WO2023175920A1 - Fraud estimation device, fraud estimation system, learning device, fraud estimation method, learning method, and recording medium - Google Patents

Abstract

Provided are a fraud estimation device and the like capable of reducing manual work.　A fraud estimation device 10 according to one aspect of the present disclosure comprises: a fraud estimation unit 130 that estimates the degree of fraud possibility of a participant in a subject race from estimation information including a race state information extracted from measurement data on the subject race, using an estimation model generated to estimate the degree of fraud possibility of a participant in a race from training information, including race state information about the race, by learning using race state information indicating the state of a past race extracted from measurement data on the past race; and an output unit 140 that outputs the degree of fraud possibility of the participant.

Description

Fraud estimation device, fraud estimation system, learning device, fraud estimation method, learning method, and storage medium

The present disclosure relates to a technique for estimating fraud.

In publicly managed competitions, fraud such as match-fixing may occur. Situations that appear to be fraudulent are determined manually based on data such as odds and winning percentage.

Patent Document 1 describes that, in a system that evaluates buy prediction data, fraud such as match-fixing can be pointed out from the statistical significance of buy predictions and specific competition results.

Japanese Patent Application Publication No. 2011-043916

With the technology of Patent Document 1, if the race result is different from what was expected in advance, there is a possibility that fraud will be pointed out, regardless of whether or not there is actual fraud. Furthermore, the technique disclosed in Patent Document 1 cannot be said to eliminate manual determination.

One of the purposes of the present disclosure is to provide a fraud estimation device that can reduce manual work.

A fraud estimation device according to an aspect of the present disclosure acquires learning information including race state information of a race by learning using race state information representing the state of the past race extracted from measurement data of the past race. Using an estimation model that is generated to estimate the degree of possibility of fraud by a participant in the target race, the number of participants in the target race is determined based on estimation information including race status information extracted from measurement data in the target race. and an output means for outputting the degree of possibility of cheating of the contestant.

A fraud estimation method according to an aspect of the present disclosure includes learning information including race state information of a race by learning using race state information representing the state of the past race extracted from measurement data of the past race. Using an estimation model that is generated to estimate the degree of possibility of fraud by a participant in the target race, the number of participants in the target race is determined based on estimation information including race status information extracted from measurement data in the target race. The degree of possibility of cheating of the contestant is estimated, and the degree of possibility of cheating of the contestant is output.

A storage medium according to an aspect of the present disclosure is configured to perform learning based on race state information of a race by learning using learning information including race state information representing a state of the past race extracted from measurement data of the past race. Using an estimation model that is generated to estimate the degree of possibility of cheating among the contestants in the target race, the number of contestants in the target race is calculated from estimation information including race status information extracted from measurement data in the target race. A program is stored that causes a computer to execute an estimation process for estimating the degree of possibility of cheating and an output process for outputting the degree of possibility of cheating by the contestant.

A learning device according to an aspect of the present disclosure includes an extraction unit that extracts race state information representing a state of the past race from measurement data of the past race, and a learning device that performs learning using learning information including the race state information. , a model generating means for generating an estimation model so as to estimate the degree of possibility of fraud by a participant in the race from estimation information including race state information of the race.

A learning method according to an aspect of the present disclosure extracts race state information representing the state of the past race from measurement data of the past race, and learns the race state by learning using learning information including the race state information. An estimation model is generated so as to estimate the degree of possibility of fraud by a contestant in the race from estimation information including race state information.

A storage medium according to an aspect of the present disclosure performs an extraction process of extracting race state information representing the state of the past race from measurement data of the past race, and learning using learning information including the race state information. , a model generation process for generating an estimation model to estimate the degree of possibility of fraud by a participant in the race from estimation information including race status information of the race.

One aspect of the present disclosure is also realized by a program stored in the storage medium described above.

The present disclosure has the effect of reducing manual work.

FIG. 1 is a block diagram illustrating an example of the configuration of a fraud estimation device according to a first embodiment of the present disclosure. FIG. 2 is a flowchart illustrating an example of the operation of the fraud estimation device according to the first embodiment of the present disclosure. FIG. 3 is a block diagram illustrating an example of the configuration of a learning device according to the second embodiment of the present disclosure. FIG. 4 is a flowchart illustrating an example of the operation of the learning device according to the third embodiment of the present disclosure. FIG. 5 is a block diagram illustrating an example of a configuration of a fraud estimation system according to a third embodiment of the present disclosure. FIG. 6 is a block diagram illustrating an example of a configuration of a fraud estimation device according to a third embodiment of the present disclosure. FIG. 7 is a block diagram illustrating an example of the configuration of a learning device according to a third embodiment of the present disclosure. FIG. 8 is a flowchart illustrating an example of the operation of the learning device according to the third embodiment of the present disclosure. FIG. 9 is a flowchart illustrating an example of the operation of the fraud estimation device according to the third embodiment of the present disclosure. FIG. 10 is a diagram illustrating an example of the hardware configuration of a computer that can implement the fraud estimation device and learning device according to the embodiment of the present disclosure.

Below, embodiments of the present disclosure will be described.

<First embodiment>
First, a first embodiment of the present disclosure will be described in detail using the drawings.

<Configuration>
FIG. 1 is a block diagram illustrating an example of the configuration of a fraud estimation device according to a first embodiment of the present disclosure. In the example shown in FIG. 1, the fraud estimation device 10 of this embodiment includes an estimation section 130 and an output section 140. The estimation unit 130 uses the estimation model to estimate the degree of possibility of cheating of the contestants in the target race from estimation information including race state information extracted from measurement data in the target race. The estimation model uses learning information that includes race state information representing the state of the past race extracted from measurement data in the past race to determine the fraud of the contestant of the race from the race state information of the race. generated to estimate the degree of likelihood. The output unit 140 outputs the degree of possibility of cheating of the contestant.

The race is, for example, a rowing competition. The race may be, for example, horse racing, bicycle racing, or other public competitions such as auto racing. The target race is a race in which the degree of possibility of cheating among contestants is estimated. The past race is one or more races of the same type as the target race that were held in the past (for example, multiple races of the same type as the target race that were held in the past).

The measurement data is, for example, an image such as a moving image captured by at least one of an imaging device fixed to the stadium and an imaging device mounted on a drone. The race status information may be, for example, changes in combinations of positions and speeds of contestants during the race. The measurement data may be a signal representing the state of an operating object (for example, an accelerator, a brake, etc.) that is measured by a sensor attached to an operating object (for example, an accelerator, a brake, etc.) operated by a race participant. The race state information may be, for example, the transition of operations by the race participants estimated from the state of the object to be investigated. Other examples of measurement data and race status information will be described in detail later.

The estimation model is, for example, an estimator configured to receive estimation information as input, estimate the degree of cheating of the contestant from the received information for estimation, and output the estimated degree of cheating of the contestant. It is. The estimator may be implemented, for example, as a processor that executes a program that implements the functions of the estimator. The estimator may be implemented, for example, as a dedicated circuit that implements the functions of the estimator. The estimator may be implemented, for example, as a combination of a processor that executes a program and dedicated circuitry that implements the functionality of the estimator.

