Store monitoring method and store monitoring device

Abstract

Provided are a store monitoring method and a store monitoring device for performing analysis on a store by monitoring a real-time status inside the store and predicting a future status, so as to provide the monitored and predicted statuses to a user terminal. The store monitoring method may comprise the steps of: receiving a first model distributed from a server; performing training on the first model on the basis of first store monitoring data collected from a store; transmitting, to the server, a first weight update generated via the training on the first model; receiving a second model distributed from the server, the second model having a second weight update generated by combining the first weight update with a weight update provided from another store; performing store analysis on the store on the basis of a model obtained by fusing the trained first model and the second model; and providing a store analysis result to a store curation server.

Description

Store monitoring method and store monitoring device

[0001] Cross-citation with related applications

[0002] This application claims the benefit of priority from Korean Patent Application No. 10-2024-0056247, filed April 26, 2024, the entire contents of which are incorporated herein by reference.

[0003] The disclosure relates to a store monitoring method and a store monitoring device.

[0004] Between stores seeking to attract customers and customers seeking stores that suit their preferences, information about nearby stores can be provided to customers based on their device's location. For example, if store information is pre-registered on an online map service, when the customer's device accesses the service, the customer's location, measured by the GPS (Global Positioning System) module installed on the customer's device, and information about stores within a predetermined range can be displayed on a map or listed on the customer's device screen. However, recently, technology that goes beyond simply providing information such as store name, business hours, and contact information, and utilizes IoT (Internet of Things) and artificial intelligence to recommend stores based on the customer's various preferences and provide real-time store status information, is gaining attention.

[0005] The task to be solved is to provide a store monitoring method and a store monitoring device that can improve the performance of analyzing the current status of a store by transmitting only the weights of a model generated through learning based on monitoring data collected for multiple stores to a server and then distributing a model with weights reflecting the general characteristics of each store generated on the server based on this.

[0006] In addition, another task to be solved is to provide a store monitoring method and a store monitoring device that can improve the performance of analyzing the status of a store by reflecting the individual characteristics of each store while based on universal characteristics of monitoring data collected for multiple stores.

[0007] A store monitoring method according to one embodiment is a store monitoring method that performs analysis on a store to monitor real-time status inside a store and predict future status and provide the same to a user terminal, the store monitoring method comprising: a step of distributing a first model from a server; a step of performing learning on the first model based on first store monitoring data collected from a store; a step of transmitting a first weight update generated by learning on the first model to the server; a step of distributing a second model from the server having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores; a step of performing store analysis on the store based on a model that combines the learned first model and the second model; and a step of providing a store analysis result to a store curation server.

[0008] In some embodiments, the store analysis results may include data regarding at least one of table or seat layout status, table or seat occupancy status, noise status, status of music played in the store, status of store music playback history, status of visiting customer gender, status of visiting customer age group, status of visiting customer behavior, status of visiting customer stay time, status of orders by menu, and status of interior design.

[0009] In some embodiments, the step of performing store analysis on the store based on a model that fuses the learned first model and the second model may include the steps of generating a third weight update by fuse the first weight update and the second weight update; obtaining a third model having the third weight update; and performing store analysis on the store using the third model.

[0010] In some embodiments, the store monitoring method further comprises a step of reading data representing a ratio set for each store from a memory, and the step of obtaining the third model may comprise a step of obtaining the third model having the third weight update calculated according to the following mathematical expression 1, when the ratio is a:b:

[0011] (Equation 1)

[0012] w3= a * w1+ b * w2

[0013] Here, w1 is the first weight update, w2 is the second weight update, and w3 is the third weight update.

[0014] In some embodiments, the sum of a and b may be 1.

[0015] In some embodiments, the store monitoring method may further include: transmitting the third weight update to the server; receiving from the server a fourth model having a fourth weight update generated by synthesizing the third weight update and weight updates provided from other stores; and performing a store analysis on the store based on a model that fuses the third model and the fourth model.

[0016] In some embodiments, the step of performing store analysis for the store based on a model that fuses the third model and the fourth model may include the steps of generating a fifth weight update by fuse the third weight update and the fourth weight update; obtaining a fifth model having the fifth weight update; and performing store analysis for the store using the fifth model.

