JP2022106636A - Method and system for estimating collision part of vehicle - Google Patents

Abstract

To provide a method and system for estimating a collision part of a vehicle.SOLUTION: A method for estimating a collision part of a vehicle comprises the steps of: determining that an impact is generated in a vehicle when the acceleration of the vehicle is measured and the acceleration has magnitude equal to or greater than a critical value; determining whether the vehicle is impacted by the collision from the output of a machine learning model by extracting a feature element for determining the collision from acceleration data before and after the generation of the impact and using the feature element as the input of the machine learning model when it is determined that the impact is generated in the vehicle; and estimating a collision part of the vehicle from the output of the machine learning model by extracting the feature element for estimating the collision part from the acceleration data and using the feature element as the input of the machine learning model when it is determined that the vehicle is impacted by the collision.SELECTED DRAWING: Figure 2

Description

The present invention relates to a vehicle contact accident site estimation method and a system.

When a vehicle is impacted when parked, a method has been developed to detect this and notify the owner of the vehicle. However, the causes of impacts generated on vehicles include not only contact accidents due to collisions with other vehicles, but also non-contact incidents such as opening and closing of doors and wind caused by the passage of other vehicles. There is a need for a method that can distinguish and detect.

Further, even when an impact is generated due to a contact accident with the vehicle, there is also a need for a method capable of detecting which part of the vehicle the impact is generated in.

Publication of Korean Patent Application No. 10-2020-0102768 (published on September 1, 2020)

One object of the present invention is to provide a method for estimating a contact accident site of a vehicle.

Another object of the present invention is to provide a vehicle contact accident site estimation system.

However, the object of the present invention is not limited to the above-mentioned object, and may be extended in various ways without departing from the idea and domain of the present invention.

In order to achieve one object of the present invention, in the vehicle contact accident site estimation method according to the embodiment of the present invention, when the acceleration of the vehicle is measured and the magnitude is equal to or larger than the critical value, the vehicle is impacted. It can be determined that it has occurred. When it is determined that an impact has occurred on the vehicle, a feature element for determining a contact accident is extracted from the acceleration data before and after the impact occurs and used as an input of a machine learning model to perform the machine learning. From the output of the model, it can be determined whether the vehicle has been impacted by a contact accident. When it is determined that the vehicle has been impacted by a contact accident, a feature element for estimating the contact accident site is extracted from the acceleration data and used as an input of the machine learning model, and the machine learning model is used. The contact accident site of the vehicle can be estimated from the output of.

In an exemplary embodiment, the output of the machine learning model when determining whether the vehicle has been impacted by a contact accident includes a contact accident and a non-contact incident, and the vehicle is impacted by the contact accident accordingly. When determining whether or not it has been received, it can include filtering when the vehicle is impacted by the non-contact incident.

In an exemplary embodiment, the impact of the non-contact incident is due to the opening and closing of the vehicle door, the impact of closing the trunk and closing of the hood, the impact of the wind generated by the passage of another vehicle, and the shaking of the bottom. It can include the impact that occurs.

In an exemplary embodiment, the vehicle is divided into 8 or 12 parts, and the output of the machine learning model when estimating the contact accident part of the vehicle is any of the divided parts of the vehicle. It can be estimated that an impact due to a contact accident has occurred in any one of the divided parts of the vehicle when estimating the contact accident part of the vehicle.

In an exemplary embodiment, the acceleration of the vehicle can be measured by a 3-axis accelerometer mounted in the vehicle.

In an exemplary embodiment, the acceleration of the vehicle can be measured when the vehicle is parked.

In the exemplary embodiment, the characteristic elements for determining the contact accident are the effective value (RMS), the maximum acceleration, and the attenuation of the acceleration for each axis before and after the time when the impact is generated, which are measured by the three-axis accelerometer. Can include at least one of the inclination of, the change of sign of positive and negative, and the duration of the impact.

In an exemplary embodiment, the feature elements for estimating the contact accident site are the effective value (RMS), maximum acceleration, and acceleration of each axis before and after the impact occurrence time point measured by the three-axis accelerometer. It can include at least one of the slope of the damping, the change in sign of positive and negative, and the duration of the impact.

In an exemplary embodiment, the learning method of the machine learning model used when determining whether the vehicle has been impacted by a contact accident or when estimating the contact accident site of the vehicle is a decision tree, a support vector. It can include at least one of a machine (SVM), a neural network, and deep learning.

In an exemplary embodiment, the learning methods of the machine learning model used when determining whether the vehicle has been impacted by a contact accident and when estimating the contact accident site of the vehicle may differ from each other.

In an exemplary embodiment, in the learning process of the machine learning model used at the stage of estimating the contact accident part of the vehicle, i) before and after the time of occurrence of each collision by carrying out a collision experiment for each part of the vehicle. Acceleration data of the vehicle can be collected and the location of each collision can be recorded. ii) A feature element is obtained from the collected acceleration data and used as an input of the machine learning model, and a collision part corresponding to each acceleration data among the recorded collision parts is used as an output of the machine learning model. be able to. Along with this, the machine learning model can learn appropriate parameters for the input and the corresponding output.