The estimated model is generated by the learning described above. As a method for learning the estimation model, various existing learning methods including, for example, heterogeneous mixture learning can be used.

The degree of possibility of a contestant's cheating may be expressed by either a value indicating cheating or a value indicating not cheating. The degree of possibility of a contestant's cheating may be expressed by a value that is greater than or equal to a lower limit value and less than or equal to an upper limit value. The degree of possibility of a contestant's fraud may be expressed by any one of three or more values including an upper limit value and a lower limit value.

<Operation>
FIG. 2 is a flowchart illustrating an example of the operation of the fraud estimation device 10 according to the first embodiment of the present disclosure. In the example shown in FIG. 2, the estimation unit 130 uses an estimation model to estimate the degree of possibility of cheating by a contestant in the target race from estimation information including race state information extracted from estimated data in the target race. Estimate (step S11). Then, the output unit 140 outputs the degree of cheating of the contestant estimated by the estimation unit 130 (step S12).

<Effect>
This embodiment has the effect of reducing manual work. This is because the estimation unit 130 uses the estimation model to estimate the degree of possibility of cheating of the contestants in the target race from the estimation information.

In order to identify fraud, for example, check information on races where the degree of possibility of fraud is higher than a predetermined standard, and omit checking information on races where the degree of possibility of fraud is lower than a predetermined standard. I can do it. As a result, manual work can be reduced.

<Learning information and estimation information>
The learning information includes the same type of information as the estimation information. The learning information may further include information for identifying contestants who cheated in past races, for each past race. Contestant fraud in past races may be manually determined. The learning information may include information representing a portion indicating fraud in race state information and pre-race state information of past races.

The learning information and the estimation information each include race status information extracted from measurement data in the race (hereinafter also referred to as race measurement data), as well as measurement data of participants before the race (pre-race measurement data). may include pre-race condition information extracted from. Race status information may be a progression of combinations of one or more status values, each representing a contestant's status measured during a race. Then, the race state information is generated such that one or more state values included in one combination represent the states of the contestants at the same timing. The pre-race condition information may be a progression of combinations of one or more condition values, each representing a condition of the contestant measured before the race. The pre-race state information is generated such that one or more state values included in one combination represent the states of the contestants at the same timing.

The learning information may include race information. The race information of the learning information includes, for example, a combination of odds information and race result information. The estimation information may include race information of the estimation information. The race information of the estimation information includes odds information. The race information of the learning information and the race information of the estimation information may each include changes in the bias of the number of votes.

The learning information and estimation information may include state values representing attributes of the contestant. The status value representing the attributes of the contestant may include, for example, information on the branch to which the contestant belongs, and information indicating whether the stadium where the race is held is the contestant's home stadium. The state value representing the contestant's attributes may include information representing the relationship between the contestant and other contestants. Information representing the relationship between a contestant and other contestants includes, for example, information representing a teacher-pupil relationship, information representing siblings, information representing friendships, information representing whether or not they are in the same training school. . This information is used, for example, to extract patterns that are likely to occur in the state value transitions of race state information from learning information when there is a specific relationship between contestants and fraud is being committed. It's fine. Then, when the extracted pattern exists in the estimation information, it may be used to estimate the degree of possibility of fraud according to the strength of the existing pattern.

<Race status information>
As mentioned above, race status information is, for example, the evolution of a combination of one or more status values each representing a contestant's status measured during a race. The lace measurement data is an image (for example, a plurality of images or a moving image) obtained by imaging the lace with one or more imaging devices as described above. The race status information is a change in the combination of position and speed for each participant, extracted from images obtained by imaging the race. In this case, position and velocity are each the above-mentioned state values. If the competition is a boat race, the position of the contestant may be the position of the boat on which the contestant rides. The position of the boat is the position of an appropriately defined point on the boat. If the competition is another competition, the position of the contestant may be the position of a point appropriately set on the device on which the contestant is riding. The equipment that contestants ride in competitions is referred to as competition equipment. Examples of competition equipment include boats, motorcycles, bicycles, and the like. The position of the contestant may be acquired, for example, by a position acquisition device that uses a distance acquisition system such as a GPS (Global Positioning System) installed in the competition device on which the contestant is riding. The measurement data in this case may be, for example, an image and data representing the position output from the position acquisition device. The speed of the competition device may be the speed measured by a speedometer of the competition device.

The measurement data may include, for example, the acceleration of the competition device (that is, the acceleration of the contestant), which is measured by an acceleration sensor installed in the competition device, as the above-mentioned state value. In this case, the race state information may include changes in combinations of position and acceleration. The race state information may include changes in combinations of position, velocity, and acceleration. In this case, the state values are position, velocity, and acceleration.

The race status information may include, as the above-mentioned status value, posture information representing the posture of the contestant, which is recognized using existing image recognition technology from measurement data that is an image of the race. The value of the posture information may be defined as appropriate depending on the posture. The race status information includes, as the above-mentioned status value, the prime mover output of the competition device that the contestant rides during the race, as measured by a sensor that measures the output of a prime mover such as a motor (hereinafter referred to as prime mover output). You can stay there.

The measurement data may be a signal representing the state of the operating object output by a sensor attached to the operating object (for example, an accelerator) operated by the contestant and measuring the state of the operating object. The race state information includes a state value representing information on the operating object (for example, a value representing the opening degree of the accelerator, a value representing the depth of the accelerator, etc.) generated from a signal representing the state of the operating object. ) may be included. In this case, the race state information may include, for example, a change in a combination of a position and a value representing the depth of the accelerator.

The race state information may include, as a state value, at least one of the relative positions and relative velocities of the two contestants for at least one of the combinations of the two contestants selected from the contestants. The race state information may include, as a state value, at least one of the relative positions and relative velocities of the two contestants for each combination of the two contestants selected from the contestants.

The combination of status values included in the race status information is not limited to the above examples. The race state information may be a transition of a combination including at least one of a plurality of state values including the state values raised above. Further, the state value is not limited to the above example.

Note that measurement data obtained by measurement by a measuring device such as a sensor mounted on the competition device may be stored in a storage device included in the competition device. Measurement data obtained by measurement by a measuring device such as a sensor mounted on the competition device may be collected by a communication device mounted on the competition device, for example, via wireless communication.

<Pre-race condition information>
As mentioned above, pre-race condition information is, for example, the evolution of a combination of one or more condition values, each representing a condition of a contestant measured before a race.

The pre-race condition information may include biological information of the contestant measured before the race (for example, heart rate measured by a heart rate monitor, blood pressure measured by a sphygmomanometer, etc.). The pre-race condition information may include information extracted from the contestant's biometric information (eg, the contestant's eye movements extracted from the contestant's facial image) measured before the race.

The pre-race status information may include status values extracted from images taken of events involving contestants (for example, a boat race start display and a horse racing paddock) that are held before the race. .

