Method and device for predicting the severity of a person's injury in a collision with a motor vehicle, and motor vehicle

Abstract

Method (20) for predicting the severity of an injury (28) to a person in the event of a collision between the person and a motor vehicle (11), wherein an impending collision is detected by means of an environmental sensor (12, 13) of the motor vehicle (11) and, after the detection of the impending collision, the severity of the injury (28) to the person as a result of the impending collision is predicted by a control unit (14) of the motor vehicle (11) depending on collision state parameters (25, 26) by means of a prediction function parameterized by processing parameters; device (10) in a motor vehicle (11) for predicting the severity of an injury (28) to a person in the event of a collision between the person and the motor vehicle (11) and motor vehicle (11) with such a device (10).

Description

The invention relates to a method for predicting a severity of injury of a person in the event of a collision of the person with a motor vehicle. The invention further relates to a device in a motor vehicle for predicting a severity of injury of a person in the event of a collision of the person with the motor vehicle, and to a motor vehicle having such a device.Currently, approximately half of the persons sacrificed in road traffic are so-called "Vulnerable Road User" (VRU in the meaning of unprotected road users) worldwide. This term covers those road users who are particularly at risk of being injured or sacrificed in road traffic, since they are not surrounded by a "protective cover" such as, for example, a driver's cab. The group of VRUs includes pedestrians, cyclists, and motor cyclists.Collisions between motor vehicles and VRUs result not only in damage to the object, but frequently in severe to fatal injuries. These arise as a result of a so-called primary impact of the body with the vehicle and as a result of the body being thrown onto the roadway and / or into oncoming traffic (also referred to as secondary impact). The forces and accelerations occurring at the VRU damage, for example, soft parts, connective tissue and bone.An accident or a collision is today detected by vehicle-inherent inertial sensor systems or predictive systems, such as cameras or radar. An emergency call signal (so-called "eCall") is currently initiated at the earliest after the detection of a severe accident with an activation of protection and / or safety systems. Rapid help and assessment of injury is particularly important for higher probability of survival. There is no live access of a rescue control center to rescue-relevant accident and / or vehicle data.Predictive systems such as emergency braking assistants may detect pre-collision VRU accidents and initiate emergency braking that reduces impact velocity.The impact will damage VRUs. The impact not only has an influence on trauma but also, among other things, the position of hit on the vehicle, posture during the collision, age of the VRU and secondary impact, e.g. with a roadway according to a spin scenario.Diagnosis is difficult because imaging examination methods usually do not have any alterations and injuries to the body. Thus, there is no clear evidence and a description of the injuries. A doctor must rely on the description of accident participants and derive the diagnosis and further treatment measures therefrom.DE 10 2014 100 384 A1 discloses a system for predicting injuries to a person in the event of a collision with a vehicle. The vehicle has an image sensor which is used to predict first and second collision points of a head region of the person with a bonnet and a windshield region of the vehicle, which are used as the basis of a prediction of a severity of injury for a control of a protective device of the vehicle and for information to an emergency room of a hospital.EP 3 404 565 A1 describes a method for automatically generating estimation data relating to a potential injury condition of a person involved in a vehicle collision. The person is monitored with respect to a kinetic interaction during the vehicle collision by means of vehicle sensors and / or traffic cameras, and an anticipated severity of injury is determined via an accident database. Estimate data may be provided to an emergency service.Against this background, the object of the invention is to provide a method for predicting a severity of injury of a person in the event of a collision of the person with a motor vehicle, a device of this type in a motor vehicle and a motor vehicle which make possible a rapid, comprehensive and accurate diagnosis of injuries to the person as a result of the collision of the person with the motor vehicle.This object is achieved by a method having the features of claim 1, by a device having the features of claim 9 and by a motor vehicle having the features of claim 10. Further, particularly advantageous embodiments of the invention are disclosed in the respective dependent claims.It should be noted that the features listed individually in the claims can be combined with one another in any technically expedient manner (even beyond category limits, for example between method and apparatus) and reveal further configurations of the invention. The description additionally characterizes and specifies