In an exemplary embodiment, the ii) step can be performed repeatedly, with which the accuracy of the parameters can be improved.

In an exemplary embodiment, only a portion of the collected acceleration data can be used to determine the feature element, and the learning process of the machine learning model used in the step of estimating the contact accident site of the vehicle. In, verification for evaluating the accuracy of the machine learning model can be performed using the acceleration data that is not used to obtain the feature element among the collected acceleration data.

In an exemplary embodiment, in paragraph 1, in the learning process of the machine learning model used in the vehicle contact accident determination stage, i) perform an impact experiment including a contact accident and a non-contact incident with the vehicle. Acceleration data of the vehicle before and after the occurrence of each impact can be collected, and a contact accident or a non-contact incident that is the cause of each impact can be recorded. ii) The feature element is obtained from the collected acceleration data and used as the input of the machine learning model, and one of the recorded contact accidents and non-contact incidents corresponding to each acceleration data is the machine learning model. Can be the output of. Along with this, the machine learning model can learn appropriate parameters for the input and the corresponding output.

In an exemplary embodiment, the ii) step can be performed repeatedly, with which the accuracy of the parameters can be improved.

In order to achieve one object of the present invention, in the method for estimating a contact accident site of a vehicle according to another embodiment of the present invention, when the acceleration of the vehicle is measured and the value is equal to or larger than the critical value, the vehicle is subjected to the above-mentioned vehicle. It can be determined that an impact has occurred. When it is determined that an impact has occurred on the vehicle, characteristic elements for determining the contact accident and estimating the contact accident site are extracted from the acceleration data before and after the impact occurrence time, and this is used as the input of the machine learning model. It is used to determine whether the vehicle has been impacted by a contact accident from the output of the machine learning model, and if it is determined that the vehicle has been impacted by a contact accident, the contact accident site of the vehicle is estimated. can do.

In an exemplary embodiment, the acceleration of the vehicle can be measured while the vehicle is parked by a 3-axis accelerometer mounted in the vehicle.

In an exemplary embodiment, the feature element for determining the contact accident and estimating the contact accident site is the effective value (RMS) of the acceleration for each axis before and after the impact occurrence time point measured by the three-axis accelerometer. , Maximum acceleration, slope of acceleration attenuation, change of positive and negative sign, and at least one of the duration of the impact.

In an exemplary embodiment, in the process of learning the machine learning model, although a contact accident and a non-contact incident with the vehicle are included, in the case of the contact accident, an impact experiment including a collision of each part of the vehicle is performed. It is possible to collect acceleration data of the vehicle before and after the occurrence of each impact, and record the contact accident or non-contact incident that is the cause of each impact and the occurrence site of each collision. ii) The feature element is obtained from the collected acceleration data and used as the input of the machine learning model, and one of the recorded contact accident and the non-contact incident corresponding to each acceleration data is the machine learning. Although the output of the model is used, in the case of the contact accident, the collision site corresponding to each acceleration data among the collision sites can be output of the machine learning model. Along with this, the machine learning model can learn appropriate parameters for the input and the corresponding output.

In an exemplary embodiment, the ii) step can be performed repeatedly, with which the accuracy of the parameters can be improved.

In order to achieve another object of the present invention, the vehicle contact accident site estimation system according to the embodiment of the present invention is a sensor unit that measures the acceleration of the vehicle; a memory that stores the acceleration data measured by the sensor unit. Unit; Control unit that determines that an impact has occurred on the vehicle when the acceleration measured by the sensor unit has a magnitude equal to or greater than a critical value; The control unit determines that an impact has occurred on the vehicle. In this case, a feature element for determining a contact accident is extracted from the acceleration data before and after the occurrence of the impact stored in the memory unit, and this is used as an input of the machine learning model, and is used from the output of the machine learning model. A contact accident determination unit that determines whether the vehicle has been impacted by a contact accident; and an acceleration stored in the memory unit when the contact accident determination unit determines that the vehicle has been impacted by a contact accident. It includes a contact accident part estimation unit that extracts a feature element for estimating a contact accident part from data, uses it as an input of a machine learning model, and estimates the contact accident part of the vehicle from the output of the machine learning model. be able to.

In an exemplary embodiment, the output of the machine learning model used by the contact accident determination unit can include contact accidents and non-contact incidents, thereby determining whether the vehicle has been impacted by the contact accident. This can include filtering when the vehicle is impacted by the non-contact incident.

In an exemplary embodiment, the vehicle may be divided into 8 or 12 parts, and the output of the machine learning model in the contact accident part estimation unit is any one of the divided parts of the vehicle. In connection with this, estimating the contact accident site of the vehicle can include presuming that an impact due to the contact accident has occurred in any one of the classified parts of the vehicle.

In an exemplary embodiment, the sensor unit may include a 3-axis accelerometer mounted in the vehicle, the acceleration of the vehicle may be measured by the 3-axis accelerometer when the vehicle is parked. ..