The pre-race state information may include state values representing the behavior of the contestant before the race. State values representing the behavior of the contestant before the race are extracted using existing techniques for e.g. behavior estimation, e.g. from an image of the contestant taken at a location where the contestant can perform the behavior before the race. It may be a value representing the behavior of the contestant. The value representing the contestant's behavior may be a predetermined value for a predefined behavior type. The state value representing the behavior of the contestant before the race is determined using existing technology that estimates the suspiciousness of the behavior, for example, from an image of the contestant taken at a place where the contestant can perform the behavior before the race. It may also be a value representing the degree of suspiciousness of the contestant's behavior, which is extracted as follows. The state value representing the behavior of the contestant before the race can be obtained, for example, from an image of the contestant taken at a location where the contestant can perform the behavior before the race, using existing techniques for estimating emotions from behavior, for example. It may also be a value extracted by the contestant that represents the emotional state of the contestant. The value representing the emotional state may be, for example, any value predetermined for each of a plurality of predetermined emotions.

<Estimated model>
The estimation model includes a function that, when a predetermined pattern is detected in the race state information, estimates the degree of possibility of cheating of the contestant related to the detected pattern according to the strength of the detected pattern. It may be generated as follows. This predetermined pattern may include, for example, deceleration on a straight course, drastic changes in rankings, occurrence of a roundabout course due to excessive acceleration at a corner, unnatural course taking (for example, a contestant in front bulges out at a corner) (in some cases, take a course that does not attack the inside).

<Second embodiment>
Next, a second embodiment of the present disclosure will be described in detail using the drawings.

<Configuration>
FIG. 3 is a block diagram illustrating an example of the configuration of a learning device according to the second embodiment of the present disclosure. In the example shown in FIG. 3, the learning device 20 includes an extraction section 220 and a model generation section 230. The extraction unit 220 extracts race state information representing the state of the past race from measurement data of the past race. The model generation unit 230 estimates the degree of possibility of cheating of the contestants in the race based on the estimation information including the race state information of the race through learning using the learning information including the race state information. Generate the model. Specifically, the extraction unit 220 may extract the above-mentioned state values from measurement data in past races, and use the extracted state values to generate race state information.

<Operation>
FIG. 4 is a flowchart illustrating an example of the operation of the learning device 20 according to the third embodiment of the present disclosure. In the example shown in FIG. 4, the extraction unit 220 extracts learning information including race state information from measurement data in past races (step S21). Next, the model generation unit 230 generates an estimation model so that, through learning using the learning information, the estimation information including the race status information of the race estimates the degree of possibility of cheating by the contestants in the race. is generated (step S22).

<Effect>
This embodiment described above has the same effects as the first embodiment. The reason for this is that the model generation unit 230 estimates the degree of possibility of cheating of a contestant in a race from the estimation information including the race state information of the race through learning using the learning information including the race state information. This is because the estimation model is generated as follows.

<Third embodiment>
Next, a third embodiment of the present disclosure will be described in detail using the drawings.

<Configuration>
FIG. 5 is a block diagram illustrating an example of a configuration of a fraud estimation system according to a third embodiment of the present disclosure. In the example shown in FIG. 5, the fraud estimation system 1 includes a fraud estimation device 100, a learning device 200, a measuring device 300, a data storage device 400, and an output destination device 500. The fraud estimation device 100 is communicably connected to each of the learning device 200, the measuring device 300, and the output destination device 500. The learning device 200 is communicably connected to each of the fraud estimation device 100 and the data storage device 400. The data storage device 400 is communicably connected to each of the measurement device 300 and the learning device 200.

The race status information, pre-race status information, race information, and participant attributes of the past race in this embodiment are the race status information, pre-race status information, race information, and participant attributes of the past race, as described above. is the same as In addition, the race status information, pre-race status information, race information, and participant attributes of the target race are also the same as the race status information, pre-race status information, race information, and participant attributes of the target race described above. be. The race information of the estimation information of the target race and the attributes of the contestants of the target race are given to the fraud estimation device 100 in advance.

<Measuring device 300>
The measuring device 300 is a device that measures the race and the participants before the race. The measuring device 300 is, for example, the above-mentioned imaging device that images the race, the imaging device that images the contestants before the race, the sensor installed in the competition device, and the device that receives the measurement data from the sensor. may contain. The sensor installed in the competition device may be one of the above-mentioned sensors, such as a position acquisition device that measures the position of a contestant, an acceleration sensor, a speedometer, a sensor that measures an object to be operated such as an accelerator, etc. It's good.

The measurement device 300 transmits the measurement data obtained by the measurement device 300 (i.e., the measurement data of the race and the measurement data of the contestants before the race) to the data storage device 400. Furthermore, the measurement device 300 transmits the measurement data obtained by the measurement device 300 (that is, the measurement data of the race and the measurement data of the contestants before the race) to the fraud estimation device 100.

<Data storage device 400>
The data storage device 400 receives the measurement data of past races (ie, the measurement data of the race and the measurement data of the contestants before the race) obtained by the measurement device 300, and stores the received measurement data. The data storage device 400 further stores race information of past races and attributes of contestants of past races. Race information on past races and attributes of contestants in past races are input into the data storage device 400, for example, by a user of the fraud estimation system 1.

<Output destination device 500>
The output destination device 500 may be, for example, either an information processing device or a storage device. The output destination device 500 may be another device.

<Fraud estimation device 100>
FIG. 6 is a block diagram illustrating an example of the configuration of a fraud estimation device 100 according to the third embodiment of the present disclosure. In the example shown in FIG. 6, the fraud estimation device 100 includes a target data receiving section 110, an extracting section 120, an estimating section 130, an output section 140, a model receiving section 150, and a model storage section 160.

<Model reception department 150>
The model reception unit 150 receives information on the above-mentioned estimated model from the learning device 200. The estimation model of this embodiment is the same as the estimation model of the first embodiment described above. In the description of this embodiment, the information on the estimated model is information necessary for operating the processor as the estimated model, including, for example, a program and parameters for realizing the estimated model. The model reception unit 150 stores information on the received estimated model in the model storage unit 160.

<Model storage unit 160>
The model storage unit 160 stores information on estimated models. The information on the estimation model is read by the estimation unit 130, for example, before the fraud estimation device 100 estimates the degree of possibility of fraud by a contestant in the target race.

<Target data receiving unit 110>
The target data receiving unit 110 receives the measurement data of the target race (ie, the measurement data of the race and the measurement data of the contestants before the race) from the measurement device 300 . A subject race is a race in which the contestant is subject to an estimation of the degree of likelihood of cheating. The target data receiving unit 110 sends the received measurement data of the target race to the extracting unit 120.

<Extraction unit 120>
The extraction unit 120 receives the measurement data of the target race (that is, the measurement data of the target race and the measurement data of the contestants before the target race) from the target data receiving unit 110. The extraction unit 120 extracts the above-mentioned state value from the received measurement data of the target race using an existing method for extracting the above-mentioned state value. The extraction unit 120 generates race state information using the state values extracted from the measurement data of the target race. The extraction unit 120 extracts the above-mentioned state value from the received measurement data of the previous participant in the target race, using an existing method for extracting the above-mentioned state value. The extraction unit 120 generates pre-race state information using the state values extracted from the measurement data of the contestants before the target race. The extraction unit 120 sends the race state information and pre-race state information to the estimation unit 130.

