JP7156741B1 - Wound detection system, wound detection method and program - Google Patents

Abstract

An object of the present invention is to quickly identify the occurrence of a scratch or the like on a vehicle and the area of occurrence without depending on various conditions. A flaw detection system (1) comprising a processing device (100) provided inside a vehicle and at least a plurality of microphones (200A, 200B, 200C, 200D), wherein the processing device (100) includes a plurality of microphones (200A, 200B). , 200C, and 200D are stored in a storage area determined for each microphone; and a detection unit 130 for detecting the impact sound data based on the sound data of all the microphones 200A, 200B, 200C, and 200D when the detection unit 130 detects impact sound data from the audio data of one microphone. and an estimating unit 140 for estimating the occurrence area of . [Selection drawing] Fig. 2

Description

The present invention relates to a flaw detection system, flaw detection method, and program.

In recent years, along with the decrease in preference for private cars, especially among young people, the number of cases of using rental cars, car sharing, etc. is increasing.
In such rental cars and car sharing, multiple users use the vehicle, and if the rental vehicle, etc. is damaged during use by the user, the user bears the repair costs. Therefore, the person in charge compares the condition of the outer circumference of the rental vehicle when the rental vehicle is rented and when it is returned. I'm trying to make sure.
However, there are problems that this kind of confirmation is labor intensive and that it is difficult to accurately detect the occurrence of scratches or the like.

As a means for solving the above problems, an impact detection means mounted on a rental vehicle for detecting an impact on the vehicle, an imaging means mounted on the rental vehicle for imaging at least the outer periphery of the vehicle, and an imaging means recording means for recording the image data captured by the image data; and image data captured by the imaging means in response to the detection of the impact by the impact detection means are associated with information related to identification of the user of the rental vehicle. and a control means for storing data in a recording means (see, for example, Patent Document 1).

JP 2006-195715 A

However, the technique described in Patent Document 1 detects the situation when a scratch or the like occurs from image data, particularly a minor scratch or the like. The image quality of the image data is greatly affected by the image pickup distance, image pickup angle, vehicle type, etc., so there was a problem that it was not possible to detect scratches on the vehicle with a high probability in any case. rice field.

Moreover, even if image data of the flaw or the like can be obtained, specifying the area in which the flaw or the like is generated still requires additional manual effort.

Accordingly, the present invention has been made in view of the above-mentioned problems, and provides a flaw detection system and a flaw detection system that can accurately identify the occurrence of a flaw on a vehicle and the area where the flaw occurs, regardless of various conditions. An object is to provide a detection method and program.

Mode 1: One or more embodiments of the present invention are a wound detection system comprising a processing device provided inside a vehicle and at least a plurality of microphones, wherein the processing device comprises a plurality of microphones. a storage unit that stores voice data input from the microphone in a storage area determined for each microphone; and machine learning using a learning model is performed on the voice data stored in the storage unit to generate impact sound data. and a detection unit for detecting, when the impact sound data is detected from the audio data of one of the microphones in the detection unit, based on the audio data of all the microphones, the generation area of the impact sound data and an estimating unit for estimating .

Mode 2: One or more embodiments of the present invention propose a flaw detection system, wherein the learning model is a learning model obtained by learning the voice data of the vehicle.

Mode 3: One or more embodiments of the present invention propose a flaw detection system in which the learning model is different for each model of the vehicle.

Mode 4: In one or more embodiments of the present invention, the detection unit detects that a learning result for the voice data and a learning result for the voice data during normal running of the vehicle exceed a predetermined threshold. A flaw detection system is proposed that detects that the sound data is the impact sound data when the deviation is the degree.

Mode 5: In one or more embodiments of the present invention, the estimation unit detects the impact sound data from the audio data of the one microphone when the detection unit detects the impact sound data from the audio data of the one microphone. proposed a flaw detection system that estimates the area where the impact sound data is generated from the difference in sound volume.

Mode 6: In one or more embodiments of the present invention, when the detecting unit detects the impact sound data from the audio data of the one microphone, the estimating unit detects each microphone of the audio data We have proposed a flaw detection system that estimates the generation area of the impact sound data from the difference in arrival time.

Mode 7: In one or more embodiments of the present invention, the estimation unit detects the impact sound data from the audio data of the one microphone when the detection unit detects the impact sound data from the audio data of the one microphone. proposes a flaw detection system that estimates the generation area of the impact sound data from the difference in the degree of divergence of the learning results with respect to.

Mode 8: In one or more embodiments of the present invention, the processing device comprises a communication unit for transmitting information to an external device, and a network detection unit for detecting a predetermined network, the network detection unit comprising: A damage detection system has been proposed in which, when the predetermined network is detected, the communication unit transmits a message indicating which area of the vehicle may be damaged to the external device.

Mode 9: One or more embodiments of the present invention propose a wound detection system including a server and transmitting data including the impact sound data acquired while the vehicle is running to the server.

