value, impact value, pressure value, roll value, pitch value, and yaw value, measured by the sensing device 250 (310). The vehicle motion information can be transmitted via the ASW 204 of the in Fig. The airbag control device 202 described in section 2 is transferred to the probability model for human injury 206.

[0065] The human injury probability model 206, which receives the vehicle motion information, can calculate the human injury prediction probability for an actual impact based on the vehicle motion information (330). In other words, the vehicle motion information measured by the detection device 250 can be fed into a predetermined human injury model to predict a human injury probability value. The human injury prediction probability can include a human injury condition probability. The human injury condition probability is the same as P(impulse ≤ injury severity) and can refer to the probability that a certain magnitude of vehicle impact impulse occurs according to the injury severity.In other words, the human injury condition probability can refer to the probability of the degree of injury at a given time according to the impact impulse of the vehicle.

[0066] In other words, the multiple measurement values ​​acquired by the sensing device 250 are input into the human injury probability model 206 and used to calculate a degree of human injury probability (human injury condition probability) for a head, neck, and thorax. The human injury probability model 206 can have a total of 12 degrees of freedom of translation (XY) and rotation (pitch, roll, and yaw) of the vehicle. The measurement value from the sensing device 250, which is input into the human injury probability model 206, can be calculated as a total of six predictive probabilities of human injury for each part of the human body.The six predictive probabilities for a human injury can represent the probabilities of a frontal head injury, a lateral head injury, a frontal neck injury, a lateral neck injury, a frontal chest injury, and a lateral chest injury. The human injury condition probability value can be output by the human injury probability model 206, along with the six predictive probabilities for a human injury, and input into the Bayesian network logic 208.

[0067] The Bayesian network logic 208 can perform machine learning using a Bayesian network based on the human injury prediction probability and the human injury condition probability calculated by the human injury probability model 206. The Bayesian network logic 208 can also generate a value for the pre-human injury probability and a value for the post-human injury probability as a result of machine learning (350).

[0068] A probabilistic Bayesian network machine learning 350 in the Bayesian network logic 208 can include obtaining the human injury condition probability (352), obtaining the pre-injury human probability (354), and calculating the post-injury human probability (356). The probabilistic Bayesian network machine learning of the vehicle according to the embodiment can include real-time feedback learning of the post-injury human probability.

[0069] In other words, the value for the pre-injury probability of the person can be updated in the next cycle by feeding back the value for the post-injury probability of the person, obtained by calculating the post-injury probability of the person 356, into the pre-injury probability of the person 354 of the next cycle in real time. This update corrects the value of the post-injury probability of the person, allowing for a more accurate determination of whether the airbag needs to be deployed.

[0070] First, Bayesian network logic 208 can obtain the human injury condition probability and the pre-human injury probability from the human injury prediction probability of the human injury probability model 206 (352, 354). Among these, the (e.g., pre-)human injury condition probability can be obtained by the following equation 1. P(xt|ut,xt−1),P(zt|xt)

[0071] Equation 1 is described in particular as follows.

[0072] The expression P(x t | u t ,x t-1 ) represents the human injury prediction probability at a current time (t) according to the measured values ​​of the impact sensors 102 and 106 and a previous (t-1) probability of human injury.

[0073] The expression P(z t | x t) represents the human injury prediction probability, which is predicted according to the occupant injuries measured / obtained through simulation.

[0074] The expression x t represents the actual probability of injury to a person in each of the six areas of the head, neck and chest at the current time (t).

[0075] The expression u t represents the measured value of the detection device 250 at the current time (t).

[0076] The expression x t-1 represents the actual probability of injury to the person at the previous time (t-1).

[0077] The expression z t represents the probability of human injury prediction for each of the six areas of the head, neck, and chest at the current time (t).

[0078] Then, in Bayesian network logic 208, the update of the pre-injury probability of the human can be performed, as shown in Equation 2 below (354). The injury-of-the-human condition probability obtained in step 352 is then multiplied by the value of the pre-injury probability of the human obtained in step 354 to obtain the post-injury probability of the human (356).

[0079] At this point, the value for the pre-injury probability of the human may be an updated value obtained by converting the previous value for the post-injury probability of the human. The pre-injury probability of the human and the post-injury probability of the human can be determined using Equation 2 and Equation 3 below, respectively. (bel(x))¯=∫P(xt|ut,xt−1)bel(xt−1)dxt−1

[0080] Equation 2 is the pre-injury probability of the human, and Equation 2 is described in detail as follows.