<Estimation unit 130>
The estimation unit 130 receives the race state information and the pre-race state information from the extraction unit 120. Further, the estimation unit 130 is provided with race information for estimation information of the target race and attributes of contestants of the target race, for example, by the user of the fraud estimation device 100. Furthermore, when the fraud estimation device 100 starts operating, the estimation unit 130 reads information on the estimated model from the model storage unit 160.

Similarly to the estimation unit 130 in the first embodiment, the estimation unit 130 estimates the degree of possibility of cheating among the contestants in the target race. Specifically, the estimation unit 130 uses the estimation model to estimate the degree of possibility of cheating of the contestants in the target race from estimation information including race state information and pre-race state information of the target race. do. The degree of possibility of cheating by a contestant in a target race represents the degree of possibility that a contestant in a target race has committed a fraudulent act during the target race.

<Output section 140>
The output unit 140 receives from the estimation unit 130 the degree of possibility of cheating of the contestants in the target race. The output unit 140 outputs the received degree of possibility of cheating of the contestant in the target race.

<Learning device 200>
FIG. 7 is a block diagram illustrating an example of the configuration of a learning device 200 according to a third embodiment of the present disclosure. In the example shown in FIG. 7, the learning device 200 includes a data acquisition section 210, an extraction section 220, a model generation section 230, and a model output section 240.

<Data acquisition unit 210>
The data acquisition unit 210 acquires measurement data of past races (specifically, measurement data of past races and measurement data of previous contestants in past races), race information, and participation data from the data storage device 400. Get the attributes of the person. That is, the data acquisition unit 210 reads measurement data of past races, race information, and attributes of contestants from the data storage device 400. The data acquisition unit 210 sends the read measurement data of past races, race information, and attributes of contestants to the extraction unit 220.

<Extraction unit 220>
The extraction unit 220 extracts measurement data of past races (specifically, measurement data of past races and measurement data of previous contestants in past races), race information, and contestants from the data acquisition unit 210. Receive the attributes of. The extraction unit 220 extracts the above-mentioned state value from the measurement data of the past race, and uses the extracted state value to generate race state information and pre-race state information of the past race. Specifically, the extraction unit 220 extracts state values from measurement data of past races, and generates race state information using the state values extracted from measurement data of past races. The extraction unit 220 extracts state values from measurement data of previous contestants in past races, and uses the state values extracted from measurement data of previous contestants in past races to obtain pre-race state information. generate. The method by which the extraction unit 220 extracts the state value is the same as the method by which the extraction unit 120 of the fraud estimation device 100 extracts the state value.

The extraction unit 220 sends the generated race status information and pre-race status information of the past race, and the received race information and participant attributes of the past race to the model generation unit 230.

<Model generation unit 230>
The model generation unit 230 receives race status information, pre-race status information, race information, and participant attributes of past races from the extraction unit 220. The model generation unit 230 generates an estimated model by learning using the received race state information, pre-race state information, race information, and attributes of contestants (ie, learning information) of past races.

The model generation unit 230 sends information about the generated estimated model to the model output unit 240.

<Model output unit 240>
The model output unit 240 receives information about the estimated model from the model generation unit 230. The model output unit 240 transmits information on the estimated model to the fraud estimation device 100.

<Operation>
Next, operations of the fraud estimation device 100 and the learning device 200 according to the third embodiment of the present disclosure will be described in detail using the drawings.

FIG. 8 is a flowchart illustrating an example of the operation of the learning device 200 according to the third embodiment of the present disclosure. In the example shown in FIG. 8, the data acquisition unit 210 of the learning device 200 acquires measurement data of past races (step S101). The data acquisition unit 210 acquires measurement data of previous contestants in past races (step S102). The data acquisition unit 210 may perform the operation in step S101 and the operation in step S102 at the same time. The data acquisition unit 210 may perform the operation in step S101 and the operation in step S102 in parallel. The data acquisition unit 210 may perform the operation in step S101 and the operation in step S102 in the reverse order. The data acquisition unit 210 may perform a combination of the operation in step S101 and the operation in step S102 for each past race. The data acquisition unit 210 may acquire race information of past races and attributes of contestants in the operation of step S101 or step S102.

Next, the extraction unit 220 extracts race status information from the measurement data of past races (step S103). Specifically, as described above, the extraction unit 220 extracts state values from measurement data of past races, and generates race state information using the state values extracted from measurement data of past races. do.

The extraction unit 220 further extracts pre-race state information from the measurement data of previous participants in past races (step S104). Specifically, the extraction unit 220 extracts the state value from the measurement data of the previous contestant in the past race, and uses the state value extracted from the measurement data of the previous contestant in the past race. Generate pre-race state information. The data acquisition unit 210 may perform the operation in step S103 and the operation in step S104 simultaneously. The data acquisition unit 210 may perform the operation in step S103 and the operation in step S104 in parallel. The data acquisition unit 210 may perform the operation in step S103 and the operation in step S104 in the reverse order. The data acquisition unit 210 may perform a combination of the operation in step S103 and the operation in step S104 for each past race.

The model generation unit 230 generates an estimated model by learning using race state information and pre-race information (step S105). The model generation unit 230 may generate the estimated model by learning using race information and attributes of contestants in addition to race state information and pre-race information.

The model output unit 240 outputs the estimated model (step S106). Specifically, the model output unit 240 outputs information about the generated estimation model to the fraud estimation device 100.

FIG. 9 is a flowchart illustrating an example of the operation of the fraud estimation device 100 according to the third embodiment of the present disclosure. Before starting the operation in FIG. 9, the model reception unit 150 receives information on the estimated model from the learning device 200, and stores the received information on the estimated model in the model storage unit 160. The estimating unit 130 then reads information on the estimated model from the model storage unit 160.

In the example shown in FIG. 9, the target data receiving unit 110 receives measurement data of previous participants in the target race (step S111). The target data receiving unit 110 further receives measurement data of the target race (step S112).

The extraction unit 120 extracts pre-race state information of the target race from the measurement data of the participants before the target race (step S113). Specifically, the extraction unit 120 extracts a state value from the measured data of the contestant before the target race, uses the state value extracted from the measured data of the contestant before the target race, and extracts the state value from the measured data of the contestant before the target race. Generate pre-race state information for.

The extraction unit 120 further extracts race status information of the target race from the measurement data of the target race (step S114). Specifically, the extraction unit 120 extracts a state value from the measurement data of the target race, and generates race state information of the target race using the state value extracted from the measurement data of the target race.

Next, the estimation unit 130 uses the estimation model to estimate the degree of possibility of cheating by the contestants of the target race from the status information of the target race and the pre-race status information (step S115). In step S115, the estimating unit 130 may estimate the degree of possibility of fraud by the contestants of the target race from the state information of the target race, the pre-race state information, the race information, and the attributes of the contestants.