the invention, in particular in conjunction with the figures.It should be further noted that a conjunction "and / or" used herein, between two features and linking them to one another, should always be interpreted such that in a first configuration of the subject matter according to the invention only the first feature may be present, in a second configuration only the second feature may be present, and in a third configuration both the first and the second feature may be present.The subject matter of the invention is a method for predicting a severity of injury of a person (also referred to herein as VRU) in the event of a collision of the person with a motor vehicle, wherein an imminent collision is detected by means of a surroundings sensor system of the motor vehicle and, after the detection of the imminent collision, the severity of injury of the person as a result of the imminent collision is predicted by a control unit of the motor vehicle as a function of collision state parameters by means of a prediction function parameterized by processing parameters.The environment sensor system is used to detect an environment of the motor vehicle, so that, among other things, an imminent (i.e. considered unavoidable) collision of the person with the motor vehicle can be detected from sensor data of the environment sensor system. The detection of the imminent collision from the sensor data of the environment sensor system can be carried out by means of a control unit of the motor vehicle (e.g. ECU), which can be the control unit for predetermination of the severity of injury or another control unit provided in the motor vehicle. The environment sensor system may include, for example, one or more external camera(s) and / or radar, lidar sensors and the like.The control unit is preferably an electronic control unit (e.g. ECU=electronic control unit).The predicted severity of injury is a measure of the anticipated injury of the colliding person resulting from the collision with the motor vehicle. The severity of injury can be represented by qualitative attributes and / or on the basis of numerical values. The predicted severity of injury may include a probability of an expected injury.The predicted severity of injury is dependent on the collision state parameters. These may be obtained from sensor data of the environment sensor system and / or represent a state of the motor vehicle. Such a state of the motor vehicle can be, for example, a speed of the motor vehicle relative to the person or a trim position of the motor vehicle (i.e. a height of a vehicle structure relative to a roadway) or a steering angle of the motor vehicle. Collision state parameters which relate to the person and which can be determined from the sensor data of the environment sensor system can be, for example, a stature of the person, its size, age, posture, step position, speed and direction of travel and the like.Likewise, properties of the environment in which the imminent collision will take place can be determined from the sensor data of the environment sensor system, e.g. a current traffic situation, a presence and / or properties of environmental objects which can represent a risk of injury for the person on the basis of a spin scenario as a result of the collision with the motor vehicle. These properties of the environment represent information which can be used in particular in assessing the severity of injury with respect to a secondary impact.The invention provides a collision-specific optimal prediction of the severity of injury of the person colliding with the motor vehicle. The information about the severity of injury predicted by the vehicle can be made available to helpers (e.g. doctors) arriving at the collision site directly and promptly there, so that a rapid, comprehensive and accurate diagnosis of the injuries that the person may have suffered or suffered from the collision can be created. This significantly increases the probability of survival of the person.According to a preferred embodiment, the collision state parameters relating to the motor vehicle comprise a position relative to the person, a speed relative to the person, a trim position and / or a steering angle of the motor vehicle, and relating to the person a stature, a size, an age, a posture, a step position, a running speed and / or a running direction, and relating to a spatial environment in which the collision will take place, a current traffic position, a presence and / or a property (e.g. geometry, material) of surrounding objects.In further preferred embodiments, the prediction of the severity of injury of the person comprises both a severity of injury as a result of a primary impact of the imminent collision and a severity of injury as a result of a secondary impact of the imminent collision. The primary impact essentially comprises the impact of the person's body with the motor vehicle, while the secondary impact also comprises injury consequences, for example by a spin of the body onto the roadway, into oncoming traffic or against an object or obstacle located in the collision environment (also referred to herein as surrounding object). To assess the severity of injury due to the secondary impact, sensor data provided in particular by the environment sensor system (e.g. camera, radar, lidar sensors, etc.) can be