[0081] The expression P(x t | u t , x t-1 ) represents the human injury prediction probability according to the measurements of the impact sensors 102 and 106 and the current time (t) according to the previous (t-1) human injury probability.

[0082] The expression bel(x t-1 ) represents the previous (t-1) post-injury probability of the human. bel(x)=ηP(zt|xt)bel(x)¯

[0083] Equation 3 is the post-injury probability of the human, and Equation 3 is described in detail as follows.

[0084] The term η represents a normalization factor.

[0085] The expression P(z t | x t) represents the human injury prediction probability, which is predicted according to the occupant injury measured / obtained by the simulation.

[0086] The expression (bel(x)) represents the pre-injury probability of the human being.

[0087] Here, the calculation of the post-injury probability of a human being is performed using a probabilistic Bayesian network machine learning according to the embodiment with reference to Fig. 4 described.

[0088] Fig. 4 is a view that uses a probability-based Bayesian network machine learning approach from the in Fig. The procedure for controlling airbag deployment is shown in Figure 3.

[0089] Referring to Fig. 4. The probabilistic Bayesian network machine learning of Bayesian network logic 208 can exhibit the preservation of the human injury condition probability, the preservation of the pre-injury human injury probability, and the calculation of the post-injury human injury probability. Furthermore, the probabilistic Bayesian network machine learning can exhibit real-time feedback learning of the post-injury human injury probability. In other words, by feeding the calculated result of the post-injury human injury probability back to the pre-injury human injury probability in real time, thus updating the value of the pre-injury human injury probability, the decision of whether to deploy the airbag can be more accurately corrected. As in Fig. As shown in Figure 4, the cycle for determining the post-injury probability of the human can be repeated continuously. At this point, an output (post-injury probability of the human) from the previous cycle becomes an input (pre-injury probability of the human) for the next cycle.

[0090] Referring again to Fig. 3. The airbag deployment logic 210 can determine airbag deployment based on whether the value for the post-injury probability of the person, calculated by the Bayesian network logic 208, exceeds a predetermined threshold for airbag deployment (370). In other words, if the value for the post-injury probability of the person exceeds the predetermined threshold, it is determined that airbags 108 and 112 must be deployed. Conversely, it is determined that airbags 108 and 112 will not be deployed if the value for the post-injury probability of the person is less than or equal to the predetermined threshold.

[0091] In other words, the airbag deployment logic 210 can determine a sum of frontal injury probabilities based on the values ​​of the frontal head injury probability, the frontal neck injury probability, and the frontal chest injury probability (372).

[0092] Furthermore, the airbag deployment logic 210 can determine whether the sum of the frontal injury probabilities exceeds a predetermined frontal airbag deployment threshold (374).

[0093] Additionally, the airbag deployment logic 210 can determine a sum of the lateral injury probability based on the values ​​of the lateral head injury probability, the lateral neck injury probability and the lateral chest injury probability (376).

[0094] Furthermore, the airbag deployment logic 210 can determine whether the sum of the lateral injury probabilities exceeds a predetermined lateral airbag deployment threshold (378).

[0095] The airbag deployment logic 210 can generate the airbag deployment command to deploy the airbag if the sum of the injury probabilities exceeds the predetermined airbag deployment threshold (JA in 374 or JA in 378) (390). If the sum of the frontal injury probabilities exceeds the predetermined frontal airbag deployment threshold (JA in 374), the airbag deployment logic 210 can generate a frontal airbag deployment command to deploy the frontal airbag. If the sum of the lateral injury probabilities exceeds the predetermined threshold for side airbag deployment (JA in 378), the airbag deployment logic 210 can generate a side airbag deployment command to deploy the side airbag. Both the command to deploy the front airbag and the command to deploy the side airbag can occur (e.g. simultaneously).

[0096] Fig. Figure 5 is a view in which the result of the determination of the airbag deployment according to one embodiment is compared with that of the related technique.

[0097] As in Fig. As shown in Figure 5A, in a conventional case the airbag deployment threshold can be determined based on limited impact data. With conventional threshold-based control, the airbag deployment threshold is set incorrectly or insufficiently. Therefore, it may not be possible to accurately determine the situations in which the airbag deploys and in which it does not.