Next, the output unit 140 outputs the estimated degree of possibility of fraud by the contestant in the target race (step S116).

<Effect>
This embodiment described above has the same effects as the first embodiment. The reason for this is the same as the reason for the effect of the first embodiment. Furthermore, the effects of the second embodiment are also produced due to the reasons why the effects of the second embodiment are produced.
<First modification of the second embodiment>
This modification is the same as the second embodiment except for the differences described below.

In this modification, the model generation unit 230 uses a learning method such as heterogeneous mixture learning that can generate a model that can further estimate the factors that contribute to the estimation result and the magnitude of the contribution of the factors. and generate an estimation model.

Using such an estimation model, the estimation unit 130 further estimates the factors that contribute to the estimation result (i.e., the degree of possibility of cheating of the contestants in the target race) and the magnitude of the contribution caused by the factors. do.

The output unit 140 further outputs, in addition to the degree of possibility of cheating by the contestants in the target race, factors that contribute to the degree of possibility of cheating by the contestants in the target race, and the magnitude of the contribution by the factors. .

<Second modification of second embodiment>
This modification is the same as the second embodiment except for the differences described below.

The data storage device 400 may further accumulate race state information and pre-race state information of past races extracted from the measurement data obtained by the measurement device 300.

The data acquisition unit 210 of the learning device 200 reads measurement data, race information, and attributes of contestants for races in which race status information and pre-race status information are not stored among past races. Among past races, for races in which race status information and pre-race status information have been accumulated, the data acquisition unit 210 acquires the accumulated race status information, pre-race status information, race information, and attributes of participants. read out.

The extraction unit 220 generates race state information and pre-race state information from the read measurement data. The extraction unit 220 stores the generated race state information and pre-race state information in the data storage device 400.

<Other modifications of the second embodiment>
Fraud estimation device 100 and learning device 200 may be implemented as the same device.

Furthermore, the target data receiving unit 110 of the fraud estimation device 100 may receive measurement data of previous contestants in the target race before the target race is held. The extraction unit 120 may extract the pre-race state information from the measurement data of the participants before the target race, before the target race is held. The estimating unit 130 uses an estimation model that estimates the degree of possibility of a contestant committing fraud based on pre-race state information before the target race is held. The degree of likelihood of a contestant in a race committing fraud may be estimated. The output unit 140 may output the degree of possibility that a contestant in the target race will commit fraud before the target race is held. The output destination device 500 in this case may be an information processing device or a display device that can be viewed by the organizer or related parties of the target race. The output destination device 500 in this case may be an information processing device or a display device that displays received information using a screen to a person within a range where the screen of the output destination device 500 can be viewed. The output destination device 500 in this case may be an information processing device that distributes information about the target race. The information processing device may distribute information representing the degree of possibility that a contestant in the target race will commit fraud to a terminal of a registrant who is registered as a destination of information distribution.

<Other embodiments>
The fraud estimation device and learning device according to the above-described embodiments of the present disclosure can be realized by a computer including a memory loaded with a program read from a storage medium and a processor that executes the program. The fraud estimation device and learning device according to the embodiments described above can also be realized by dedicated hardware. The fraud estimation device and learning device according to the embodiments described above can also be realized by a combination of the computer described above and dedicated hardware.

FIG. 10 is a diagram illustrating an example of the hardware configuration of a computer 1000 that can implement the fraud estimation device and learning device according to the above-described embodiment. In the example shown in FIG. 10, the computer 1000 includes a processor 1001, a memory 1002, a storage device 1003, and an I/O (Input/Output) interface 1004. Additionally, the computer 1000 can access a storage medium 1005. The memory 1002 and the storage device 1003 are, for example, a RAM (Random Access Memory), a hard disk, or the like. The storage medium 1005 is, for example, a storage device such as a RAM or a hard disk, a ROM (Read Only Memory), or a portable storage medium. The storage device 1003 may be the storage medium 1005. The processor 1001 can read and write data and programs to and from the memory 1002 and the storage device 1003. Processor 1001 can access other devices via I/O interface 1004. Processor 1001 can access storage medium 1005. The storage medium 1005 stores a program that causes the computer 1000 to operate as the fraud estimation device according to the embodiment described above or a program that causes the computer 1000 to operate as the learning device according to the embodiment described above.

The processor 1001 loads into the memory 1002 a program stored in the storage medium 1005 that causes the computer 1000 to operate as the fraud estimation device according to the above-described embodiment. Then, by the processor 1001 executing the program loaded into the memory 1002, the computer 1000 operates as the fraud estimation device according to the embodiment described above.

The processor 1001 loads into the memory 1002 a program stored in the storage medium 1005 that causes the computer 1000 to operate as the learning device according to the above-described embodiment. Then, by the processor 1001 executing the program loaded into the memory 1002, the computer 1000 operates as the learning device according to the above-described embodiment.

The target data receiving unit 110, the extracting unit 120, the estimating unit 130, the output unit 140, and the model accepting unit 150 can be realized by, for example, a processor 1001 that executes a program loaded into the memory 1002. The data acquisition unit 210, the extraction unit 220, the model generation unit 230, and the model output unit 240 can be realized, for example, by the processor 1001 that executes a program loaded into the memory 1002. The model storage unit 160 can be realized by a memory 1002 included in the computer 1000 or a storage device 1003 such as a hard disk device. Part or all of the target data receiving section 110, the extracting section 120, the estimating section 130, the output section 140, the model receiving section 150, and the model storage section 160 can be realized by a dedicated circuit that realizes the functions of each section. Part or all of the data acquisition section 210, extraction section 220, model generation section 230, and model output section 240 can be realized by a dedicated circuit that realizes the functions of each section.

Further, part or all of the above embodiments may be described as in the following supplementary notes, but the present invention is not limited to the following.

(Additional note 1)
Estimating the degree of possibility of cheating by a participant in the race based on the race status information through learning using learning information that includes race status information representing the status of the past race extracted from measurement data in the past race. estimating means for estimating the degree of possibility of cheating by a contestant in the target race from estimation information including race state information extracted from measurement data in the target race using the estimation model generated in the manner; ,
output means for outputting the degree of possibility of cheating of the contestant;
A fraud estimation device comprising:

(Additional note 2)
The fraud estimation device according to supplementary note 1, wherein the race state information includes a change in a combination of position and speed for each participant in the race, which is extracted from an image obtained as the measurement data by imaging the race.

(Additional note 3)
The race status information includes changes in the combination of relative positions and relative velocities between the participants in the race, which are extracted from images obtained as the measurement data by imaging the race. Estimation device.

(Additional note 4)
The fraud estimation device according to any one of Supplementary Notes 1 to 3, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.

(Appendix 5)
The estimation model estimates the degree of fraud for each participant based on pre-race state information extracted from measurement data of participants before the race and the race state information. generated by the learning using the learning information further including pre-race state information,
The estimating means estimates the degree of the possibility of fraud using the estimation information including the pre-race state information of the contestant in the target race before the target race. The fraud estimation device according to any one of the items.