In an exemplary embodiment, the feature element for determining the contact accident or the feature element for estimating the contact accident site is the acceleration for each axis before and after the impact generation time measured by the three-axis accelerometer. The effective value (RMS), maximum acceleration, can include at least one of the slope of velocity attenuation, the change of sign of positive or negative sign, and the duration of the impact.

In an exemplary embodiment, the learning methods of the machine learning model used in the contact accident determination unit or the contact accident site estimation unit include a decision tree, a support vector machine (SVM), a neural network, and deep learning ( At least one of deep learning) can be included.

In an exemplary embodiment, in the learning process of the machine learning model used in the contact accident site estimation unit of the vehicle, i) the collision experiment is performed for each part of the vehicle, and the collision experiment is performed before and after the occurrence of each collision. It is possible to collect the acceleration data of the vehicle and record the occurrence site of each of the collisions. ii) A feature element is obtained from the collected acceleration data and used as an input of the machine learning model, and a collision part corresponding to each acceleration data among the recorded collision parts is used as an output of the machine learning model. be able to. Along with this, the machine learning model can learn appropriate parameters for the input and the corresponding output.

In an exemplary embodiment, the ii) step can be performed repeatedly, with which the accuracy of the parameters can be improved.

In an exemplary embodiment, in the learning process of the machine learning model used in the vehicle contact accident determination unit, i) an impact experiment including a contact accident and a non-contact incident with the vehicle is performed to generate each of the impacts. Acceleration data of the vehicle before and after the time point can be collected, and a contact accident or a non-contact incident that is the cause of each impact can be recorded. ii) A feature element is obtained from the collected acceleration data and used as an input of the machine learning model, and one of the contact accident and the non-contact incident corresponding to each acceleration data is output of the machine learning model. Can be. Along with this, the machine learning model can learn appropriate parameters for the input and the corresponding output.

In an exemplary embodiment, the ii) step can be performed repeatedly, with which the accuracy of the parameters can be improved.

In order to achieve another object of the present invention, the vehicle contact accident site estimation system according to another embodiment of the present invention is a sensor unit that measures the acceleration of the vehicle; and stores the acceleration data measured by the sensor unit. A memory unit; and an accident estimation unit can be provided. When the acceleration measured by the sensor unit has a magnitude equal to or higher than the critical value, the accident estimation unit determines that an impact has occurred on the vehicle, and is stored in the memory unit before and after the occurrence of the impact. A characteristic element for determining a contact accident is extracted from the acceleration data of the machine learning, and this is used as an input of a machine learning model. Judgment unit; and when the contact accident determination unit determines that the vehicle has been impacted by a contact accident, a feature element for estimating the contact accident site is extracted from the acceleration data stored in the memory unit. This can be used as an input of the machine learning model, and can include a contact accident part estimation unit that estimates the contact accident part of the vehicle from the output of the machine learning model.

According to the vehicle contact accident site estimation method and system according to the embodiment of the present invention, when an impact occurs on a parked vehicle using a machine learning model, non-contact incidents can be filtered, and in the case of a contact accident. , The contact accident site can be easily estimated.

However, the effect of the present invention is not limited to the above-mentioned effect, and may be expanded in various ways without departing from the idea and domain of the present invention.

It is a block diagram which shows the contact accident part estimation system of the vehicle which concerns on embodiment of this invention. It is a flowchart which shows the contact accident part estimation method of the vehicle which concerns on embodiment of this invention. It is a graph which showed the change of the acceleration for each axis by the type of the impact generated in a vehicle. It is a drawing which showed the vehicle divided into 8 or 12 parts on a plane composed of X-axis and Y-axis. It is a flowchart which shows the contact accident part estimation method of the vehicle which concerns on other Examples of this invention.

With respect to the examples of the present invention disclosed in the text, specific structural or functional explanations are merely exemplified for the purpose of explaining the examples of the present invention, and the examples of the present invention are illustrated. It can be implemented in a variety of forms and should not be construed as being limited to the examples described in the text.

The present invention can be modified in various ways and can have various forms, and a specific embodiment will be illustrated in the drawings and described in detail in the text. However, this is not intended to limit the invention to any particular form of disclosure and should be understood to include all modifications, equivalents or alternatives contained within the ideas and technical scope of the invention. be.

The terms first, second, etc. can be used to describe various components, but the components should not be limited by the terms. The term may be used to distinguish one component from another. For example, the first component can be named the second component without departing from the scope of rights of the present invention, and similarly the second component can be named the first component.

When it is mentioned that one component is "connected" or "connected" to another component, it may or may not be directly connected to the other component. It should be understood that there may be other components in between. On the other hand, when it is mentioned that one component is "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between. Other expressions that describe the relationships between the components, such as "between" and "immediately between" or "adjacent to" and "directly adjacent to", should be interpreted in the same way. be.

The terms used in this application are used solely to describe a particular embodiment and are not intended to limit the invention. A singular expression includes multiple expressions unless they are meant to be explicitly different in context. In this application, terms such as "contain" or "have" are intended to specify the existence of the features, numbers, stages, actions, components, components or combinations thereof described. It should be understood that it does not preclude the possibility of existence or addition of one or more other features or numbers, stages, actions, components, components or combinations thereof.