[0098] Alternatively, in the embodiment of the invention, as in Fig.5B, by predicting the degree of human injury probability through real-time machine learning with feedback using the Bayesian network based on the probability model, and by determining and correcting whether the airbag should be deployed based on the predicted human injury probability, to determine more accurately whether the airbag should be triggered / deployed.

[0099] According to the embodiments of the invention, it is possible to ensure the robustness of the airbag triggering logic and to protect passengers more effectively by determining whether the airbag should be triggered based on the post-injury probability of the human being calculated by a probability model and Bayesian network learning (feedback learning).

Claims

[1] A device (202) for controlling a vehicle airbag, wherein the device comprises a human injury probability calculation unit (206), a learning unit (208) and an airbag deployment detection unit (210), characterized by , that the human injury probability calculation unit (206) is set up to calculate a human injury condition probability and a human injury prediction probability based on vehicle movement information measured by a detection device (250), the learning unit (208) is set up to calculate a post-human injury probability by performing real-time probabilistic machine learning based on the human injury condition probability and the human injury prediction probability, and the airbag deployment determination unit (210) is set up to determine whether the airbag should be deployed based on the post-injury probability of the person. [2] Device (202) according to claim 1, wherein the learning unit (208) is configured to: to calculate a pre-injury probability of the human being based on the injury-of-the-human condition probability and the injury-of-the-human prediction probability, and To calculate the post-injury probability of the human being, multiply the injury-of-the-human-condition probability by the pre-injury probability of the human being. [3] Device (202) according to claim 2, wherein the probability-based machine learning with real-time feedback is set up to update the pre-injury probability of the human by tracing a current post-injury probability of the human back to a previous pre-injury probability of the human. [4] Device (202) according to claim 2 or 3, wherein the human injury condition probability is configured to be calculated using Equation 1: P(xt|ut,xt−1),P(zt|xt) where P(x t | u t ,x t-1 ) a human injury prediction probability at a current time (t) according to the measured values ​​of the impact sensors and a previous (t-1) probability of human injury, P(z t | x t) the human injury prediction probability is predicted according to the occupant injuries measured by a simulation, x t an actual probability of injury to the person for each of the six areas of the head, neck and chest at the current time (t), u t a measured value of a detection device (250) at the current time (t) is, x t-1 an actual probability of injury to the person at the previous time (t-1) and z t a human injury prediction probability for each of the six areas of the head, neck and chest at the current time (t). [5] The device (202) according to any one of claims 2 to 4, wherein the pre-injury probability of the human is set up to be calculated using equation 2: (bel(x))¯=∫P(xt|ut,xt−1)bel(xt−1)dxt−1 where P(x t | u t ,x t-1 ) a human injury prediction probability according to the impact sensor readings and the current time (t) according to the previous (t-1) probability of human injury, and bel(x t-1 ) a previous (t-1) post-injury probability of the human. [6] The device (202) according to claim 2, wherein the post-injury probability of the human is set up to be calculated using equation 3: bel(x)=ηP(zt|xt)bel(x)¯ where η is a normalization factor, P(z t | x t ) is a human injury prediction probability that is predicted according to an occupant injury measured by a simulation, and (bel(x)) is a pre-injury probability of the human. [7] Device (202) according to any of the preceding claims, wherein the airbag deployment detection unit (201) is configured to determine the deployment of the airbag based on the post-injury probability of the person exceeding a predetermined reference value. [8] Device (202) according to any of the preceding claims, wherein the vehicle motion information includes an acceleration value and an angular velocity value of the vehicle, an impact value, a pressure value, a roll value, a pitch value and a yaw value. [9] A method for controlling a vehicle airbag, wherein the method comprises: Calculate (330, 352), by a human injury probability calculation unit (206), a human injury condition probability, and a human injury prediction probability, Calculate (356), through a learning unit (208), a post-injury probability of the human, and Determine (374, 378) by means of an airbag deployment determination unit (201) whether the airbag should be deployed, characterized by , that the human injury condition probability and the human injury prediction probability are calculated based on vehicle movement information measured by a detection device (250), The post-injury probability of the human is calculated by performing probability-based machine learning with real-time feedback based on the human injury condition probability and the human injury prediction probability, and Based on the post-injury probability of the person, it is determined whether the airbag should be deployed. [10] Method according to claim 