(Appendix 6)
The fraud estimation device according to appendix 5, wherein the pre-race state information includes biometric information extracted from the biometric data of the contestant obtained as the measured data.

(Appendix 7)
The fraud estimation device according to appendix 5 or 6, wherein the pre-race state information includes a transition of the estimated behavior of the contestant extracted from an image taken of the contestant before the race.

(Appendix 8)
The pre-race state information includes the transition of the estimated state of the contestant in the event, which is extracted from an image of the contestant in the event before the race. Fraud estimation device described.

(Appendix 9)
The estimation model is generated by the learning to estimate factors contributing to the degree of possibility and the magnitude of contribution of the factors to the degree of possibility,
The estimation means uses the estimation model to estimate factors contributing to the degree of possibility of cheating of the contestant in the target race and the magnitude of contribution of the factors to the degree of possibility. ,
The fraud estimation device according to any one of Supplementary Notes 1 to 8, wherein the output means further outputs the factor and the magnitude of the contribution of the factor.

(Appendix 10)
a learning device comprising learning means for performing the learning of the estimated model;
A fraud estimation device according to any one of Supplementary Notes 1 to 9;
Fraud estimation system including.

(Appendix 11)
Extracting means for extracting race state information representing the state of the past race from measurement data of the past race;
Model generation that generates an estimation model to estimate the degree of possibility of cheating of a contestant in the race from estimation information including race state information of the race by learning using learning information including the race state information. means and
A learning device equipped with.

(Appendix 12)
The learning device according to appendix 11, wherein the race state information includes a transition in a combination of position and speed for each participant in the race, which is extracted from an image obtained as the measurement data by imaging the race.

(Appendix 13)
The learning according to appendix 11 or 12, wherein the race state information includes changes in the combination of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. Device.

(Appendix 14)
The learning device according to any one of Supplementary Notes 11 to 13, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.

(Appendix 15)
The model generating means is configured to calculate the estimation model based on the estimation information further including pre-race state information representing the pre-race state of the contestant extracted from measurement data of the contestant before the race. The estimation model is generated by the learning using the learning information further including pre-race state information of past races so as to estimate the degree of fraud for each person. learning device.

(Appendix 16)
The learning device according to appendix 15, wherein the pre-race state information includes biological information extracted from biometric data of the contestant obtained as the measured data.

(Appendix 17)
The learning device according to supplementary note 15 or 16, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image of the contestant before the race.

(Appendix 18)
The pre-race state information includes a transition in the estimated state of the contestant in the event, which is extracted from an image of the contestant in the event before the race. The learning device described.

(Appendix 19)
The model generation means generates the estimation model by the learning so that the estimation model estimates a factor contributing to the degree of possibility and a magnitude of contribution of the factor to the degree of possibility. The learning device according to any one of Supplementary Notes 11 to 18.

(Additional note 20)
Estimating the degree of possibility of cheating by a participant in the race based on the race status information through learning using learning information that includes race status information representing the status of the past race extracted from measurement data in the past race. using the estimation model generated to estimate the degree of possibility of cheating of the contestants in the target race from estimation information including race status information extracted from measurement data in the target race;
outputting the degree of likelihood of cheating of the contestant;
Fraud estimation method.

(Additional note 21)
The fraud estimation method according to appendix 20, wherein the race state information includes changes in combinations of position and speed for each participant in the race, which are extracted from images obtained as the measurement data by imaging the race.

(Additional note 22)
The race status information includes changes in the combination of relative positions and relative velocities between contestants in the race, extracted from images obtained as the measurement data by imaging the race. Fraud as set forth in Appendix 20 or 21 Estimation method.

(Additional note 23)
The fraud estimation method according to any one of Supplementary Notes 20 to 22, wherein the race state information includes a change in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.

(Additional note 24)
The estimation model estimates the degree of fraud for each participant based on pre-race state information extracted from measurement data of participants before the race and the race state information. generated by the learning using the learning information further including pre-race state information,
According to any one of Supplementary Notes 20 to 23, the estimation information including the pre-race state information of the contestant in the target race before the target race is used to estimate the degree of possibility of fraud. Described fraud estimation method.

(Additional note 25)
The fraud estimation method according to attachment 24, wherein the pre-race state information includes biometric information extracted from biometric data of the contestant obtained as the measured data.

(Additional note 26)
The fraud estimation method according to attachment 24 or 25, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image taken of the contestant before the race.

(Additional note 27)
The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. Described fraud estimation method.

(Additional note 28)
The estimation model is generated by the learning to estimate factors contributing to the degree of possibility and the magnitude of contribution of the factors to the degree of possibility,
using the estimation model to estimate factors contributing to the degree of possibility of cheating of the contestant in the target race and the magnitude of the contribution of the factors to the degree of possibility;
s The fraud estimation method according to any one of appendices 20 to 27, further comprising outputting the factor and the magnitude of the contribution of the factor.

(Additional note 29)
extracting race state information representing the state of the past race from measurement data in the past race;
Generating an estimation model to estimate the degree of possibility of fraud by a contestant in the race from the estimation information including the race state information of the race by learning using the learning information including the race state information;
How to learn.

(Additional note 30)
The learning method according to appendix 29, wherein the race state information includes a change in a combination of position and speed for each participant in the race, which is extracted from an image obtained as the measurement data by imaging the race.

(Appendix 31)
The learning according to appendix 29 or 30, wherein the race state information includes changes in the combination of relative positions and relative speeds between contestants in the race, extracted from images obtained as the measurement data by imaging the race. Method.

(Appendix 32)
The learning method according to any one of Supplementary Notes 29 to 31, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.

(Appendix 33)
The estimation model determines the degree of fraud for each contestant based on the estimation information further including pre-race state information representing the contestant's pre-race state extracted from measurement data of the contestant before the race. The learning method according to any one of appendices 29 to 32, wherein the estimation model is generated by the learning using the learning information further including pre-race state information of past races so as to estimate the pre-race state information of past races.

(Appendix 34)
The learning method according to attachment 33, wherein the pre-race state information includes biometric information extracted from the biometric data of the contestant obtained as the measured data.

(Appendix 35)
The learning method according to appendix 33 or 34, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image taken of the contestant before the race.

(Appendix 36)
The pre-race state information includes a transition in the estimated state of the contestant in the event, which is extracted from an image of the contestant in the event before the race. The learning method described.

(Additional note 37)
The estimation model is generated by the learning so that the estimation model estimates a factor contributing to the degree of possibility and a magnitude of contribution of the factor to the degree of possibility. The learning method described in any one of the paragraphs.

(Appendix 38)
Estimating the degree of possibility of cheating by a participant in the race based on the race status information through learning using learning information that includes race status information representing the status of the past race extracted from measurement data in the past race. an estimation process of estimating the degree of possibility of cheating by a contestant in the target race from estimation information including race status information extracted from measurement data in the target race using the estimation model generated as described above; ,
output processing that outputs the degree of possibility of cheating of the contestant;
A storage medium that stores a program that causes a computer to execute.