Mode 10: One or more embodiments of the present invention provide a vehicle flaw detection method in a vehicle flaw detection system comprising a processing device provided inside the vehicle and at least three or more microphones, a first step in which the processing device stores the audio data input from the at least three or more microphones in a storage area determined for each microphone; and the processing device stores the audio data stored in the storage unit On the other hand, a second step of performing machine learning using a learning model to detect impact sound data; and a third step of estimating an impact sound data generation area based on the audio data of all microphones.

Mode 11: One or more embodiments of the present invention is a program for causing a computer to execute a flaw detection method in a flaw detection system comprising a processor provided inside a vehicle and at least three or more microphones. a first step in which the processing device stores audio data input from the at least three or more microphones in a storage area determined for each microphone; a second step of performing machine learning using a learning model on the sound data to detect impact sound data; A program is proposed for causing a computer to execute a third step of estimating an impact sound data generation area based on the audio data of all microphones when the impact sound data is detected.

Advantageous Effects of Invention According to one or more embodiments of the present invention, there is an effect that it is possible to accurately and promptly identify the occurrence of a scratch or the like on a vehicle and the occurrence area regardless of various conditions.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows the outline|summary of the wound detection system which concerns on the 1st Embodiment of this invention. It is a figure which shows the structure of the processing apparatus which concerns on the 1st Embodiment of this invention. It is a figure which shows the outline|summary of the memory|storage part which concerns on the 1st Embodiment of this invention. It is a figure which shows the estimation method of the estimation part which concerns on the 1st Embodiment of this invention. It is a figure which shows the structure of the detection part which concerns on the 1st Embodiment of this invention. It is a figure which shows the processing flow of the wound detection system which concerns on the 1st Embodiment of this invention. It is a figure which shows the structure of the processing apparatus which concerns on the 2nd Embodiment of this invention. It is a figure which shows the processing flow of the flaw detection system which concerns on the 2nd Embodiment of this invention. It is a figure which shows the estimation method of the estimation part based on the 2nd Embodiment of this invention. It is a figure which shows the structure of the processing apparatus which concerns on the 3rd Embodiment of this invention. It is a figure which shows the processing flow of the flaw detection system which concerns on the 3rd Embodiment of this invention. It is a figure which shows the estimation method of the estimation part which concerns on the 3rd Embodiment of this invention.

An embodiment of the present invention will be described below with reference to FIGS. 1 to 12. FIG.

<First embodiment>
A wound detection system 1 according to the present embodiment will be described with reference to FIGS. 1 to 6. FIG.

<Configuration of flaw detection system 1>
As shown in FIG. 1, the wound detection system 1 according to the present embodiment includes a processing device 100 provided inside a vehicle VC, and microphones 200A, 200B, 200C, and 200D.

The processing device 100 performs machine learning using a learning model on audio data input from a plurality of microphones 200A, 200B, 200C, and 200D, detects impact sound data, and extracts audio data from one microphone. When the impact sound data is detected from the microphone, the area where the impact sound data is generated is estimated based on the audio data of all the microphones.
A detailed configuration of the processing device 100 will be described later.

The microphones 200A, 200B, 200C, and 200D are devices that collect audio data while the vehicle VC is running.
The microphones 200A, 200B, 200C, and 200D are arranged, for example, at four corners inside the vehicle VC, and audio data collected by each of the microphones 200A, 200B, 200C, and 200D is output to the processing device 100.
Also, the ON/OFF state of each of the microphones 200A, 200B, 200C, and 200D is controlled by the processing device 100. FIG.
The microphones 200A, 200B, 200C, and 200D may have directivity, or may be placed on a pedestal and rotated.

<Configuration of processing device 100>
As shown in FIG. 2, the processing device 100 includes a voice data information input unit 110, a storage unit 120, a detection unit 130, an estimation unit 140, a communication unit 150, a network detection unit 160, and a data transmission/reception unit. 170 and .

Voice data information input unit 110 inputs voice data obtained from microphones 200A, 200B, 200C, and 200D.
The voice data information input unit 110 stores the voice data information attached to the voice data to which the time information from the clock unit (not shown) indicating the input time of the voice data information is input, in a predetermined storage area in the storage unit 120 to be described later. do.

Storage unit 120 stores the audio data input by audio data information input unit 110 in storage areas determined for each of microphones 200A, 200B, 200C, and 200D.
As shown in FIG. 3, the storage unit 120 includes an audio data information storage area 1, an audio data information storage area 2, an audio data information storage area 3, an audio data information storage area 4, a learning model storage area, and a normal audio data storage area. , an impact sound audio data storage area, a learning result storage area, and the like.
As for the audio data information from the microphones 200A, 200B, 200C, and 200D, for example, information on the microphone 200A is stored in the audio data information storage area 1, and information on the microphone 200B is stored in the audio data information storage area 2. , information about the microphone 200C is stored in the audio data information storage area 3, and information about the microphone 200D is stored in the audio data information storage area 4. FIG.

