30 that is installed in the self-service checkout register 50. For example, the video image data DB 105 stores therein video image data obtained by each of the self-service checkout registers 50 or each of the cameras 30.

The fraud rule DB 106 is a database in which a motion of the hand or a motion of a commodity product that is determined to be a fraudulent behavior is defined. Specifically, the fraud rule DB 106 stores therein a fraudulent behavior pattern that is specified by a past history or the like.

FIG. 5 is a diagram illustrating an example of information stored in the fraud rule DB 106. As illustrated in FIG. 5, the fraud rule DB 106 stores therein items of “a fraud category, a target, and a fraud pattern” in an associated manner. The “fraud category” stored here is information that indicates the content of a fraud (an abnormal behavior). The “target” is information that is used to specify whether a determination target of a fraud is a motion of a hand or a motion of a commodity product. The “fraud pattern” is a change in a movement of the hand that is targeted for determination of a fraud or a change in a movement of a plurality of commodity products that are targeted for determination of a fraud.

In the example illustrated in FIG. 5, a fraud 1 is a fraud pattern that is targeted for a movement of a plurality of commodity products, and that is an abnormal behavior determined to be a fraud in the case where a motion of gripping two commodity products and piling up the two gripped commodity products, and then putting the two gripped commodity products into a bag (a carry-out bag) has been detected. That is, the fraud 1 corresponds to a fraud of a read error called as a label switch or the like.

A fraud 2 is a fraud pattern that is targeted for a movement of a plurality of commodity products, and that is an abnormal behavior determined to be a fraud in the case where a motion of taking out, after a plurality of small commodity products have been scanned multiple times, another commodity product from a shopping basket and putting the other commodity product into a bag has been detected. A fraud 3 is a fraud pattern that is targeted for a movement of a hand, and that is an abnormal behavior determined to be a fraud in the case where a motion of holding small commodity products in both hands has been detected. That is, the fraud 3 corresponds to a fraud of a read error called as a label switch or the like.

A fraud 4 is a fraud pattern that is targeted for a movement of a hand, and that is an abnormal behavior determined to be a fraud in the case where a motion of gripping a commodity product in each of the right hand and the left hand, scanning only one of the commodity products, and putting both of the commodity products into a bag has been detected. A fraud 5 is a fraud pattern that is targeted for a movement of a hand and a movement of a plurality of commodity products, and that is an abnormal behavior determined to be a fraud in the case where a motion of gripping a plurality of commodity products, registering a single piece of commodity product at a checkout register screen, and then putting the plurality of commodity products into a bag has been detected. That is, the fraud 5 corresponds to a fraud called a banana trick or the like or a fraud of bar code hiding or the like.

A description will be given here by referring back to FIG. 3. The control unit 110 is a processing unit that manages the entirety of the information processing apparatus 100 and is implemented by, for example, a processor or the like. The control unit 110 includes a machine learning unit 111, a video image acquisition unit 112, a region specifying unit 113, a tracking unit 114, a fraud detection unit 115, and a warning control unit 116. In addition, the machine learning unit 111, the video image acquisition unit 112, the region specifying unit 113, the tracking unit 114, the fraud detection unit 115, and the warning control unit 116 are implemented by an electronic circuit that is included in a processor or implemented by a process or the like that is executed by the processor.

Machine Learning

The machine learning unit 111 is a processing unit that performs machine learning on the machine learning model 104 by using each of the pieces of training data that are stored in the training data DB 103. FIG. 6 is a diagram illustrating machine learning performed on the machine learning model 104. FIG. 6 illustrates an example in which the HOID is used for the machine learning model 104. As illustrated in FIG. 6, the machine learning unit 111 inputs the input data on the training data to the HOID, and acquires an output result of the HOID. In the output result, the class of the person, the class of the object, the interaction between the person and the object, and the like that are detected by the HOID are included. Then, the machine learning unit 111 calculates error information indicating an error between correct answer information on the training data and the output result of the HOID, and performs machine learning that updates the parameters of the HOID by using error back propagation such that the error is decreased.