(Appendix 39)
39. The storage medium according to appendix 38, wherein the race state information includes a change in a combination of position and speed for each participant in the race, which is extracted from an image obtained as the measurement data by imaging the race.

(Additional note 40)
The memory according to appendix 38 or 39, wherein the race state information includes a change in a combination of relative positions and relative speeds between contestants in the race, which is extracted from an image obtained as the measurement data by imaging the race. Medium.

(Appendix 41)
The storage medium according to any one of Supplementary Notes 38 to 40, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.

(Additional note 42)
The estimation model estimates the degree of fraud for each participant based on pre-race state information extracted from measurement data of participants before the race and the race state information. generated by the learning using the learning information further including pre-race state information,
The estimation process estimates the degree of the possibility of fraud using the estimation information including the pre-race state information of the contestant in the target race before the target race. The storage medium according to any one of the items.

(Appendix 43)
The storage medium according to attachment 42, wherein the pre-race condition information includes biometric information extracted from the biometric data of the contestant obtained as the measured data.

(Appendix 44)
The storage medium according to attachment 42 or 43, wherein the pre-race state information includes a transition of the estimated behavior of the contestant extracted from an image taken of the contestant before the race.

(Additional note 45)
The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. Storage medium as described.

(Appendix 46)
The estimation model is generated by the learning to estimate factors contributing to the degree of possibility and the magnitude of contribution of the factors to the degree of possibility,
The estimation process uses the estimation model to estimate factors contributing to the degree of possibility of cheating of the contestant in the target race and the magnitude of contribution of the factors to the degree of possibility. ,
46. The storage medium according to any one of appendices 38 to 45, wherein the output process further outputs the factor and the magnitude of the contribution of the factor.

(Additional note 47)
Extraction processing for extracting race state information representing the state of the past race from measurement data of the past race;
Model generation that generates an estimation model to estimate the degree of possibility of cheating of a contestant in the race from estimation information including race state information of the race by learning using learning information including the race state information. process,
A storage medium that allows a computer to execute.

(Additional note 48)
48. The storage medium according to appendix 47, wherein the race state information includes changes in combinations of position and speed for each participant in the race, which are extracted from images obtained as the measurement data by imaging the race.

(Additional note 49)
The memory according to appendix 47 or 48, wherein the race status information includes a transition in a combination of relative positions and relative speeds between contestants in the race, which is extracted from images obtained as the measurement data by imaging the race. Medium.

(Additional note 50)
50. The storage medium according to any one of appendices 47 to 49, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.

(Appendix 51)
In the model generation process, the estimation model is based on the estimation information that further includes pre-race state information representing the pre-race state of the contestant extracted from measurement data of the contestant before the race. The estimation model is generated by the learning using the learning information further including pre-race state information of past races so as to estimate the degree of fraud for each person. storage medium.

(Appendix 52)
The storage medium according to appendix 51, wherein the pre-race condition information includes biological information extracted from biometric data of the contestant obtained as the measurement data.

(Appendix 53)
The storage medium according to appendix 51 or 52, wherein the pre-race state information includes a transition of the estimated behavior of the contestant extracted from an image taken of the contestant before the race.

(Appendix 54)
The pre-race state information includes a transition in the estimated state of the contestant in the event, which is extracted from an image of the contestant in the event before the race. Storage medium as described.

(Appendix 55)
The model generation process includes generating the estimation model by the learning so that the estimation model estimates a factor contributing to the degree of possibility and a magnitude of contribution of the factor to the degree of possibility. The storage medium according to any one of Supplementary Notes 51 to 54.

Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above embodiments. Various changes can be made to the structure and details of the present disclosure that can be understood by those skilled in the art within the scope of the present disclosure.

1 Fraud estimation system 10 Fraud estimation device 20 Learning device 100 Fraud estimation device 110 Target data receiving unit 120 Extracting unit 130 Estimating unit 140 Output unit 150 Model receiving unit 160 Model storage unit 200 Learning device 210 Data acquisition unit 220 Extracting unit 230 Model generation Section 240 Model output section 300 Measuring device 400 Data storage device 500 Output destination device 1000 Computer 1001 Processor 1002 Memory 1003 Storage device 1004 I/O interface 1005 Storage medium

Claims (55)