The detection unit 130 detects impact sound data by executing machine learning using a learning model on the audio data information stored in the storage unit 120 .
Note that the learning model used for machine learning in the detection unit 130 is obtained from the server 300 via the data transmission/reception unit 170 described later, for example, and is stored in the learning model storage area.
Here, the learning model is a learning model obtained by learning voice data of a vehicle, and differs for each type of vehicle.
Specifically, the detection unit 130 detects that the sound data is impact sound data when the degree of divergence between the learning result for the sound data and the learning result for the sound data during normal running of the vehicle exceeds a predetermined threshold value. detect something.
It should be noted that the degree of divergence can be appropriately changed according to the correct answer rate of machine learning.
Further, the detection unit 130 detects that the audio data is impact sound data when the degree of divergence between the learning result for the audio data and the learning result for the impact sound data is equal to or less than a predetermined threshold value. good too.
A detailed configuration of the detection unit 130 will be described later.

The learning data used to generate the learning model is obtained from not only the running sound that can be observed inside the vehicle during normal driving, but also scrap cars obtained from scrap factories that can be scratched without causing problems. (for example, 20 Km/h, 40 Km/h, 60 Km/h) and the collision sound of the vehicle is also included.
A tool such as a slingshot is used to collide with pebbles, and the speed at the time of collision is recorded using a speed gun. A small personal computer and a microphone array are used to record anomalous sounds in the same way as usual.
In addition, since it is not always possible to obtain a scrapped vehicle in a state where it can be driven, if the scrapped vehicle cannot be driven, the scratch generation sound does not contain the noise detailed below. Pseudo abnormal data is obtained by superimposing the scratch occurrence sound.
The running sound includes (1) noise generated outside the vehicle, such as road surface conditions and rain, and (2) noise generated inside the vehicle, such as voices of passengers and sounds emitted by devices used. By collecting these noise components and adding them to AI learning data as training data, it is expected that the accuracy of detecting the sound when a scratch occurs can be improved.
Here, (1) noise generated outside the vehicle can be mainly classified into three types: (i) vibration noise between the road surface and tires, (ii) engine noise, and (iii) noise due to weather conditions.
Further, by collecting a sufficient amount of data, noise generated outside the vehicle can be made statistically negligible.
Regarding the road surface and tire vibration noise, the model number of the tire, the period of use of the tire, the distance traveled since the tire was installed, and the road surface type are used as training data.
Here, the period of use and the mileage do not necessarily have to be exact values, and are given as discrete data of limited types, such as within a week, within a month, and within a year.
Also, the road surface type gives the type of road surface at the travel point as a categorical variable. Categorical variables are, for example, asphalt, dirt, gravel roads, and the like. Road surface type data is created by annotation.
For the engine sound, the model number of the engine, the number of months since the car inspection, etc. are given as training data. It is also conceivable to give the engine rotation amount as training data.
For noise due to weather conditions, temperature, cloudiness, rainfall in the past hour, hourly average rainfall on the previous day, etc. are provided as teaching data. Here, the reason why the amount of rainfall of the previous day is added to the data is to consider the muddy road surface.
There are mainly two types of noise generated in the passenger compartment: (i) passenger conversation, and (ii) noise caused by the use of in-vehicle equipment.
Further, by collecting a sufficient amount of data, noise generated outside the vehicle can be made statistically negligible.
Concerning the conversation of passengers, the presence or absence of conversation is given as an annotation. This is category data having two values, "0" when not speaking and "1" when speaking.
Regarding the use of audio equipment among in-vehicle equipment, the presence or absence of music playback is given as an annotation. This is category data having two values, "0" when not speaking and "1" when speaking.
In addition, regarding the use of air conditioners among the in-vehicle devices, the presence or absence of operation of the devices is given as an annotation. This is category data having two values, "0" when not operating and "1" when operating.

When the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140 estimates the generation area of the impact sound data based on the audio data of all the microphones.
Specifically, when the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140 estimates the generation area of the impact sound data from the difference in volume of the audio data of each microphone. .
4, FIG. 4 shows that when the detection unit 130 detects impact sound data from the audio data of one microphone, the volume of the audio data of the microphones 200B and 200D changes to the volume of the microphone 200A. , is sufficiently larger than the volume of the audio data of the microphone 200C.
Therefore, an example of estimating the generation area of the impact sound data based on the sound volume of the audio data of the microphones 200B and 200D is shown.
Here, the ratio of the volume of the audio data of the microphone 200B to the volume of the audio data of the microphone 200D is 3:2, and the volume of the audio data of the microphone 200B is 1.5 times the volume of the audio data of the microphone 200D. If it is large, the distance to the impact sound data generating area is 1.5 times longer for the microphone 200D than for the microphone 200B.
Therefore, a circle having a radius corresponding to this ratio is drawn from the center of the microphone 200D and the microphone 200B, and the estimating unit 140 designates the vicinity of the location where these two arcs and the body line BD of the vehicle are close to each other as the impact sound data generation area. presume.

The communication unit 150 transmits information on the processing device 100 to an external device.
Here, the information transmitted from the communication unit 150 to the external device includes the occurrence time of the impact sound, the area where the impact sound data is generated, and the like.
When the network detection unit 160, which will be described later, detects a predetermined network, the communication unit 150 transmits a message such as the time when the impact sound was generated and the area where the impact sound data was generated to the external device.