In addition, the machine learning unit 111 is also able to generate a model that outputs, in accordance with an input of the image data, a region of a hand, a region of a commodity product, and skeleton information on the hand. FIG. 7 is a diagram illustrating machine learning performed on the machine learning model 104. As illustrated in FIG. 7, the machine learning unit 111 inputs, to the machine learning model 104, the training data in which “image data” is used as an explanatory variable and the items of “the region of the hand, the region of the commodity product, and the skeleton information on the hand” are used as objective variables (correct answer information), and then, calculates error information indicating an error between the output result of the items of “the region of the hand, the region of the commodity product, and skeleton information on the hand” and the objective variable of the items of “the region of the hand, the region of the commodity product, and the skeleton information on the hand” in the machine learning model 104. Then, the machine learning unit 111 performs machine learning that updates the parameters of the machine learning model 104 by using error back propagation such that the error is decreased.

Here, skeleton information is information illustrated in, for example, FIG. 8. FIG. 8 is a diagram illustrating an example of the skeleton information. It is possible to use, for the skeleton information, 18 pieces (number 0 to number 17) of definition information that are obtained by numbering each of joints specified by a known skeleton model. For example, a right shoulder joint (SHOULDER RIGHT) is indicated by number 7, a left elbow joint (ELBOW_LEFT) is indicated by number 5, a left knee joint (KNEE LEFT) is indicated by number 11, and a right hip joint (HIP RIGHT) is indicated by number 14. Therefore, it is possible to acquire 18 pieces of coordinate information on the skeleton illustrated in FIG. 8 from the image data, and, for example, “the X coordinates=X7, the Y coordinates=Y7, and the Z coordinates=Z7” are acquired as the position of the right shoulder joint indicated by number 7. Furthermore, for example, the Z-axis may be defined as a horizontal direction, and the Y-axis may be defined as a vertical direction.

Furthermore, as the machine learning model 104, it may be possible to use the machine learning model illustrated in FIG. 6 or the machine learning model illustrated in FIG. 7, or it may be possible to use both the machine learning models illustrated in FIG. 6 and FIG. 7.

Video Image Acquisition

The video image acquisition unit 112 is a processing unit that acquires video image data from the camera 30. For example, the video image acquisition unit 112 acquires video image data from the camera 30 installed in the self-service checkout register 50 at any time, and stores therein the video image data DB 105.

Specifying of Region

The region specifying unit 113 is a processing unit that specifies, from the video image data by inputting video image data to the machine learning model 104, a region of a hand of a person and a region of a commodity product that is being gripped in the hand of the person. For example, the region specifying unit 113 specifies, by using the HOID, the first region that includes the hand of the person, the second region that includes the commodity product, and the relationship between the first region and the second region, and specifies a behavior of the person exhibited with respect to the commodity product.

FIG. 9 is a diagram illustrating behavioral identification performed by using the HOID. As illustrated in FIG. 9, the region specifying unit 113 inputs each of the pieces of image data included in the video image data to the HOID, and acquires an output result obtained from the HOID. As described above, in the output result obtained from the HOID, the Bbox of the person, the Bbox of the object, a probability value of the interaction between the person and the object, a class name, and the like are included. Then, the region specifying unit 113 identifies one of the behaviors indicated by the following drawings (a), (b), and (c) on the basis of the output result obtained from the HOID.