1. Estimating the degree of possibility of cheating by a participant in the race based on the race status information through learning using learning information that includes race status information representing the status of the past race extracted from measurement data in the past race. estimating means for estimating the degree of possibility of cheating by a contestant in the target race from estimation information including race state information extracted from measurement data in the target race using the estimation model generated in the manner; ,
   output means for outputting the degree of possibility of cheating of the contestant;
   A fraud estimation device comprising:
2. The fraud estimation device according to claim 1, wherein the race state information includes changes in combinations of positions and speeds for each participant in the race, which are extracted from images obtained as the measurement data by imaging the race.
3. 3. The race status information includes changes in combinations of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. Fraud estimation device.
4. The fraud estimation device according to any one of claims 1 to 3, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.
5. The estimation model estimates the degree of fraud for each participant based on pre-race state information extracted from measurement data of participants before the race and the race state information. generated by the learning using the learning information further including pre-race state information,
   The estimation means estimates the degree of possibility of fraud using the estimation information including the pre-race state information of the contestant in the target race before the target race. The fraud estimation device according to any one of the above.
6. The fraud estimation device according to claim 5, wherein the pre-race state information includes biometric information extracted from biometric data of the contestant obtained as the measurement data.
7. The fraud estimation device according to claim 5 or 6, wherein the pre-race state information includes a transition in estimated behavior of the contestant extracted from an image of the contestant before the race.
8. The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. Fraud estimation device described in .
9. The estimation model is generated by the learning to estimate factors contributing to the degree of possibility and the magnitude of contribution of the factors to the degree of possibility,
   The estimation means uses the estimation model to estimate factors contributing to the degree of possibility of cheating of the contestant in the target race and the magnitude of contribution of the factors to the degree of possibility. ,
   The fraud estimation device according to any one of claims 1 to 8, wherein the output means further outputs the factor and the magnitude of the contribution of the factor.
10. a learning device comprising learning means for performing the learning of the estimated model;
    A fraud estimation device according to any one of claims 1 to 9;
    Fraud estimation system including.
11. Extracting means for extracting race state information representing the state of the past race from measurement data of the past race;
    Model generation that generates an estimation model to estimate the degree of possibility of cheating of a contestant in the race from estimation information including race state information of the race by learning using learning information including the race state information. means and
    A learning device equipped with.
12. The learning device according to claim 11, wherein the race state information includes a change in a combination of position and speed for each contestant in the race, which is extracted from an image obtained as the measurement data by imaging the race.
13. 13. The race status information includes changes in combinations of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. learning device.
14. The learning device according to any one of claims 11 to 13, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.
15. The model generating means is configured to calculate the estimation model based on the estimation information further including pre-race state information representing the pre-race state of the contestant extracted from measurement data of the contestant before the race. The estimation model is generated by the learning using the learning information further including pre-race state information of past races so as to estimate the degree of fraud for each person. The learning device described.
16. The learning device according to claim 15, wherein the pre-race state information includes biometric information extracted from biometric data of the contestant obtained as the measurement data.
17. The learning device according to claim 15 or 16, wherein the pre-race state information includes a transition in estimated behavior of the contestant extracted from an image taken of the contestant before the race.
18. The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. The learning device described in .
19. The model generation means generates the estimation model by the learning so that the estimation model estimates a factor contributing to the degree of possibility and a magnitude of contribution of the factor to the degree of possibility. The learning device according to any one of claims 11 to 18.
20. Estimating the degree of possibility of cheating by a participant in the race based on the race status information through learning using learning information that includes race status information representing the status of the past race extracted from measurement data in the past race. using the estimation model generated to estimate the degree of possibility of cheating of the contestants in the target race from estimation information including race status information extracted from measurement data in the target race;
    outputting the degree of likelihood of cheating of the contestant;
    Fraud estimation method.
21. The fraud estimation method according to claim 20, wherein the race state information includes changes in combinations of position and speed for each contestant in the race, which are extracted from images obtained as the measurement data by imaging the race.
22. 22. The race status information includes changes in combinations of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. Fraud estimation method.
23. The fraud estimation method according to any one of claims 20 to 22, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.
24. The estimation model estimates the degree of fraud for each participant based on pre-race state information extracted from measurement data of participants before the race and the race state information. generated by the learning using the learning information further including pre-race state information,
    Any one of claims 20 to 23, wherein the estimation information including the pre-race state information of the contestant in the target race before the target race is used to estimate the degree of possibility of fraud. Fraud estimation method described in .
25. The fraud estimation method according to claim 24, wherein the pre-race state information includes biometric information extracted from biometric data of the contestant obtained as the measured data.
26. The fraud estimation method according to claim 24 or 25, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image taken of the contestant before the race.
27. The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. Fraud estimation method described in .
28. The estimation model is generated by the learning to estimate factors contributing to the degree of possibility and the magnitude of contribution of the factors to the degree of possibility,
    using the estimation model to estimate factors contributing to the degree of possibility of cheating of the contestant in the target race and the magnitude of the contribution of the factors to the degree of possibility;
    s The fraud estimation method according to any one of claims 20 to 27, further comprising outputting the factor and the magnitude of the contribution of the factor.
29. extracting race state information representing the state of the past race from measurement data in the past race;
    Generating an estimation model to estimate the degree of possibility of fraud by a contestant in the race from the estimation information including the race state information of the race by learning using the learning information including the race state information;
    How to learn.
30. The learning method according to claim 29, wherein the race state information includes changes in combinations of position and speed for each contestant in the race, which are extracted from images obtained as the measurement data by imaging the race.
31. 31. The race status information includes changes in combinations of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. How to learn.
32. The learning method according to any one of claims 29 to 31, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.
33. The estimation model determines the degree of fraud for each contestant based on the estimation information further including pre-race state information representing the contestant's pre-race state extracted from measurement data of the contestant before the race. The learning method according to any one of claims 29 to 32, wherein the estimation model is generated by the learning using the learning information further including pre-race state information of past races so as to estimate.
34. The learning method according to claim 33, wherein the pre-race state information includes biometric information extracted from biometric data of the contestant obtained as the measured data.
35. The learning method according to claim 33 or 34, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image taken of the contestant before the race.
36. The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. The learning method described in.
37. The estimation model is generated by the learning so that the estimation model estimates a factor contributing to the degree of possibility and a magnitude of contribution of the factor to the degree of possibility. The learning method described in any one of the following.
38. Estimating the degree of possibility of cheating by a participant in the race based on the race status information through learning using learning information that includes race status information representing the status of the past race extracted from measurement data in the past race. an estimation process of estimating the degree of possibility of cheating by a contestant in the target race from estimation information including race status information extracted from measurement data in the target race using the estimation model generated as described above; ,
    output processing that outputs the degree of possibility of cheating of the contestant;
    A storage medium that stores a program that causes a computer to execute.
39. 39. The storage medium according to claim 38, wherein the race status information includes changes in combinations of position and speed for each participant in the race, which are extracted from images obtained as the measurement data by imaging the race.
40. 40. The race status information includes changes in combinations of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. storage medium.
41. The storage medium according to any one of claims 38 to 40, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.
42. The estimation model estimates the degree of fraud for each participant based on pre-race state information extracted from measurement data of participants before the race and the race state information. generated by the learning using the learning information further including pre-race state information,
    The estimation process estimates the degree of possibility of fraud using the estimation information including the pre-race state information of the contestant in the target race before the target race.Claims 38 to 41 The storage medium according to any one of .
43. The storage medium according to claim 42, wherein the pre-race condition information includes biometric information extracted from biometric data of the contestant obtained as the measurement data.
44. The storage medium according to claim 42 or 43, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image of the contestant before the race.
45. The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. The storage medium described in .
46. The estimation model is generated by the learning to estimate factors contributing to the degree of possibility and the magnitude of contribution of the factors to the degree of possibility,
    The estimation process uses the estimation model to estimate factors contributing to the degree of possibility of cheating of the contestant in the target race and the magnitude of contribution of the factors to the degree of possibility. ,
    The storage medium according to any one of claims 38 to 45, wherein the output process further outputs the factor and the magnitude of the contribution of the factor.
47. Extraction processing for extracting race state information representing the state of the past race from measurement data of the past race;
    Model generation that generates an estimation model to estimate the degree of possibility of cheating of a contestant in the race from estimation information including race state information of the race by learning using learning information including the race state information. process,
    A storage medium that allows a computer to execute.
48. 48. The storage medium according to claim 47, wherein the race state information includes changes in combinations of position and speed for each participant in the race, which are extracted from images obtained as the measurement data by imaging the race.
49. 49. The race status information includes changes in combinations of relative positions and relative velocities between contestants in the race, which are extracted from images obtained as the measurement data by imaging the race. storage medium.
50. The storage medium according to any one of claims 47 to 49, wherein the race state information includes a transition in an estimated operation amount extracted from measurement data of an operation object operated by the contestant.
51. In the model generation process, the estimation model is based on the estimation information that further includes pre-race state information representing the pre-race state of the contestant extracted from measurement data of the contestant before the race. The estimation model is generated by the learning using the learning information further including pre-race state information of past races so as to estimate the degree of fraud for each person. Storage medium described.
52. 52. The storage medium according to claim 51, wherein the pre-race condition information includes biometric information extracted from biometric data of the contestant obtained as the measurement data.
53. The storage medium according to claim 51 or 52, wherein the pre-race state information includes a transition in the estimated behavior of the contestant extracted from an image taken of the contestant before the race.
54. The pre-race state information includes a transition in the estimated state of the contestant at the event, which is extracted from an image of the contestant at the event before the race. The storage medium described in .
55. The model generation process includes generating the estimation model by the learning so that the estimation model estimates a factor contributing to the degree of possibility and a magnitude of contribution of the factor to the degree of possibility. The storage medium according to any one of claims 51 to 54.

Images