Network detection unit 160 detects a predetermined network.
Here, the predetermined network can be Wi-Fi (registered trademark) or the like, but short-range wireless communication such as Bluetooth (registered trademark) may be used regardless of the network.

The data transmission/reception unit 170 transmits data including impact sound data acquired while the vehicle is running to the server 300 .
The information that the data transmission/reception unit 170 transmits to the server 300 may include normal speech data, learning results, and the like.

<Configuration of detection unit 130>
The detection unit 130 includes a detection target data set 131, a learned data set 132, and a machine learning unit 133, as shown in FIG.

The detection target data set 131 is a positive example data set used for deep learning by the machine learning unit 133, and is stored in the audio data information storage area 1, the audio data information storage area 2, and the audio data information storage area 3 of the storage unit 120. , the data group read from the voice data information storage area 4 is stored in the detection object data set 131 .

The learned data set 132 is a so-called learning model, and is a data set that has been learned in advance based on the voice data from the microphones 200A, 200B, 200C, and 200D, and is read from the learning model storage area of the storage unit 120. The learned model is stored in the trained data set 132 .

The machine learning unit 133 uses the voice data obtained from the microphones 200A, 200B, 200C, and 200D input to the voice data information input unit 110 as input data, and uses the trained model stored in the trained data set 132 to perform machine learning;
More specifically, the machine learning unit 133 calculates a score for, for example, similarity or deviation between the detection target data set 131 and the learned data set 132 .

<Processing of processing device 100>
The processing of the processing device 100 according to this embodiment will be described with reference to FIG.

The voice data information input unit 110 of the processing device 100 inputs voice data obtained from the microphones 200A, 200B, 200C, and 200D at predetermined acquisition timings.
Then, the audio data information input unit 110 stores the audio data information attached to the audio data to which the time information from the clock unit (not shown) indicating the input time of the audio data information is input, in a predetermined storage area in the storage unit 120. (step S110).

The detection unit 130 performs machine learning using a learning model on the audio data information stored in the storage unit 120 to detect impact sound data (step S120).
Specifically, the detection unit 130 detects that the sound data is impact sound data when the degree of divergence between the learning result for the sound data and the learning result for the sound data during normal running of the vehicle exceeds a predetermined threshold value. detect something.

When the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140 estimates the generation area of the impact sound data based on the audio data of all the microphones (step S130). .
Specifically, when the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140 estimates the generation area of the impact sound data from the difference in volume of the audio data of each microphone. .

The network detection unit 160 detects whether or not the vehicle has entered the communication area of a predetermined network (step S140).
When the network detection unit 160 determines that the vehicle is not within the communication area of the predetermined network ("NO" in step S140), the network detection unit 160 shifts to the standby mode.

On the other hand, when network detection unit 160 determines that the vehicle is within the communication area of the predetermined network (“YES” in step S140), communication unit 150 transmits the time and A message such as an impact sound data generation area is transmitted to the external device (step S150).

Then, the data transmission/reception unit 170 transmits data including impact sound data acquired while the vehicle is running to the server 300, and ends the process (step S160).

<Action/effect>
As described above, the processing device 100 in the flaw detection system 1 according to the present embodiment determines audio data input from the plurality of microphones 200A, 200B, 200C, and 200D for each of the microphones 200A, 200B, 200C, and 200D. The detection unit 130 detects impact sound data by executing machine learning using a learning model on the stored audio data.
Therefore, regardless of various conditions, it is possible to accurately and quickly detect the occurrence of a scratch or the like on the vehicle.

In addition, the processing device 100 in the flaw detection system 1 according to the present embodiment stores audio data input from the plurality of microphones 200A, 200B, 200C, and 200D in storage areas determined for each of the microphones 200A, 200B, 200C, and 200D. Then, when the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140 determines the impact sound data based on the audio data of all the microphones 200A, 200B, 200C, and 200D. Estimate the area of occurrence.
Therefore, regardless of various conditions, it is possible to quickly identify the occurrence of a scratch or the like on the vehicle and the occurrence area.

Also, the learning model in the processing device 100 in the flaw detection system 1 according to the present embodiment is a learning model obtained by learning voice data of the vehicle.
That is, the learning model according to the present embodiment is a learning model obtained by learning a huge amount of voice data during normal running of the vehicle and impact sound data when a scratch or the like occurs.
Therefore, regardless of various conditions, it is possible to accurately and quickly detect the occurrence of a scratch or the like on the vehicle.

In addition, in the processing device 100 in the flaw detection system 1 according to the present embodiment, the learning model differs for each type of vehicle.
In other words, depending on the type of vehicle, the material and thickness of the vehicle body, the type of paint, the size of the vehicle interior space, etc., differ. difference occurs.
However, in the processing device 100 of the present embodiment, learning models are prepared for each type of vehicle.
Therefore, regardless of the type of vehicle, even under various conditions, it is possible to quickly identify the occurrence of a scratch or the like on the vehicle and the occurrence area.