In addition, terms such as "-part" described in the text mean a unit for processing at least one function or operation, which may be embodied in hardware or software or a combination of hardware and software.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by those with ordinary knowledge in the technical field to which the present invention belongs. Is. Terms such as those defined in commonly used dictionaries should be construed as having a meaning consistent with the context of the relevant technology and should be ideal or excessive unless expressly defined in this application. Is not interpreted in a formal sense.

Hereinafter, examples of the present invention will be described in more detail with reference to the attached drawings. The same components on the drawings are designated by the same or similar reference numerals.

FIG. 1 is a block diagram showing a vehicle contact accident site estimation system according to an embodiment of the present invention.

Referring to FIG. 1, the vehicle contact accident site estimation system 100 according to the embodiment of the present invention can include a sensor unit 200, a memory unit 300, a control unit 400, and an accident estimation unit 500. Further, the vehicle contact accident site estimation system 100 may further include a photographing device such as a camera and a communication device such as a data transmission modem.

The vehicle contact accident site estimation system 100 as described above may be, for example, a vehicle black box, a built-in black box (build-in cam), or the like.

The sensor unit 200 can be mounted in the vehicle and may include an accelerometer such as a 3-axis accelerometer. The three-axis accelerometer is formed by an X-axis that crosses the vehicle in the front-rear direction, a Y-axis that crosses the left and right sides of the vehicle, and an X-axis and a Y-axis, with a position mounted in the vehicle as a reference point (zero point). The acceleration of the vehicle can be measured using a three-dimensional local coordinate system composed of a Z-axis perpendicular to the plane. In the 3-axis accelerometer, the X-axis, the Y-axis, and the Z-axis can be freely arranged as long as they are perpendicular to each other.

However, in some cases, the sensor unit 200 may include a biaxial accelerometer that includes only the X-axis and the Y-axis.

The memory unit 300 can store various data related to the acceleration measured by the sensor unit 200. The memory unit 300 can include, for example, a ram (RAM), a ROM (ROM), a hard disk drive (HDD), a solid state drive (SSD), and the like.

The control unit 400 can control each operation performed by the sensor unit 200, the memory unit 300, and the accident estimation unit 500, and can perform various operations necessary for the operation. The control unit 400 can include, for example, a microcontroller unit (MCU), a microprocessor unit (MPU), an application processor (AP), and the like.

In an exemplary embodiment, the control unit 400 can determine that an impact has occurred on the vehicle when the acceleration measured by the sensor unit 200 has a magnitude equal to or greater than a critical value.

The accident estimation unit 500 can include a contact accident determination unit 510 and a contact accident site estimation unit 520. The accident estimation unit 500 can include, for example, a microcontroller unit (MCU), a microprocessor unit (MPU), an application processor (AP), and the like.

In an exemplary embodiment, when the contact accident determination unit 510 is determined by the control unit 400 that an impact has occurred on the vehicle, machine learning is performed from the acceleration data before and after the impact occurrence point stored in the memory unit 300. The model can be used to determine if the vehicle was impacted by a contact accident. Along with this, the contact accident determination unit 510 can drive software including an algorithm for determining a vehicle contact accident.

In an exemplary embodiment, when the contact accident determination unit 510 determines that the vehicle has been impacted by the contact accident, the contact accident site estimation unit 520 is a machine learning model from the acceleration data stored in the memory unit 300. Can be used to estimate the contact accident site of the vehicle. Along with this, the contact accident determination unit 510 can drive software including an algorithm for estimating the contact accident portion of the vehicle.

Until now, the accident estimation unit 500 exists separately from the control unit 400, the control unit 400 executes the impact determination operation of the vehicle, and the accident estimation unit 500 performs the contact accident determination operation and the contact accident part estimation operation of the vehicle. However, the concept of the present invention is not limited to this.

That is, the accident estimation unit 500 may be arranged in the control unit 400, and accordingly, one integrated unit, that is, the control unit 400 or the accident estimation unit 500 may not only perform various control and calculation operations but also impact. The determination, the determination of the contact accident, and the contact accident site estimation operation may be performed together.

In the following, for convenience, only the accident estimation unit 500 exists separately from the control unit 400 and they operate separately, but it is also possible for them to form one unit and perform all the above operations. be.

FIG. 2 is a flowchart showing a method for estimating a contact accident portion of a vehicle according to an embodiment of the present invention.

Referring to both FIGS. 1 and 2, in the first stage (S110), the sensor unit 200 included in the vehicle contact accident site estimation system 100 (hereinafter referred to as system 100) mounted in the vehicle is When the acceleration of the vehicle can be measured and the acceleration has a magnitude equal to or higher than the critical value in the second stage (S120), the control unit 400 included in the system 100 causes an impact on the vehicle. Can be determined.

In an exemplary embodiment, the vehicle can be a parked vehicle.

In an exemplary embodiment, the acceleration of the vehicle can be measured by a 3-axis accelerometer, the size of the acceleration vector formed by the acceleration measured for each axis is calculated, and the size is equal to or greater than a certain value. Only in the case of, it can be determined that the impact has occurred on the vehicle.