used, which describe the current traffic situation, the spatial environment, etc., for example. Taking into account the consequences of both the primary impact and the secondary impact enables an even more comprehensive and more accurate prediction of the severity of injury of the person colliding with the motor vehicle.According to advantageous further embodiments, in addition to the prediction of the severity of injury, a severity of damage relating to the motor vehicle is additionally predicted as a function of the collision state parameters. In this way, both damage to persons and damage to things as a result of the collision of the motor vehicle with the person can be determined. An automatic transmission of the severity of damage to a vehicle-external recipient, e.g. an insurance company, a motor vehicle service station, etc., is optionally possible.In yet further preferred embodiments, after the accident severity has been determined beforehand, an emergency call to an emergency call point external to the vehicle is triggered on the vehicle side, e.g. police, emergency doctor, fire fight or the like. The triggering can take place in particular automatically and without user intervention. The emergency call can be initiated even before the actual collision.After the emergency call has been triggered, the predetermined severity of injury is likewise transmitted to the emergency call location. In this way, targeted preparations and measures can be initiated by the emergency call site already before the arrival of assistance forces at the location of the collision, which further mitigate the collision consequences for the person colliding with the motor vehicle.In other preferred embodiments, a vehicle-side crash sensor system is used to determine whether the previously detected collision is plausible and has actually taken place, and the emergency call is triggered only after the determination that the collision is plausible and has taken place. The plausibility of the collision may be checked, for example, on the basis of the collision state parameters that are not intended to conflict. In this way, even if an unavoidable collision was initially detected by the environment sensor system, an additional plausibility check of the collision conditions can be carried out before the emergency call is triggered. Preferred crash sensors comprise, for example, vehicle-side acceleration sensors and / or pressure hose sensors, with which the occurrence of the collision can be unambiguously confirmed if a corresponding measured value (i.e. acceleration or pressure) exceeds a predefined threshold value over a predeterminable period of time, for example more than 300 ms.Further preferred embodiments provide that the processing parameters of the prediction function are trained during a training phase by machine learning using training data sets.The prediction function can be formed, for example, by means of data-based models that are trained by the training data sets. These models can be, for example, correspondingly trained neural networks or statistical data modelling. For example, a neural network based on radial basis functions can be used. It is also possible to provide a plurality of, in particular uncorrelated, decision trees, the results of which are combined. In particular, if at least one numerical prediction result (i.e. a numerically represented severity of injury) is determined, the prediction function can also be parameterized by a regression method.The training data sets can each comprise input data and output data, wherein the prediction function can be trained, for example, by an error feedback, also referred to as backpropagation of error, in such a way that, by applying the prediction function to the input data, output data are generated which are as similar as possible to the output data of the respective training data sets.The training data sets are preferably generated by means of a FEM crash simulation based on the collision state parameters and FEM human model simulations (FEM, finite element method). Here, information about the colliding person (e.g., stature, size, etc.) and the collision conditions (e.g., posture, speed, and direction of the person relative to the motor vehicle, etc.) including the FEM human models are used to determine collision-specific loads of the person.The prognosis or prediction of the severity of injury is based on the use of FEM human model simulations on physical relationships and the biomechanics of the human. In the FEM crash models, geometries and stiffnesses of the motor vehicle, a physical structure, physical size, posture, running direction etc. of the colliding person and surrounding objects at the location of the colliding event can be taken into account.With the aid of these data, the collision can be adjusted in great detail and while preserving the cause-effect principle. Loads on each body part of the colliding person can be determined and the severity of injury can be predicted individually as a function of the load variables. The use of the trained, parameterized prediction function enables the prediction of the severity of injury in a short time, i.e. at the latest immediately after the occurrence of the collision.The processing parameters of the prediction function can be trained with a multiplicity of possible