For example, as indicated by the drawing (a) illustrated in FIG. 9, the region specifying unit 113 identifies (specifies) a person and a commodity product and identifies (specifies) that the person is gripping the commodity product from the output result obtained from the HOID. Furthermore, the position of the shopping basket may be assigned by an administrator or the like, or may be automatically assigned by using another machine learning model or the like. In addition, as indicated by the drawing (b) illustrated in FIG. 9, by identifying the person and the commodity product and identifying that the person is gripping the commodity product from the output result obtained from the HOID, the region specifying unit 113 also specifies a commodity product that is allowed to pass through the scan position of each of the codes, such as a bar code or two-dimensional code, at the self-service checkout register 50. In addition, as indicated by the drawing (c) illustrated in FIG. 9, by identifying the person and the commodity product, identifying that the person is gripping the commodity product, identifying a person and a checkout bag, and identifying that the person is gripping the checkout bag from the output result obtained from the HOID, the region specifying unit 113 also specifies the commodity product that has been put into the checkout bag by the user.

In addition, by inputting the video image data to the machine learning model 104, the region specifying unit 113 is also able to specify, from the video image data, the region of the hand of the person, the region of the commodity product that is being gripped in the hand of the person, and the skeleton information on the person. In addition, by associating each of the behaviors with a transition of the skeleton information obtained at the time of each of the behaviors, the region specifying unit 113 is also able to specify, from each of the regions and the skeleton information specified from the video image data, a behavior of the person exhibited with respect to the commodity product and a behavior of the person performed on the self-service checkout register 50 included in the video image data.

Tracking

The tracking unit 114 is a processing unit that tracks, in the case where the number of commodity products that are being gripped in a hand of a person is plural, either a movement of the hand of the person who is gripping the plurality of commodity products, or, a movement of the plurality of commodity products. Specifically, the tracking unit 114 performs tracking in the case where a person who is gripping a plurality of commodity products has been detected in a specific result obtained from the frame included in the video image data by the region specifying unit 113. In other words, the tracking unit 114 tracks a motion related to the same commodity products in continuous frames that are present subsequent to a certain frame, in which the plurality of commodity products have been identified, included in the video image data. Then, the tracking unit 114 stores the tracking result in the storage unit 102, and outputs the tracking result to the fraud detection unit 115.

In addition, in the case where the number of bounding boxes of the commodity products is plural in the output result obtained from the HOID, the tracking unit 114 determines whether or not the positions of the plurality of bounding boxes are overlapped, and, in the case where the positions of the bounding boxes are overlapped, the tracking unit 114 is also able to perform tracking.

Fraud Detection

The fraud detection unit 115 is a processing unit that detects an abnormality of a behavior of registering a commodity product to the self-service checkout register 50 on the basis of a change in a movement of a hand that has been tracked by the tracking unit 114 or a change in a movement of a plurality of commodity products that have been tracked by the tracking unit 114. Specifically, the fraud detection unit 115 detects a fraudulent behavior in the case where a pattern of a movement of a tracked hand or a pattern of a movement of a tracked commodity product corresponds to the pattern that is stored in the fraud rule DB 106.

For example, the fraud detection unit 115 detects a fraudulent behavior indicated by the fraud 1 in the case where “a motion of gripping two commodity products and piling up the two gripped commodity products, and then putting the two gripped commodity products into a bag (carry-out bag)” has been detected.

FIG. 10 is a diagram illustrating an example of a fraud detection in the case where two commodity products have been gripped. As illustrated in FIG. 10, the fraud detection unit 115 detects a fraud in the case where the fraud detection unit 115 has detected a motion of piling up the commodity products from a result of specifying a region obtained by the region specifying unit 113. For example, in the case where the fraud detection unit 115 has detected a motion of piling a high-priced commodity product A on top of a low-priced commodity product B, the fraud detection unit 115 detects a fraud (for example, a label switch) conducted by scanning a bar code of the low-priced commodity product instead of a bar code of the high-priced commodity product.