At this time, the size of the acceleration vector is the maximum acceleration which is the maximum value of the size of the acceleration vector for a certain period of time before and after the impact generation, or the root mean square of the acceleration vector for a certain period of time before and after the impact generation. It can be an effective value (Root Mean Square: RMS).

So far, the description has been made based on the acceleration of the vehicle, but it is assumed that the impact is generated on the vehicle even when the impact force, which is the force obtained by integrating the mass of the vehicle, is equal to or higher than the critical value. Needless to say, it can be judged.

On the other hand, the acceleration data measured by the sensor unit 200 can be stored in the memory unit 300 included in the system 100.

When the control unit 400 determines in the third stage (S130) that an impact has occurred on the vehicle, the contact accident determination unit 510 included in the system 100 determines that the impact is due to a contact accident through a machine learning model. It can be determined whether it is due to a non-contact case or not.

That is, when a vehicle is impacted, for example, the impact of opening and closing the door of the vehicle, the closing of the trunk and the closing of the bonnet, the impact of the wind generated by the passage of another vehicle, And can be the impact caused by the shaking of the bottom. However, since the impact caused by such a non-contact incident does not substantially damage the vehicle, it is not necessary to detect it.

Along with this, even if an impact occurs on the vehicle, it can be determined whether the vehicle has been impacted by the contact accident by filtering the case where the vehicle is impacted by the non-contact incident.

In an exemplary embodiment, whether or not the vehicle has been impacted by a contact accident is determined by the contact accident determination unit 510 for determining the contact accident from the acceleration data before and after the occurrence of the impact stored in the memory unit 300. By extracting an element and using it as an input of a machine learning model, it is possible to make a judgment through the output of the machine learning model.

In an exemplary embodiment, the characteristic elements for determining a contact accident of the vehicle are the effective value (RMS) of the acceleration for each axis before and after the impact occurrence time, the maximum acceleration, the slope of the acceleration attenuation, and the positive and negative signs. Changes in the impact, the duration of the impact, and the like can be included.

FIG. 3 is a graph showing changes in acceleration for each axis depending on the type of impact generated on the vehicle.

With reference to FIG. 3 together, it can be seen that the change in the special acceleration of each axis in a contact accident such as a collision accident has a different pattern from the change in the acceleration for each axis in a non-contact case such as a door closing. ..

For example, the effective value at the acceleration for each axis, the maximum acceleration, the inclination of the attenuation of the acceleration, the change of the positive and negative signs, the duration of the impact, etc. for each type of impact show different aspects in the contact accident and the non-contact incident. Therefore, it may correspond to a characteristic element for distinguishing these. However, the concept of the present invention is not limited to this, and items other than the items exemplified above may also correspond to the above-mentioned characteristic elements.

In an exemplary embodiment, in the learning process of the machine learning model, impact experiments including contact accidents and non-contact incidents with respect to the vehicle are performed a plurality of times, and acceleration data of the vehicle before and after the occurrence of each impact is collected. At this time, for each of the impacts, one of the contact accident and the non-contact incident that causes the impact can be recorded. After that, a feature element is obtained from the collected acceleration data and used as an input of the machine learning model, and one of the causes of the impact recorded corresponding to each acceleration data, that is, a contact accident and a non-contact incident. Is set to be the output of the machine learning model, so that the parameters of the machine learning model can be learned.

In an exemplary embodiment, the learning of the machine learning model as described above, that is, the feature element obtained from the acceleration data collected by the impact experiment is input, and the cause of the impact corresponding to the acceleration data, that is, the contact accident. Learning to output whether it is a non-contact case or a non-contact case can be repeatedly performed to improve the accuracy of the parameter, and accordingly, the machine learning model can be applied to the input and the corresponding output. Optimal parameters can be obtained. By using the machine learning model learned in this way, it may be possible to more accurately determine the cause of the impact of the vehicle, that is, whether it is a contact accident or a non-contact incident.

In an exemplary embodiment, only a portion of the collected acceleration data can be used to determine the feature element, and of the collected acceleration data the acceleration not used to determine the feature element. The data can be used to perform validations that assess the accuracy of the machine learning model.

In an exemplary embodiment, the learning method of the machine learning model may include a decision tree, a support vector machine (SVM), a neural network, deep learning, and the like. can.

In the fourth stage (S140), when the contact accident determination unit 510 determines that the vehicle has been impacted by the contact accident, the contact accident site estimation unit 520 included in the system 100 uses the machine learning model to determine that the vehicle has been impacted. The contact accident site can be estimated.

FIG. 4 is a drawing showing the vehicle divided into 8 or 12 parts on a plane composed of an X-axis and a Y-axis.

With reference to FIG. 4 together, when the contact accident determination unit 510 determines that the vehicle has been impacted by the contact accident, the contact accident site estimation unit 520 uses a machine learning model to classify the vehicle into eight or more. Any one of the 12 parts can be estimated as the contact accident part.