collision scenarios (i.e. variation of all collision state parameters in the FEM crash simulation) in order to generate a personalized prediction of the severity of injury in the event of a collision in a very short time. Since the training data is subject to physical relationships, apparent correlations can be excluded when sufficient training data is used. The prediction quality is consequently very high and can be further increased by a further development of the FEM models used.Preferably, the prediction of the severity of injury is not only limited to the primary impact (i.e., impact of the person with the motor vehicle), but also involves the secondary impact (i.e., collision of the person with surrounding objects). Both the primary impact and the secondary impact can be taken into account accordingly in the training data and can consequently be trained.Furthermore, the subject matter of the invention is a device in a motor vehicle for predicting a severity of injury of a person in the event of a collision of the person with the motor vehicle, which device has a surroundings sensor system (e.g. camera, radar, lidar sensors and the like) and a control unit (e.g. ECU), wherein the control unit is configured to execute a method according to one of the embodiments disclosed herein.It is to be understood that, with regard to device-related term definitions and the effects and advantages of device-related features, the disclosure of analogous definitions, effects and advantages of the method according to the invention can be fully resorted to and vice versa. Repetition of explanations of correspondingly identical features, the effects and advantages thereof can thus be dispensed with in favor of a more compact description, without such omissions having to be interpreted as a restriction for one of the disclosed subject matter of the invention.Preferred embodiments provide that the device has a data remote transmission module for automatically transmitting an emergency call and the severity of injury predicted for the imminent collision. An optionally predetermined severity of damage with respect to the motor vehicle can likewise be transmitted by means of the remote data transmission module, for example to a service station of the motor vehicle or an insurance company.In further embodiments, the device preferably has a crash sensor system (e.g. acceleration sensors and / or pressure tube sensors), with which the plausibility and the actual occurrence of the previously detected collision can be checked, e.g. before the actual triggering of the emergency call and / or before the transmission of the severity of damage.Furthermore, the subject matter of the invention is a motor vehicle having a device for predicting a severity of injury of a person in the event of a collision of the person with the motor vehicle according to one of the embodiments disclosed herein.Further features and advantages of the invention will become apparent from the following description of exemplary embodiments of the invention, which are to be understood as non-limiting and are explained in more detail below with reference to the drawings. In this drawing, the following are shown schematically: FIG. 1 shows an apparatus and a motor vehicle according to an embodiment of the invention; and FIG. 2 is a flow chart for explaining a method according to one embodiment of the invention.In the different figures, parts which are equivalent in terms of their function are always provided with the same reference symbols, so that they are generally also described only once.FIG. 1 schematically illustrates a device 10 and a motor vehicle 11 having the device 10 for predicting a severity of injury 28 (FIG. 2 ) of a person (not illustrated) in the event of a collision of the person with a motor vehicle 11, in each case according to one embodiment of the invention.FIG. 1 shows that the device 10 or the motor vehicle 11 have a surroundings sensor system which, in the present case, comprises, for example, one or more external camera(s) 12 and radar and / or lidar sensors 13. Furthermore, the device 10 or the motor vehicle 11 has a control unit 14 which is configured to execute a method as disclosed herein for predicting the severity of injury 28 of a person in the event of a collision of the person with the motor vehicle 11.As can be further seen from FIG. 1, the exemplary device 10 or the motor vehicle 11 also has a crash sensor system 15, which can comprise acceleration sensors and / or so-called pressure hose sensors, for example. These can be used, for example, to confirm that the collision between the person and the motor vehicle 11 is plausible and has actually taken place.Furthermore, the protection control device 10 or the motor vehicle 11 in the present case has a data remote transmission module 16, with which an emergency call and / or data about the predetermined severity of injury can / can be transmitted from the control unit 14 to an emergency call location external to the vehicle, e.g. police, emergency doctor, fire fight and the like. The control unit 14 can actuate the data transmission module 16 accordingly after it has predetermined the severity of injury 28 to be expected for the imminent collision. A severity of damage 32 (FIG. 2 ) relating to the motor vehicle 11 optionally predicted by the control