In addition, even when a pattern is other than the patterns that are stored in the fraud rule DB 106, the fraud detection unit 115 is also able to detect a fraudulent behavior in the case where the fraud detection unit 115 has detected a motion that has been defined as a fraudulent behavior and that indicates that “a movement of a hand is a motion of piling up a plurality of commodity products”. Similarly, the fraud detection unit 115 is also able to detect a fraudulent behavior in the case where the fraud detection unit 115 has detected a motion that has been defined as a fraudulent behavior in advance and that indicates that “a motion of piling up the commodity products” or “a motion of piling up the bar codes”.

FIG. 11 is a diagram illustrating an example of fraud detection performed on the basis of the fraudulent behavior that has been defined in advance. As illustrated in FIG. 11, the fraud detection unit 115 detects a fraud in the case where, after a motion of gripping a commodity product has been detected from the result of specifying a region obtained by the region specifying unit 113, the fraud detection unit 115 detects from the self-service checkout register 50 that the commodity product has been registered without a scan motion being detected. For example, the fraud detection unit 115 detects a fraud (for example, a banana trick) in the case where, after a high-priced commodity product (for example, a melon) to which a bar code is not assigned has been gripped, a low-priced commodity product (for example, a banana) that is the same kind of commodity product, though, has been registered at a registration screen of the self-service checkout register 50.

For example, the fraud detection unit 115 detects a fraudulent behavior indicated by the fraud 4 by determining the behavior corresponds to the fraud 4 included in the fraud rule illustrated in FIG. 5 in the case where “a motion of gripping a wine in the left hand, gripping a wine in the right hand, and then, scanning only the wine gripped in the right hand” has been detected as the pattern of a movement of the hand.

For example, the fraud detection unit 115 specifies, from the output result obtained from the HOID, the positional relationship with the area that is set in order to scan a commodity product to the self-service checkout register 50. Then, the fraud detection unit 115 detects a fraudulent behavior when it is determined, from the output result obtained from the HOID, that a bar code of each of the two commodity products has not been scanned to the self-service checkout register 50. In other words, the fraud detection unit 115 detects a fraudulent behavior in the case where both of the tow commodity products do not pass through the scan region that is set to each of the cameras.

FIG. 12 is a diagram illustrating an example of fraud detection on the basis of the positional relationship. As illustrated in FIG. 12, the fraud detection unit 115 detects a fraud in the case where a motion of holding commodity products in both hands, and a motion of scanning a commodity product at a position that is different from a supposed bar code position have been detected from a result of specifying a region obtained by the region specifying unit 113. For example, the fraud detection unit 115 detects a fraud conducted by scanning the bar code of a single piece of can that is included in a package of six cans instead of scanning a bar code of a 6-can packaged commodity product containing six cans each of which a bar code is assigned.

In addition, the fraud detection unit 115 is able to detect a further complicated motion by using the skeleton information that is specified by the machine learning model 104, so that the fraud detection unit 115 is also able to improve the accuracy of the determination of the fraudulent behavior described above. For example, the fraud detection unit 115 specifies, from the skeleton information, a pose of a user, a movement of a finger, a movement of a joint, and the like, and detects a fraudulent behavior by comparing these pattern with the pattern that is prepared in advance.

Alert Report

The warning control unit 116 is a processing unit that generates an alert and that performs report control of the alert in the case where a fraudulent behavior (fraudulent motion) has been detected by the fraud detection unit 115. For example, the warning control unit 116 generates an alert indicating that the commodity product that has not been registered to the self-service checkout register 50 by a person, or, an alert indicating that the commodity product that has been registered to the self-service checkout register 50 by a person is abnormal, and outputs the generated alert to the self-service checkout register 50 and the administrator terminal 60.

In addition, if the warning control unit 116 generates an alert related to an abnormality of a behavior of registering a commodity product to the self-service checkout register 50, the warning control unit 116 outputs, from the self-service checkout register 50, a voice or a screen that makes the person located at the self-service checkout register 50 aware of a registration omission of the commodity product.