9, further comprising: Calculate, through the learning unit (208), a pre-injury probability of the human being based on the injury-of-the-human condition probability and the injury-of-the-human prediction probability, and Calculate (356), through the learning unit (208), the post-injury probability of the human by multiplying the injury-of-the-human condition probability with the pre-injury probability of the human. [11] Method according to claim 10, wherein the probability-based machine learning with real-time feedback is set up to update the pre-injury probability of the human by tracing a current post-injury probability of the human back to a previous pre-injury probability of the human. [12] Method according to claim 10 or 11, wherein the human injury condition probability is set up to be calculated by equation 1: P(xt|ut,xt−1),P(zt|xt) where P(x t | u t, x t-1 ) a human injury prediction probability at a current time (t) according to the measured values ​​of the impact sensors and a previous (t-1) probability of human injury, P(z t | x t) is a human injury prediction probability that is predicted according to occupant damage measured by a simulation, x t an actual probability of injury to the person for each of the six areas of the head, neck and chest at the current time (t) is, u t a measured value of a detection device (250) at the current time (t) is, x t-1 an actual probability of injury to the person at the previous time (t-1) and z t a human injury prediction probability for each of the six areas of the head, neck and chest at the current time (t). [13] Method according to any one of claims 10 to 12, wherein the pre-injury probability of the human is set up to be calculated using equation 2: (bel(x))¯=∫P(xt|ut,xt−1)bel(xt−1)dxt−1 where P(x t | u t ,x t-1 ) a human injury prediction probability according to the impact sensor readings and the current time (t) according to the previous (t-1) probability of human injury, and bel(x t-1 ) a previous (t-1) post-injury probability of the human. [14] Method according to any one of claims 10 to 13, wherein the post-injury probability of the human is set up to be calculated using equation 3: bel(x)=ηP(zt|xt)bel(x)¯ where η is a normalization factor, P(z t | x t ) is a human injury prediction probability, which is predicted according to occupant injuries measured by a simulation, and (bel(x)) is a pre-injury probability of the human. [15] Method according to any one of claims 9 to 14, wherein the airbag deployment detection unit (210) is configured to determine the deployment of the airbag based on the post-injury probability of the human being exceeding a predetermined reference value. [16] Method according to any one of claims 9 to 15, wherein the vehicle motion information includes an acceleration value and an angular velocity value of the vehicle, an impact value, a pressure value, a roll value, a pitch value and a yaw value. [17] A device for controlling a vehicle airbag, wherein the device comprises a human injury probability calculation unit (206), a learning unit (208) and an airbag deployment detection unit (210), characterized by , that the human injury probability calculation unit (206) is set up to calculate a human injury condition probability and a human injury prediction probability based on vehicle movement information measured by a detection device (250), the learning unit (208) is set up to: to calculate a pre-injury probability of the human being based on the injury-of-the-human condition probability and the injury-of-the-human prediction probability, to calculate the post-injury probability of the human being by multiplying the injury-of-the-human-condition probability by the pre-injury probability of the human being, and to update the pre-injury probability of a human by feeding back a current post-injury probability of a human to a previous pre-injury probability of a human through probability-based machine learning with real-time feedback, and the airbag deployment determination unit (210) is set up to determine whether an airbag should be deployed based on the post-injury probability of the person. [18] A method for controlling a vehicle airbag, wherein the method comprises: Calculate (330, 352), by a human injury probability calculation unit (206), a human injury condition probability and a human injury prediction probability , Calculate, through a learning unit (208), a pre-injury probability of the human, Calculate (356), through the learning unit (208), the post-injury probability of the human, Update (354), through the learning unit (208), the pre-injury probability of the human, and Determine (374, 378) by means of an airbag deployment detection unit whether an airbag should be deployed, characterized by , that the human injury condition probability and the human injury prediction probability are calculated based on vehicle movement information measured by a detection device (250), The pre-injury probability of the human is calculated based on the injury-of-the-human condition probability and the injury-of-the-human prediction probability. The post-injury probability of the human is calculated by multiplying the injury-of-the-human condition probability by the pre-injury probability of the human. The post-injury probability of a human is updated by feedback from a current post-injury probability of a human to a previous pre-injury probability of a human through probability-based machine learning with real-time feedback, and Based on the post-injury probability of the person, it is determined whether the airbag should be deployed.

Images