In the learning process of the machine learning model according to the exemplary embodiment, collision experiments can be performed a plurality of times for each part of the vehicle, and acceleration data of the vehicle before and after the occurrence of each collision can be collected. At this time, for each of the collisions, the location where the collision occurs can be recorded. After that, a feature element is obtained from the collected acceleration data and used as an input of the machine learning model, and the collision portion recorded corresponding to each acceleration data is used as an output of the machine learning model. This makes it possible to train the parameters of the machine learning model.

In an exemplary embodiment, the learning of the machine learning model as described above, that is, the learning in which the feature elements obtained from the acceleration data collected by the collision experiment are input and the collision occurrence site corresponding to the acceleration data is output. Iterative execution can be performed to improve the accuracy of the parameters, so that the machine learning model can acquire the most appropriate parameters for the input and the corresponding output. By using the machine learning model learned in this way, it may be possible to make a more accurate determination of the collision site of the vehicle.

In an exemplary embodiment, only a portion of the collected acceleration data can be used to determine the feature element, and of the collected acceleration data the acceleration not used to determine the feature element. The data can be used to perform validations that assess the accuracy of the machine learning model.

The characteristic elements of the collected acceleration data include, for example, the effective value at each axis-specific acceleration by each contact accident site, the maximum acceleration, the inclination of the acceleration attenuation, the change of the positive / negative sign, the duration of the impact, and the like. Can include. However, the concept of the present invention is not limited to this, and items other than the items exemplified above may also correspond to the above-mentioned characteristic elements.

In an exemplary embodiment, the learning method of the machine learning model may include a decision tree, a support vector machine (SVM), a neural network, deep learning, and the like. can.

In the exemplary embodiment, the learning method of the machine learning model used for determining the contact accident in the third stage (S130) and the machine learning model used for estimating the contact accident site in the fourth stage (S140) is , They may be the same as each other, or they may be different from each other.

Further, so far, the contact accident determination in the third stage (S130) and the contact accident site estimation in the fourth stage (S140) have been described using separate machine learning models, respectively. The concept of is not necessarily limited to this.

That is, until now, the first machine learning model in the third stage (S130) has input the characteristic elements for determining the contact accident of the vehicle, outputs the contact accident and the non-contact case, and outputs the contact accident and the non-contact case, and in the fourth stage (S140). The second machine learning model of the above uses the feature element for estimating the contact accident part of the vehicle as an input, the specific part of the vehicle as an output, and the first and second machine learning models learn through their respective learning methods. It was explained that these are utilized in the third stage (S130) and the fourth stage (S140), respectively.

FIG. 5 is a flowchart showing a method for estimating a contact accident portion of a vehicle according to another embodiment of the present invention.

However, referring to FIG. 5, by integrating the first and second machine learning models and using one machine learning model, the third and fourth stages (S130, S140) become the third stage (S130). It may be integrated and carried out.

That is, the feature elements for determining the contact accident and estimating the contact accident portion may both be input of one machine learning model, and the non-contact incident and the contact accident (and the specific portion) may be output.

As described above, according to the vehicle contact accident site estimation method and system 100 according to the embodiment of the present invention, when a parked vehicle is impacted using a machine learning model, non-contact incidents can be filtered. In the case of a contact accident, the contact accident site can be easily estimated.

The vehicle contact accident site estimation method and system according to the embodiment of the present invention have been described above with reference to the drawings, but the above description is exemplary and does not deviate from the technical idea of the present invention. It could be modified and modified by someone with normal knowledge in the relevant art.

100: Vehicle contact accident site estimation system 200: Sensor unit 300: Memory unit 400: Control unit 500: Accident estimation unit 510: Contact accident determination unit 520: Contact accident site estimation unit

Claims (31)