FIG. 13 is a diagram illustrating a report example of the alert. As illustrated in FIG. 13, the warning control unit 116 causes the message indicating that “Is there any commodity product for a scan omission? Please scan the commodity products again.”, or the like to be displayed on a display screen, such as a touch panel, of the self-service checkout register 50.

In addition, the warning control unit 116 causes a warning lamp installed in the self-service checkout register 50 to be turned on, causes an identifier of the self-service checkout register 50 and a message indicating that a fraud possibly occurs to be displayed on the administrator terminal 60, and causes the identifier of the self-service checkout register 50 and a message indicating an occurrence of the fraud and indicating a need to check to be transmitted to the terminal that is used by a store clerk who is present in an inside of a store.

In addition, in the case where the warning control unit 116 generates an alert related to an abnormality of a behavior of registering a commodity product to the self-service checkout register 50, the warning control unit 116 causes the camera 30 included in the self-service checkout register 50 to capture an image of a person, and to store the image data on the captured person and the alert in an associated manner in the storage unit. By doing so, it is possible to collect information on a fraudulent person who exhibits a fraudulent behavior, so that it is possible to make use of various measures to prevent the fraudulent behavior by detecting, at the entrance of the store, a customer who visits the store and who has a history of exhibiting a fraudulent behavior or the like. In addition, the warning control unit 116 is able to detect a fraudulent person from the image data on the person who uses the self-service checkout register 50, detect a fraudulent person at the entrance of the store, and the like by performing supervised learning using the image data on the fraudulent person and by generating the machine learning models. In addition, the warning control unit 116 is also able to acquire and store information on a credit card that is used by a person who has exhibited a fraudulent behavior from the self-service checkout register 50.

Specific Example

In the following, a specific example of the process described above performed by the information processing apparatus 100 will be described. FIG. 14 is a diagram illustrating a series of processes up to an alarm report after tracking and fraud detection. As illustrated in FIG. 14, in the video image data, each of the pieces of image data to which an identifier has been assigned is included. FIG. 14 illustrates an example in which a region and a relationship are specified by using the HOID.

First, for image data 1, when the information processing apparatus 100 specifies items of “a shopping basket, a person, and a mutual relationship (grip) between the person and the shopping basket” by using the region specifying unit 113, a plurality of commodity products are not detected, so that the information processing apparatus 100 determines that this case is not a fraudulent behavior without performing tracking.

Subsequently, for image data 2, when the information processing apparatus 100 specifies items of “a single commodity product, the person, and a mutual relationship (grip) between the person and the commodity product” by using the region specifying unit 113, a plurality of commodity products are not detected, so that the information processing apparatus 100 determines that this case is not a fraudulent behavior without performing tracking.

Subsequently, for image data 3, when the information processing apparatus 100 specifies items of “the commodity product, the person, and a mutual relationship (grip, and scan) between the person and the commodity product at the scan position” by using the region specifying unit 113, a plurality of commodity products are not detected, so that the information processing apparatus 100 determines that this case is not a fraudulent behavior without performing tracking.

Subsequently, for image data 4, when the information processing apparatus 100 specifies items of “the commodity product, the person, and a mutual relationship (grip) between the person and the commodity product in the checkout bag” by using the region specifying unit 113, a plurality of commodity products are not detected, so that the information processing apparatus 100 determines that this case is not a fraudulent behavior without performing tracking.

Subsequently, for image data 5, when the information processing apparatus 100 specifies items of “two commodity products, and a mutual relationship (grip) between the person and the commodity products” by using the region specifying unit 113, a plurality of commodity products have been detected, so that the information processing apparatus 100 starts tracking performed by the tracking unit 114. Furthermore, the information processing apparatus 100 performs fraud detection, by using the fraud detection unit 115, on the basis of the change in the movement of the hand obtained after the start of the tracking or the change in the movement of the plurality of commodity products obtained after the start of the tracking.