In addition, in the processing device 100 in the flaw detection system 1 according to the present embodiment, the detection unit 130 determines the degree of divergence between the learning result of the voice data and the learning result of the voice data during normal running of the vehicle exceeding a predetermined threshold. , it is detected that the audio data is impact sound data.
That is, the detection unit 130 compares the learning result of the voice data with the learning result of the voice data during normal running of the vehicle, and if the degree of deviation of the score exceeds a predetermined threshold, , detects that the audio data is impact sound data.
Therefore, regardless of the type of vehicle, even under various conditions, it is possible to quickly identify the occurrence of a scratch or the like on the vehicle and the occurrence area.

In addition, in the processing device 100 in the flaw detection system 1 according to the present embodiment, the estimation unit 140 detects each microphone 200A, 200B, and 200C when the detection unit 130 detects impact sound data from voice data of one microphone. , 200D, the area where the impact sound data is generated is estimated.
That is, the volume of the audio data captured by the microphones 200A, 200B, 200C, and 200D varies in proportion to the distance from the scratched area.
Using this relationship, it is possible to estimate the impact sound data generation area by converting the difference in volume of the sound data of the microphones 200A, 200B, 200C, and 200D into the distance to the scratch generation area.
Therefore, regardless of the type of vehicle, even under various conditions, it is possible to quickly identify the occurrence of a scratch or the like on the vehicle and the occurrence area.

Further, in the processing device 100 in the wound detection system 1 according to the present embodiment, when the network detection unit 160 detects a predetermined network, the communication unit 150 determines which area of the vehicle may be damaged. A message to that effect is sent to the external device.
In other words, in the case of a rental car, before returning the rental car, a message is sent to the operator indicating which area of the vehicle may have been damaged.
Therefore, after returning the rental car, the operator can quickly check whether there is a scratch or not, and can clarify where responsibility lies.

In addition, in the processing device 100 in the wound detection system 1 according to the present embodiment, the data transmission/reception unit 170 transmits to the server 300 data including impact sound data acquired while the vehicle is running.
Therefore, the server 300 can optimize the learning model by re-learning using the data received from the processing device 100 and including the impact sound data acquired while the vehicle is running.

<Modification 1>
In the present embodiment, four microphones 200A, 200B, 200C, and 200D are arranged in the vehicle as an example, but if there are two or more microphones, it is also possible to identify the scratched area.
In other words, the greater the number of microphones, the better the accuracy of identifying the scratched area. If so, the cost of the entire system can be reduced by reducing the number of microphones.

<Modification 2>
In this embodiment, in order to facilitate understanding, the estimating unit 140 converts the difference in the sound volume of the audio data of the microphones 200A, 200B, 200C, and 200D into the distance to the area where the scratch occurs, and converts each , as radii from the centers of the microphones 200A, 200B, 200C, and 200D.
However, it is also possible to estimate the scratch occurrence area from the vicinity of the body line BD of the vehicle and the intersection of the spheres drawn as radii from the centers of the microphones 200A, 200B, 200C, and 200D for each converted distance. .
By doing so, it is possible to estimate the scratch occurrence area in the three-dimensional space, so that the accuracy of estimating the scratch occurrence area can be improved.

<Second embodiment>
A wound detection system 1A according to the present embodiment will be described with reference to FIGS. 7 to 9. FIG.

<Configuration of flaw detection system 1A>
A wound detection system 1A according to the present embodiment is provided inside a vehicle VC and includes a processing device 100A and microphones 200A, 200B, 200C, and 200D.

The processing device 100A performs machine learning using a learning model on audio data input from a plurality of microphones 200A, 200B, 200C, and 200D, detects impact sound data, and extracts audio data from one microphone. When the impact sound data is detected from the microphone, the area where the impact sound data is generated is estimated based on the audio data of all the microphones.
In this embodiment, in particular, when the detection unit 130 detects impact sound data from the audio data of one microphone, the processing device 100A detects the impact sound data based on the difference in arrival time of the audio data to each microphone. Estimate the area of occurrence.

<Configuration of processing device 100A>
As shown in FIG. 7, the processing device 100A includes a voice data information input unit 110, a storage unit 120, a detection unit 130, an estimation unit 140A, a communication unit 150, a network detection unit 160, and a data transmission/reception unit. 170 and .
It should be noted that the components denoted by the same reference numerals as those of the first embodiment have the same functions, and detailed description thereof will be omitted.

When the detection unit 130 detects impact sound data from the audio data of one microphone, the estimation unit 140A estimates the generation area of the impact sound data based on the audio data of all the microphones.
Specifically, when detection unit 130 detects impact sound data from audio data of one microphone, estimating unit 140A determines the generation area of impact sound data based on the difference in arrival time of the audio data to each microphone. presume.
More specifically, with reference to FIG. 8, FIG. 8 shows that when the detection unit 130 detects impact sound data from the audio data of one microphone, the arrival time of the audio data to the microphones 200B and 200D is It is sufficiently faster than the volume of audio data to the microphones 200A and 200C.
Therefore, an example of estimating the generation area of the impact sound data based on the arrival time of the audio data to the microphones 200B and 200D is shown.
Here, the ratio of the arrival time of the audio data to the microphone 200B and the arrival time of the audio data to the microphone 200D is 2:3, and the arrival time of the audio data to the microphone 200B is the same as the arrival time of the audio data to the microphone 200D. Assuming that the arrival time is 1.5 times faster than the arrival time, the distance to the impact sound data generating area is 1.5 times longer for the microphone 200D than for the microphone 200B.
Therefore, a circle having a radius corresponding to this ratio is drawn from the center of the microphone 200D and the microphone 200B, and the estimating unit 140A designates the vicinity of the location where the two arcs and the body line BD of the vehicle are close to each other as the impact sound data generation area. presume.