When the acceleration of the vehicle is measured and the magnitude is equal to or greater than the critical value, it is determined that the vehicle has an impact; when it is determined that the vehicle has an impact, before and after the impact occurs. A stage in which a characteristic element for determining a contact accident is extracted from the acceleration data of the machine learning model and used as an input of a machine learning model, and it is determined from the output of the machine learning model whether the vehicle has been impacted by a contact accident; When it is determined that the vehicle has been impacted by a contact accident, a feature element for estimating the contact accident site is extracted from the acceleration data and used as an input of a machine learning model, and the machine learning model is used. A method for estimating a contact accident part of a vehicle, which comprises a step of estimating the contact accident part of the vehicle from the output of. The output of the machine learning model at the stage of determining whether the vehicle has been impacted by a contact accident includes a contact accident and a non-contact incident, and accordingly, a stage of determining whether the vehicle has been impacted by the contact accident. The method for estimating a contact accident site of a vehicle according to claim 1, further comprising filtering the case where the vehicle is impacted by the non-contact incident. The impact of the non-contact incident includes the impact of opening and closing the door of the vehicle, the closing of the trunk and the closing of the bonnet, the impact of the wind generated by the passage of another vehicle, and the impact of the shaking of the bottom. The method for estimating a contact accident site of a vehicle according to claim 2. The vehicle is divided into 8 or 12 parts, and the output of the machine learning model at the stage of estimating the contact accident part of the vehicle includes any one of the divided parts of the vehicle. The vehicle according to claim 1, wherein the step of estimating the contact accident part of the vehicle includes presuming that an impact due to the contact accident has occurred in any one of the classified parts of the vehicle. Contact accident site estimation method. The method for estimating a contact accident site of a vehicle according to claim 1, wherein the acceleration of the vehicle is measured by a three-axis accelerometer mounted in the vehicle. The vehicle contact accident site estimation method according to claim 5, wherein the acceleration of the vehicle is measured when the vehicle is parked. The characteristic elements for determining the contact accident are the effective value (RMS) of the acceleration for each axis before and after the impact generation time measured by the three-axis accelerometer, the maximum acceleration, the slope of the acceleration attenuation, and the positive and negative signs. The vehicle contact accident site estimation method according to claim 5, which comprises at least one of the change and the duration of the impact. The characteristic elements for estimating the contact accident site are the effective value (RMS) of the acceleration for each axis before and after the impact generation time measured by the three-axis accelerometer, the maximum acceleration, the inclination of the attenuation of the acceleration, and the positive and negative. The method for estimating a contact accident site of a vehicle according to claim 5, which comprises at least one of a change in reference numeral and a duration of the impact. The learning methods of the machine learning model used in the stage of determining whether the vehicle has been impacted by a contact accident or the stage of estimating the contact accident site of the vehicle include a decision tree, a support vector machine (SVM), and a neural circuit. The vehicle contact accident site estimation method according to claim 1, which comprises at least one of a net and deep learning. The vehicle contact accident according to claim 9, wherein the learning methods of the machine learning models used in the stage of determining whether the vehicle has been impacted by the contact accident and the stage of estimating the contact accident portion of the vehicle are different from each other. Site estimation method. The learning process of the machine learning model used in the stage of estimating the contact accident part of the vehicle is to carry out a collision experiment for each part of the vehicle and collect acceleration data of the vehicle before and after the occurrence of each collision. , The stage of recording the occurrence site of each collision; and the feature element obtained from the collected acceleration data is used as the input of the machine learning model, and corresponds to each acceleration data of the recorded collision sites. The vehicle according to claim 1, wherein the machine learning model learns appropriate parameters for the input and the corresponding output, including a step of making the collision site an output of the machine learning model. Contact accident site estimation method. ii) The vehicle contact accident site estimation method according to claim 11, wherein the step is repeatedly performed, and the accuracy of the parameters is improved accordingly. Only a part of the collected acceleration data is used to obtain the feature element, and the learning process of the machine learning model used at the stage of estimating the contact accident site of the vehicle is the collected acceleration data. The vehicle contact accident site estimation method according to claim 11, further comprising a verification step of evaluating the accuracy of the machine learning model using acceleration data that is not used to obtain the feature element. In the learning process of the machine learning model used in the vehicle contact accident determination stage, impact experiments including contact accidents and non-contact incidents with respect to the vehicle are performed, and acceleration data of the vehicle before and after the occurrence of each impact is obtained. The stage of collecting and recording the contact accident or non-contact incident that is the cause of each impact; and the characteristic element obtained from the collected acceleration data is used as the input of the machine learning model, and the recorded contact accident. And, one of the non-contact cases corresponding to each acceleration data includes a step of making the output of the machine learning model, and accordingly, the machine learning model has parameters appropriate for the input and the corresponding output. The method for estimating a contact accident site of a vehicle according to claim 1, wherein the method is learned. ii) The vehicle contact accident site estimation method according to claim 14, wherein the steps are repeatedly performed, and the accuracy of the parameters is improved accordingly. When the acceleration of the vehicle is measured and the value is greater than or equal to the critical value, it is determined that the vehicle has an impact; and when it is determined that the vehicle has an impact, the time when the impact occurs. Features for determining contact accidents and estimating contact accident sites are extracted from the front and rear acceleration data and used as input to the machine learning model, and the vehicle is impacted by the contact accident from the output of the machine learning model. A method for estimating a contact accident part of a vehicle, which includes a step of determining whether or not the vehicle has been hit and estimating the contact accident part of the vehicle when it is determined that the vehicle has been impacted by a contact accident. The vehicle contact accident site estimation method according to claim 16, wherein the acceleration of the vehicle is measured by a three-axis accelerometer mounted in the vehicle when the vehicle is parked. The characteristic elements for determining the contact accident and estimating the contact accident site are the effective value (RMS) of the acceleration