unit 14 can likewise be transmitted by means of the remote data transmission module 16, for example to a service station of the motor vehicle 11 or an insurance company.FIG. 2 schematically illustrates a flow chart 20 for explaining a method for protecting a person (not shown) in the event of a collision of the person with a motor vehicle, e.g. motor vehicle 11 from FIG. 1, according to an embodiment of the invention. Method 20 may be carried out, for example, by control unit 14 of motor vehicle 11 or by control unit 14 of device 10.FIG. 2 is divided into three sections A, B and C, wherein exemplary method steps of the method during the operational execution in the motor vehicle are shown in section C (also referred to herein as field situation). Section A schematically illustrates an operation for generating training data (also referred to herein as training dataset generation phase) used in section B for training a prediction function within the scope of machine learning (also referred to herein as training phase).In the method shown in FIG. 2 (see section C), an imminent collision of the person with the motor vehicle is detected in a method step 21 by means of a surroundings sensor system of the motor vehicle (e.g. external camera, radar, lidar sensors), as has already been described elsewhere herein. In a step 22, a check is optionally made as to whether the collision is unavoidable. If the collision is avoidable, e.g. by initiating an avoidance and / or emergency braking maneuver of the motor vehicle, the method ends in step 23, in which no further actions are carried out any longer.If the collision between the person and the motor vehicle is unavoidable, the method continues with step 24, in which the severity of injury 28 of the person as a result of the imminent collision is predicted by means of a control unit, e.g. control unit 14 from FIG. 1, as a function of collision state parameters 25 and 26. The prediction of the severity of injury 28 is carried out by means of a prediction function parameterized by processing parameters.In the example illustrated here, the collision state parameters 25 relate to properties of the colliding person, such as a stature, a size, an age, a posture, a step position, a travel speed and / or a travel direction. The collision state parameters 26 relate to the motor vehicle and describe vehicle states such as a position relative to the person, a speed relative to the person, a trim position and / or a steering angle. In addition, other collision state parameters (not shown) may relate to characteristics of an environment of the location where the collision will occur, such as a current traffic situation, a presence and / or a characteristic of surrounding objects.The processing parameters of the prediction function are trained during a training phase B by machine learning using training data sets, as will be described further below in connection with the explanation of sections A and B of FIG. 2.After the severity of injury 28 has been predicted, in the exemplary embodiment shown here, a check is made in step 27 as to whether the collision is plausible and has actually taken place. For this purpose, crash sensor system 15 (see FIG. 1 ) may be used, for example. If, for example, a corresponding measured value of vehicle-side acceleration sensors and / or pressure hose sensors (i.e. acceleration or pressure) exceeds a predefined threshold value over a predeterminable period of time, e.g. more than 300 ms, it can be assumed that the collision has actually taken place.If it is determined in step 27 that the collision has not taken place, the method continues with step 23, in which the method is ended.If it is determined in step 27 that the collision has occurred, in the exemplary embodiment shown here, an emergency call 31 to an emergency call location external to the vehicle is triggered on the vehicle side and, after the emergency call 31 has been triggered, the predetermined severity of injury 28 is transmitted to the emergency call location, for example wirelessly by means of the remote data transmission module 16 shown in FIG. 1.FIG. 2 additionally shows that, in the present case, after the severity of injury 28 has been determined in a step 32, a severity of damage to the collision relating to the motor vehicle is additionally predetermined on the vehicle side as a function of the collision state parameters 25, 26, but is not necessarily limited thereto.Section B of FIG. 2 represents the training of the parameterized prediction function using training data sets 33 (cf. section A) in the context of machine learning. Step 24 shown in section B comprises the prediction of the severity of injury essentially as already described in connection with section C based on input data comprising the collision state parameters 25 and 26, wherein in the training phase B a training severity of injury 34 is predicted.The prediction function used in step 24 can be formed, for example, by means of data-based models that are trained by the training data sets 33. These models can be correspondingly trained neural networks or statistical data modelling. For example, a neural network based on radial basis functions can be used. It is also possible to provide a plurality of, in particular uncorrelated, decision trees, the results of which are combined. In