<Processing of processing device 100A>
The processing of the processing device 100A according to this embodiment will be described with reference to FIG.

The audio data information input unit 110 of the processing device 100A inputs audio data obtained from the microphones 200A, 200B, 200C, and 200D at predetermined acquisition timings.
Then, the audio data information input unit 110 stores the audio data information attached to the audio data to which the time information from the clock unit (not shown) indicating the input time of the audio data information is input, in a predetermined storage area in the storage unit 120. (step S110).

The detection unit 130 performs machine learning using a learning model on the audio data information stored in the storage unit 120 to detect impact sound data (step S120).
Specifically, the detection unit 130 detects that the sound data is impact sound data when the degree of divergence between the learning result for the sound data and the learning result for the sound data during normal running of the vehicle exceeds a predetermined threshold value. detect something.

When the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140A estimates the generation area of the impact sound data based on the audio data of all the microphones (step S210). .
Specifically, when detection unit 130 detects impact sound data from audio data of one microphone, estimating unit 140A determines the generation area of impact sound data based on the difference in arrival time of the audio data to each microphone. presume.

The network detection unit 160 detects whether or not the vehicle has entered the communication area of a predetermined network (step S140).
When the network detection unit 160 determines that the vehicle is not within the communication area of the predetermined network ("NO" in step S140), the network detection unit 160 shifts to the standby mode.

On the other hand, when network detection unit 160 determines that the vehicle is within the communication area of the predetermined network (“YES” in step S140), communication unit 150 transmits the time and A message such as an impact sound data generation area is transmitted to the external device (step S150).

Then, the data transmission/reception unit 170 transmits data including impact sound data acquired while the vehicle is running to the server 300, and ends the process (step S160).

<Action/effect>
As described above, the estimating unit 140A of the processing device 100A in the flaw detection system 1 according to the present embodiment, when the detecting unit 130 detects impulsive sound data from the audio data of the microphone 1, Based on the difference in arrival time to the microphone, the area where the impact sound data is generated is estimated.
In other words, the arrival time of the audio data to the microphones 200A, 200B, 200C, and 200D differs in proportion to the distance from the scratched area.
By using this relationship, the impact sound data generation area can be estimated by converting the difference in arrival time of the sound data to each of the microphones 200A, 200B, 200C, and 200D into the distance to the scratch generation area. can.
Therefore, regardless of the type of vehicle, even under various conditions, it is possible to quickly identify the occurrence of a scratch or the like on the vehicle and the occurrence area.

<Modification 3>
In this embodiment, in order to facilitate understanding, the estimation unit 140A converts the difference in sound volume of the audio data of the microphones 200A, 200B, 200C, and 200D into the distance to the scratch occurrence area, and converts each , as radii from the centers of the microphones 200A, 200B, 200C, and 200D.
However, it is also possible to estimate the scratch occurrence area from the vicinity of the body line BD of the vehicle and the intersection of the spheres drawn as radii from the centers of the microphones 200A, 200B, 200C, and 200D for each converted distance. .
By doing so, it is possible to estimate the scratch occurrence area in the three-dimensional space, so that the accuracy of estimating the scratch occurrence area can be improved.

<Third Embodiment>
A wound detection system 1B according to the present embodiment will be described with reference to FIGS. 10 to 12. FIG.

<Configuration of flaw detection system 1B>
A wound detection system 1B according to the present embodiment is provided inside a vehicle VC, and includes a processing device 100B and microphones 200A, 200B, 200C, and 200D.

The processing device 100B performs machine learning using a learning model on audio data input from a plurality of microphones 200A, 200B, 200C, and 200D, detects impact sound data, and extracts audio data from one microphone. When the impact sound data is detected from the microphone, the area where the impact sound data is generated is estimated based on the audio data of all the microphones.
In the present embodiment, in particular, when the detection unit 130 detects impact sound data from the audio data of one microphone, the processing device 100B detects the deviation of the learning result for the audio data of each of the microphones 200A, 200B, 200C, and 200D. Based on the difference in intensity, the area where the impact sound data is generated is estimated.

<Configuration of processing device 100B>
As shown in FIG. 10, the processing device 100B includes an audio data information input unit 110, a storage unit 120, a detection unit 130, an estimation unit 140B, a communication unit 150, a network detection unit 160, and a data transmission/reception unit. 170 and .
It should be noted that the components denoted by the same reference numerals as in the first embodiment and the second embodiment have the same functions, and detailed description thereof will be omitted.