for each axis before and after the impact occurrence time, the maximum acceleration, and the attenuation of the acceleration measured by the three-axis accelerometer. The method for estimating a contact accident site of a vehicle according to claim 17, which comprises at least one of a tilt, a change of a positive / negative sign, and a duration of the impact. The learning process of the machine learning model includes a contact accident and a non-contact incident with the vehicle, but in the case of the contact accident, an impact experiment including a collision of each part of the vehicle is performed and the time when each impact occurs. The stage of collecting the acceleration data of the front and rear vehicles and recording the contact accident or non-contact incident that is the cause of each impact and the location of each collision; and obtaining the characteristic element from the collected acceleration data. Is used as the input of the machine learning model, and one of the recorded contact accidents and the non-contact incidents corresponding to the respective acceleration data is used as the output of the machine learning model. Among the collision sites, the collision site corresponding to each acceleration data is used as the output of the machine learning model, and the machine learning model learns the input and the corresponding parameters for the output. The vehicle contact accident site estimation method according to claim 14. ii) The vehicle contact accident site estimation method according to claim 19, wherein the step is repeatedly performed, and the accuracy of the parameters is improved accordingly. Sensor unit that measures the acceleration of the vehicle; Memory unit that stores the acceleration data measured by the sensor unit; When the acceleration measured by the sensor unit has a magnitude equal to or greater than the critical value, an impact is generated on the vehicle. Control unit for determining that: When the control unit determines that an impact has occurred on the vehicle, a feature element for determining a contact accident from the acceleration data stored in the memory unit before and after the occurrence of the impact. Is extracted and used as an input of a machine learning model, and a contact accident determination unit that determines whether the vehicle has been impacted by a contact accident from the output of the machine learning model; and the vehicle is impacted by a contact accident. When it is determined by the contact accident determination unit, a feature element for estimating the contact accident portion is extracted from the acceleration data stored in the memory unit and used as an input of the machine learning model. A vehicle contact accident part estimation system including a contact accident part estimation unit that estimates the contact accident part of the vehicle from the output of a machine learning model. The output of the machine learning model used by the contact accident determination unit includes a contact accident and a non-contact incident, and it is determined whether the vehicle is impacted by the contact accident according to the non-contact incident. The vehicle contact accident site estimation system according to claim 21, which comprises filtering the case of receiving an impact. The vehicle is divided into 8 or 12 parts, and the output of the machine learning model in the contact accident part estimation unit includes any one of the divided parts of the vehicle. The vehicle contact accident site according to claim 21, wherein estimating the vehicle contact accident site includes presuming that an impact due to a contact accident has occurred in any one of the classified parts of the vehicle. Estimate system. 21. The vehicle contact according to claim 21, wherein the sensor unit includes a 3-axis accelerometer mounted in the vehicle, and the acceleration of the vehicle is measured by the 3-axis accelerometer when the vehicle is parked. Accident site estimation system. The characteristic element for determining the contact accident or the characteristic element for estimating the contact accident site is the effective value (RMS) of the acceleration for each axis before and after the impact generation time measured by the three-axis accelerometer, and the maximum. The vehicle contact accident site estimation system according to claim 24, which comprises at least one of acceleration, acceleration attenuation gradient, positive / negative sign change, and impact duration. The learning method of the machine learning model used in the contact accident determination unit or the contact accident site estimation unit is at least one of a decision tree, a support vector machine (SVM), a neural network, and deep learning. The vehicle contact accident site estimation system according to claim 21, including the above. In the learning process of the machine learning model used in the contact accident part estimation unit of the vehicle, a collision experiment is performed for each part of the vehicle, and acceleration data of the vehicle before and after the occurrence of each collision is collected, and the acceleration data of the vehicle is collected. The stage of recording each collision occurrence site; and the collision site corresponding to each acceleration data among the recorded collision sites by obtaining a feature element from the collected acceleration data and using this as an input of the machine learning model. 21. The vehicle contact according to claim 21, wherein the machine learning model learns appropriate parameters for the input and the corresponding output. Accident site estimation system. ii) The vehicle contact accident site estimation system according to claim 27, wherein the step is repeatedly performed, and the accuracy of the parameters is improved accordingly. In the learning process of the machine learning model used in the contact accident determination unit of the vehicle, an impact experiment including a contact accident and a non-contact incident with the vehicle is performed, and acceleration data of the vehicle before and after the occurrence of each impact is obtained. The stage of collecting and recording the contact accident or non-contact incident that is the cause of each impact; and the characteristic element obtained from the collected acceleration data is used as the input of the machine learning model, and the contact accident and the non-contact accident are recorded. Including a step in which one of the contact cases corresponding to each acceleration data is used as the output of the machine learning model, the machine learning model learns the input and the corresponding parameters for the output. 21. The vehicle contact accident site estimation system according to claim 21. ii) The vehicle contact accident site estimation system according to claim 29, wherein the step is repeatedly performed, and the accuracy of the parameters is improved accordingly. A sensor unit that measures the acceleration of a vehicle; a memory unit that stores acceleration data measured by the sensor unit; and an impact is generated on the vehicle when the acceleration measured by the sensor unit has a magnitude equal to or greater than a critical value. The characteristic element for contact accident determination is extracted from the acceleration data before and after the occurrence of the impact stored in the memory unit and used as the input of the machine learning model, and the machine learning model is used. Contact accident determination unit that determines whether the vehicle has been impacted by a contact accident; and if the contact accident determination unit determines that the vehicle has been impacted by a contact accident, it is stored in the memory unit. The characteristic element for estimating the contact accident part is extracted from the obtained acceleration data and used as the input of the machine learning model, and the contact accident part estimation of the vehicle is estimated from the output of the machine learning model. A vehicle contact accident site estimation system including an accident estimation section including a section.

Images