Subsequently, tracking is performed on image data 6. In other words, in a period of time for which two commodity products are detected by the region specifying unit 113, the information processing apparatus 100 performs fraud detection, by using the fraud detection unit 115, on the basis of the change in the movement of the hand obtained after the start of the tracking or the change in the movement of the plurality of commodity products obtained after the start of the tracking.

Similarly, tracking is performed on image data 7. Here, the information processing apparatus 100 detects a fraudulent behavior in the case where a motion corresponding to the fraud rule has been detected by the fraud detection unit 115 on the basis of the change in the movement of the hand obtained after the start of the tracking or the change in the movement of the plurality of commodity products obtained after the start of the tracking. Then, the information processing apparatus 100 generates an alert and send a report. At this time, the information processing apparatus 100 may ends the tracking, may stop the tracking until the end of a reaction with respect to the alert performed by a store clerk or the like, or may continue the tracking.

Flow of Process

FIG. 15 is a flowchart illustrating the flow of the process. As illustrated in FIG. 15, the information processing apparatus 100 acquires the video image data at any time (Step S101).

Subsequently, if the information processing apparatus 100 receives an instruction to start the process of fraud detection (Yes at Step S102), the information processing apparatus 100 acquires a frame included in the video image data (Step S103). Here, if the video image data is not present, the information processing apparatus 100 ends the process. In contrast, if the video image data is present, the information processing apparatus 100 specifies a region of the hand and a region of the commodity product by using the machine learning model 104 (Step S104).

Then, if a plurality of commodity products are not detected (No at Step S105), the information processing apparatus 100 repeats the process at Step S103 and the subsequent processes. In contrast, a plurality of commodity products have been detected (Yes at Step S105), the information processing apparatus 100 determines whether or not tracking is being performed (Step S106).

Here, if tracking is not being performed (No at Step S106), the information processing apparatus 100 starts the tracking (Step S107), and performs the process at Step S108. In contrast, if tracking is being performed (Yes at Step S106), the information processing apparatus 100 performs the process at Step S108.

In other words, in the case where the motion corresponding to the fraud rule is not detected (No at Step S108), the information processing apparatus 100 repeats the process at Step S103 and the subsequent processes, whereas, if the motion corresponding to the fraud rule has been detected (Yes at Step S108), the information processing apparatus 100 sends an alert report (Step S109), and ends the process.

Effects

As described above, if a number of commodity products that is being gripped in a hand of a person is plural, the information processing apparatus 100 tracks either a movement of the hand of the person who is gripping the plurality of commodity products, or a movement of the plurality of commodity products. Then, the information processing apparatus 100 generates, on the basis of a change in the tracked movement of the hand or a change in the tracked movement of the plurality of commodity products, an alert connected to an abnormality of a behavior of registering the commodity products to the self-service checkout register 50. Therefore, the information processing apparatus 100 is able to detect a fraud conducted at the self-service checkout register 50 without using a weight sensor or the like.

In addition, in the case where the movement of the commodity product has been tracked and in the case where the changes in the movements of the plurality of commodity products included in the region of the commodity products overlap, that is, correspond to a pattern that has been assigned in advance, the information processing apparatus 100 generates the alert connected to the abnormality of the behavior of the commodity products to the self-service checkout register 50. Therefore, the information processing apparatus 100 is able to detect an inevitable error, such as a scan omission, or an intentional fraud.

In addition, in the case where the movement of the hand has been tracked and in the case where the movement of the hand corresponds to the motion of piling up the plurality of commodity products, the information processing apparatus 100 generates the alert connected to the abnormality of the behavior of registering the commodity product to the self-service checkout register 50. Therefore, the information processing apparatus 100 is able to detect a fraud conducted by scanning a bar code of a low-priced commodity product instead of a high-priced commodity product.

In addition, the information processing apparatus 100 acquires, by using the machine learning model 104, a bounding box that indicates the region of the hand, a bounding box of the commodity product that indicates the region of the commodity products, and the skeleton information on the hand of the person. As a result, the information processing apparatus 100 is able to specify a motion of the person and a movement of the commodity product in a short time as compared to image analysis, so that it is possible to implement real-time fraud detection.