When the detection unit 130 detects the impact sound data from the audio data of one microphone, the estimation unit 140B estimates the generation area of the impact sound data based on the audio data of all the microphones.
Specifically, when the detection unit 130 detects the impact sound data from the sound data of one microphone, the estimation unit 140B determines the difference in the degree of divergence between the learning results for the sound data of each of the microphones 200A, 200B, 200C, and 200D. , the area where the impact sound data is generated is estimated.
More specifically, with reference to FIG. 11, FIG. 11 shows the degree of divergence of the learning result for the audio data of the microphones 200B and 200D when the detection unit 130 detects impact sound data from the audio data of one microphone. is sufficiently smaller than the degree of divergence of the learning results for the voice data of the microphones 200A and 200C.
Therefore, an example of estimating the generation area of impact sound data based on the degree of divergence of the learning results for the voice data of the microphones 200B and 200D is shown.
Here, the ratio of the deviation of the learning result with respect to the voice data of the microphone 200B and the deviation of the learning result with respect to the voice data of the microphone 200D is 2:3, and the deviation of the learning result with respect to the voice data of the microphone 200B is 2:3. If it is 1.5 times smaller than the deviation of the learning result for the sound data of 200D, the distance to the impact sound data generating area is 1.5 times longer for the microphone 200D than for the microphone 200B.
Therefore, a circle having a radius corresponding to this ratio is drawn from the center of the microphone 200D and the microphone 200B, and the estimating unit 140A designates the vicinity of the location where the two arcs and the body line BD of the vehicle are close to each other as the impact sound data generation area. presume.

<Processing of processing device 100B>
The processing of the processing device 100B according to this embodiment will be described with reference to FIG.

The audio data information input unit 110 of the processing device 100B inputs audio data obtained from the microphones 200A, 200B, 200C, and 200D at predetermined acquisition timings.
Then, the audio data information input unit 110 stores the audio data information attached to the audio data to which the time information from the clock unit (not shown) indicating the input time of the audio data information is input, in a predetermined storage area in the storage unit 120. (step S110).

The detection unit 130 performs machine learning using a learning model on the audio data information stored in the storage unit 120 to detect impact sound data (step S120).
Specifically, the detection unit 130 detects that the sound data is impact sound data when the degree of divergence between the learning result for the sound data and the learning result for the sound data during normal running of the vehicle exceeds a predetermined threshold value. detect something.

When the detection unit 130 detects impact sound data from the audio data of one microphone, the estimation unit 140B estimates the generation area of the impact sound data based on the audio data of all the microphones (step S210). .
Specifically, when the detection unit 130 detects the impact sound data from the sound data of one microphone, the estimation unit 140B determines the difference in the degree of divergence between the learning results for the sound data of each of the microphones 200A, 200B, 200C, and 200D. , the area where the impact sound data is generated is estimated.

The network detection unit 160 detects whether or not the vehicle has entered the communication area of a predetermined network (step S140).
When the network detection unit 160 determines that the vehicle is not within the communication area of the predetermined network ("NO" in step S140), the network detection unit 160 shifts to the standby mode.

On the other hand, when network detection unit 160 determines that the vehicle is within the communication area of the predetermined network (“YES” in step S140), communication unit 150 transmits the time and A message such as an impact sound data generation area is transmitted to the external device (step S150).

Then, the data transmission/reception unit 170 transmits data including impact sound data acquired while the vehicle is running to the server 300, and ends the process (step S160).

<Action/effect>
As described above, the estimating unit 140B of the processing device 100B in the flaw detection system 1 according to the present embodiment, when the detecting unit 130 detects impulsive sound data from voice data of one microphone, Based on the difference in the degree of divergence between the learning results for the audio data 200B, 200C, and 200D, the generation area of the impact sound data is estimated.
In other words, the degree of divergence of the learning results for the audio data of the microphones 200A, 200B, 200C, and 200D differs in proportion to the distance from the scratch occurrence area.
Using this relationship, the area where the impact sound data is generated can be estimated by converting the difference in the degree of divergence of the learning results for the audio data to each of the microphones 200A, 200B, 200C, and 200D into the distance to the area where the damage is generated. can do.
Therefore, regardless of the type of vehicle, even under various conditions, it is possible to quickly identify the occurrence of a scratch or the like on the vehicle and the occurrence area.

<Modification 4>
In this embodiment, in order to facilitate understanding, the estimating unit 140B converts the difference in the degree of divergence between the learning results for the audio data of the microphones 200A, 200B, 200C, and 200D into the distance to the area where the flaw occurs. , a method of estimating a scratch occurrence area from the intersection of circles drawn as radii from the centers of the microphones 200A, 200B, 200C, and 200D for each converted distance and the vicinity of the body line BD of the vehicle.
However, it is also possible to estimate the scratch occurrence area from the vicinity of the body line BD of the vehicle and the intersection of the spheres drawn as radii from the centers of the microphones 200A, 200B, 200C, and 200D for each converted distance. .
By doing so, it is possible to estimate the scratch occurrence area in the three-dimensional space, so that the accuracy of estimating the scratch occurrence area can be improved.

The processing of the processing devices 100, 100A, and 100B is recorded in a recording medium readable by a computer system, and the program recorded in the recording medium is read and executed by the processing devices 100, 100A, and 100B. can be realized. The computer system here includes hardware such as an OS and peripheral devices.