particular, if at least one numerical prediction result (i.e. severity of injury) is determined, the prediction function can also be parameterized by a regression method.The training data sets 33 can each comprise input data (i.e. the collision state parameters 25, 26) and output data (i.e. a severity of injury calculated as a function of the collision state parameters 25, 26), wherein the prediction function in step 24 of section B is trained, for example, by error feedback, as is shown in section B of FIG. 2 in the present exemplary embodiment, in such a way that, by applying the prediction function to the input data (i.e. collision state parameters 25, 26), output data (i.e. the predicted training severity of injury 34) are generated, which are as similar as possible to the output data of the respective training data sets 33.Section A of FIG. 2 schematically illustrates a phase for generating the training datasets 33. The training data sets 33 are preferably generated by means of a FEM crash simulation 35 based on the collision state parameters 25, 26 and FEM human model simulations. Here, information on the colliding person (e.g., stature, height, age, etc.) and the colliding state (e.g., posture, speed, and direction of the person relative to the automobile, etc.) including finite element human model simulations are used to determine loads of the person. In the FEM crash models, geometries and stiffnesses of the motor vehicle and surrounding objects at the location of the collision event are preferably likewise taken into account.LIST OF REFERENCE CHARACTERS:10 Protection control device 11 Motor vehicle 12 External camera 13 Radar sensor, lidar sensor 14 Control unit 15 Crash sensor system 16 Data remote transmission module 20 Flow diagram 21 Detection of a collision 22 Checking whether a collision is avoidable 23 Method end 24 Prediction of a severity of injury 25 Collision state parameter relating to person 26 Collision state parameter relating to motor vehicle 27 Checking for plausibility 28 Predicted severity of injury 31 Triggering emergency call 32 Prediction of a severity of damage 33 Training datasets 34 Predicted training severity of injury 35 FEM crash simulations A Training dataset generation phase B Training phase C Field conditionsReferences included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedDE 10 2014 100 384 A1

[0008] EP 3 404 565 A1

[0009]

Claims

Method (20) for predicting a severity of injury (28) of a person in the event of a collision of the person with a motor vehicle (11), wherein an imminent collision is detected by means of a surroundings sensor system (12, 13) of the motor vehicle (11) and, after the detection of the imminent collision, the severity of injury (28) of the person as a result of the imminent collision is predicted by a control unit (14) of the motor vehicle (11) as a function of collision state parameters (25, 26) by means of a prediction function parameterized by processing parameters.Method according to Claim 1, in which the collision state parameters (25, 26) relating to the motor vehicle (11) comprise a position relative to the person, a speed relative to the person, a trim position and / or a steering angle, and relating to the person a stature, a size, an age, a posture, a step position, a running speed and / or a running direction, and relating to a spatial environment of the collision a current traffic position, a presence and / or a property of surrounding objects.The method of claim 1 or 2, wherein predicting the severity of injury (28) comprises both a severity of injury due to a primary impact of the imminent collision and a severity of injury due to a secondary impact of the imminent collision.Method according to one of the preceding claims, in which a severity of damage (32) relating to the motor vehicle (11) is predetermined as a function of the collision state parameters (25, 26).Method according to one of the preceding claims, in which, after the prediction of the severity of injury (28), an emergency call (31) to an emergency call location external to the vehicle is triggered on the vehicle side and, after the triggering of the emergency call (31), the predicted severity of injury (28) is transmitted to the emergency call location.Method according to Claim 5, in which it is determined by means of a vehicle-side crash sensor system (15) whether the detected collision is plausible and has actually taken place, and the emergency call (31) is triggered only after the determination that the collision is plausible and has taken place.Method according to one of the preceding claims, in which the processing parameters are trained during a training phase (B) by machine learning using training data sets (33).Method according to Claim 7, in which the training data sets (33) are generated by means of a FEM crash simulation (35) based on the collision state parameters (25, 26) and FEM human model simulations.Device (10) in a motor vehicle (11) for predicting a severity of injury (28) of a person in the event of a collision of the person with the motor vehicle (11), having a surroundings sensor system (12, 13) and a control unit (14), wherein the control unit (14) is configured to execute a method (20) according to one of the preceding claims.Motor vehicle (11) having a device (10) according to Claim 9.

Images