In addition, the information processing apparatus 100 specifies, by using the machine learning model 104, the first region that includes the hand of the person, the second region that includes the commodity product, and the relationship between the first region and the second region. Then, on the basis of the first motion of the person gripping the first commodity product in the right hand and the second motion of the person gripping the second commodity product in the left hand, the information processing apparatus 100 detects that a code of one of the commodity products between the first commodity product and the second commodity product has not been scanned by the self-service checkout register 50. Therefore, the information processing apparatus 100 is able to detect a scan omission occurring when only one commodity product is scanned or detect an intentional fraud.

In addition, the information processing apparatus 100 generates an alert indicating that there is a commodity product that has not been registered to the self-service checkout register 50 by a person, or, indicating that the commodity product that has been registered to the self-service checkout register 50 by a person is abnormal. Therefore, a store clerk or the like is able to react by listening the circumstances by using the information processing apparatus 100 before the person who has exhibited the fraudulent behavior leaves the store.

In addition, in the case where an alert related to an abnormality of a behavior of registering the commodity product to the self-service checkout register 50 has been generated, the information processing apparatus 100 outputs, from the self-service checkout register 50, a voice or a screen that makes the person located at the self-service checkout register 50 aware of a registration omission of the commodity product. Therefore, the information processing apparatus 100 is able to directly alert the person who is performing a scan even in a case of an inevitable mistake or even in a case of an intentional fraud, so that it is possible to reduce the mistake or the intentional fraud.

In addition, when an alert related to an abnormality of a behavior of registering the commodity product to the self-service checkout register 50 is generated, the information processing apparatus 100 causes the camera included in the self-service checkout register 50 to capture an image of the person, and stores the image data on the captured person and the alert in an associated manner in the storage unit. Accordingly, the information processing apparatus 100 is able to collect and hold the information on the fraudulent person who exhibits the fraudulent behavior, so that it is possible to make use of various measures to prevent the fraudulent behavior by detecting a visit of the fraudulent person to the store from the captured data obtained by the camera that captures a customer who visits the store. In addition, the information processing apparatus 100 is also able to acquire and store information on a credit card that is used by a person who has exhibited a fraudulent behavior from the self-service checkout register 50, so that it is possible to charge a fee via a credit card company in the case where the fraudulent behavior has been confirmed.

(b) Second Embodiment

In the above explanation, a description has been given of the embodiments according to the present invention; however, the present invention may also be implemented with various kinds of embodiments other than the embodiments described above.

Numerical Value, Etc.

The number of self-service checkout registers and cameras, examples of numerical values, examples of the training data, the number of pieces of training data, the machine the learning models, each of the class names, the number of classes, the data formats, and the like that are used in the embodiment described above are only examples and may be arbitrarily changed. Furthermore, the flow of the processes descried in each of the flowcharts may be changed as long as the processes do not conflict with each other. In addition, a model generated from various algorithms, such as a neural network, may be used for each of the models.

In addition, regarding a scan position and a position of a shopping basket, the information processing apparatus 100 is also able to use a known technology, such as another machine learning model for detecting a position, an object detection technology, or a position detection technology. For example, the information processing apparatus 100 is able to detect a position of a shopping basket on the basis of a difference between frames (image data) and a change in frames in time series, so that the information processing apparatus 100 may also perform detection by using the difference between frames and the change in frames in time series, or may also generate another model by using the difference between frames and the change in frames in time series. Furthermore, by designating a size of the shopping basket in advance, the information processing apparatus 100 is also able to identify the position of the shopping basket in the case where an object with that size has been detected from the image data. In addition, the scan position is a position that is fixed to an extent, so that the information processing apparatus 100 is also able to identify the position designated by an administrator or the like as the scan position.