Also, the "computer system" includes a home page providing environment (or display environment) if the WWW (World Wide Web) system is used. Further, the above program may be transmitted from a computer system storing this program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in a transmission medium. Here, the "transmission medium" for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.

Further, the program may be for realizing part of the functions described above. Further, it may be a so-called difference file (difference program) that can realize the above-described functions in combination with a program already recorded in the computer system.

Although the embodiment of the present invention has been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes design within the scope of the gist of the present invention.

1; wound detection system 1A; wound detection system 1B; wound detection system 100; processing device 100A; processing device 100B; processing device 110; communication unit 160; network detection unit 170; data transmission/reception unit 200A; microphone 200B; microphone 200C; microphone 200D;

Claims (10)

A flaw detection system for a vehicle, comprising: a processor provided inside the vehicle; and at least a plurality of microphones,
The processing device is
a storage unit that stores audio data input from the plurality of microphones in a storage area determined for each microphone;
a detection unit that performs machine learning using a learning model on the audio data stored in the storage unit to detect impact sound data;
an estimating unit for estimating a generation area of the impact sound data based on the audio data of all the microphones when the impact sound data is detected from the audio data of one of the microphones in the detection unit;
with
The learning model is based on the driving sound data that can be observed inside the vehicle during normal driving, the noise generated outside the vehicle such as road conditions and rain during normal driving, the voice of the passenger, and the voice emitted by the equipment used. Learning data that uses the impact sound data obtained by superimposing the sound of scratches caused by colliding small pebbles with a plurality of speeds on the vehicle on the running sound that can be observed inside the vehicle, including the noise generated in the room, and the impact sound data as teacher data. A wound detection system characterized by being generated by . 2. The flaw detection system according to claim 1 , wherein the learning model differs for each model of the vehicle. When the degree of divergence between the learning result for the sound data and the learning result for the sound data during normal running of the vehicle exceeds a predetermined threshold, the detection unit detects that the sound data is the impact sound data. 3. The flaw detection system according to any one of claims 1 and 2 , wherein the flaw detection system detects that The estimating unit estimates a generation area of the impact sound data from a difference in volume of the audio data of each microphone when the detection unit detects the impact sound data from the audio data of the one microphone. 4. The flaw detection system according to any one of claims 1 to 3 , wherein: When the detection unit detects the impact sound data from the audio data of the one microphone, the estimating unit determines the generation of the impact sound data from the difference in arrival time of the audio data to each microphone. 4. The flaw detection system according to any one of claims 1 to 3 , wherein an area is estimated. When the detecting unit detects the impact sound data from the audio data of the one microphone, the estimating unit calculates the impact sound data from the difference in the degree of divergence of the learning result with respect to the audio data of each microphone. 4. The flaw detection system according to claim 3 , wherein the area of occurrence of . a communication unit in which the processing device transmits information to an external device;
a network detection unit that detects a predetermined network;
with
When the network detection unit detects the predetermined network, the communication unit transmits a message to the external device indicating which area of the vehicle may have been damaged. The wound detection system according to any one of claims 1 to 6 . including a server,
The wound detection system according to any one of claims 1 to 7 , wherein data including the impact sound data acquired while the vehicle is running is transmitted to the server. A flaw detection method in a flaw detection system comprising a processor provided inside a vehicle and at least three or more microphones,
a first step in which the processing device stores the audio data input from the at least three or more microphones in a storage area determined for each microphone;
a second step in which the processing device performs machine learning using a learning model on the audio data stored in the storage unit to detect impact sound data;
a third step of estimating a generation area of the impact sound data based on the audio data of all the microphones when the processing device detects the impact sound data from the audio data of one of the microphones;
with
The learning model is based on sound data that can be observed inside the vehicle during normal driving, noise generated outside the vehicle such as road conditions and rain during normal driving, voices of passengers and sounds emitted by equipment used, etc. Learning data using the impact sound data obtained by superimposing the sound of scratches caused by colliding small pebbles with a plurality of speeds on the vehicle on the running sound that can be observed inside the vehicle, including the noise generated in the vehicle, as teacher data. A flaw detection method characterized by : A program for causing a computer to execute a flaw detection method in a flaw detection system comprising a processor provided inside a vehicle and at least three or more microphones,
a first step in which the processing device stores the audio data input from the at least three or more microphones in a storage area determined for each microphone;
a second step in which the processing device performs machine learning using a learning model on the audio data stored in the storage unit to detect impact sound data;
a third step of estimating a generation area of the impact sound data based on the audio data of all the microphones when the processing device detects the impact sound data from the audio data of one of the microphones;
with
The learning model is based on sound data that can be observed inside the vehicle during normal driving, noise generated outside the vehicle such as road conditions and rain during normal driving, voices of passengers and sounds emitted by equipment used, etc. Learning data using the impact sound data obtained by superimposing the sound of scratches caused by colliding small pebbles with a plurality of speeds on the vehicle on the running sound that can be observed inside the vehicle, including the noise generated in the vehicle, as teacher data. A program for causing a computer to execute a wound detection method characterized by being generated .

Images