[0017] In some embodiments, the store monitoring method further comprises a step of reading data representing a ratio set for each store from a memory, and the step of obtaining the fifth model may comprise a step of obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 2 when the ratio is a:b; and a step of obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 3 when the ratio is changed to c:d:

[0018] (Equation 2)

[0019] w5= a * w3+ b * w4

[0020] (Equation 3)

[0021] w5= c * w3+ d * w4

[0022] Here, w3 is the third weight update, w4 is the fourth weight update, and w5 is the fifth weight update.

[0023] In some embodiments, the store monitoring method further comprises a step of reading data representing a ratio set for each store from a memory, and the step of performing an analysis on the store by fusing the learned first model and the second model may comprise a step of performing an analysis on the store according to the following mathematical expression 4 when the ratio is a:b:

[0024] (Equation 4)

[0025] y3= a * y1+ b * y2

[0026] Here, y1 is the predicted value of the learned first model, y2 is the predicted value of the second model, and y3 is the predicted value of the third model.

[0027] In some embodiments, the sum of a and b may be 1.

[0028] According to one embodiment, a store monitoring device is a store monitoring device that performs analysis on a store to monitor real-time conditions inside a store and predict future conditions and provide them to a user terminal by executing program codes loaded on one or more memory devices through one or more processors, wherein the program code is executed to receive a first model from a server, perform learning on the first model based on first store monitoring data collected from a store, transmit a first weight update generated by learning on the first model to the server, receive a second model having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores, perform store analysis on the store based on a model that combines the learned first model and the second model, and provide a store analysis result to a store curation server.

[0029] In some embodiments, the store analysis results may include data regarding at least one of table or seat layout status, table or seat occupancy status, noise status, status of music played in the store, status of store music playback history, status of visiting customer gender, status of visiting customer age group, status of visiting customer behavior, status of visiting customer stay time, status of orders by menu, and status of interior design.

[0030] In some embodiments, performing store analysis for the store based on a model that fuses the learned first model and the second model may include generating a third weight update by fuse the first weight update and the second weight update, obtaining a third model having the third weight update, and performing store analysis for the store using the third model.

[0031] In some embodiments, the program code may read data representing a ratio set for each store from the memory of the store monitoring device, and obtaining the third model may include obtaining the third model having the third weight update calculated according to the following mathematical expression 1, when the ratio is a:b:

[0032] (Equation 1)

[0033] w3= a * w1+ b * w2

[0034] Here, w1 is the first weight update, w2 is the second weight update, and w3 is the third weight update.

[0035] In some embodiments, the sum of a and b may be 1.

[0036] In some embodiments, the program code may transmit the third weight update to the server, receive from the server a fourth model having a fourth weight update generated by synthesizing the third weight update and weight updates provided from other stores, and perform a future status prediction for the store based on a model that fuses the third model and the fourth model.

[0037] In some embodiments, performing store analysis for the store based on a model that fuses the third model and the fourth model may include generating a fifth weight update by fuse the third weight update and the fourth weight update, obtaining a fifth model having the fifth weight update, and performing store analysis for the store using the fifth model.

[0038] In some embodiments, the program code may include reading data representing a ratio set for each store from a memory of the store monitoring device, and obtaining the fifth model, wherein when the ratio is a:b, the fifth model having the fifth weight update calculated according to the following mathematical expression 2 is obtained, and when the ratio is changed to c:d, the fifth model having the fifth weight update calculated according to the following mathematical expression 3 is obtained:

[0039] (Equation 2)

[0040] w5= a * w3+ b * w4

[0041] (Equation 3)

[0042] w5= c * w3+ d * w4

[0043] Here, w3 is the third weight update, w4 is the fourth weight update, and w5 is the fifth weight update.

[0044] In some embodiments, the program code may include reading data representing a ratio set for each store from a memory of the store monitoring device, and performing an analysis on the store by fusing the learned first model and the second model, wherein when the ratio is a:b, the analysis on the store may be performed according to the following mathematical expression 4.

[0045] (Equation 4)

[0046] y3= a * y1+ b * y2

[0047] Here, y1 is the predicted value of the learned first model, y2 is the predicted value of the second model, and y3 is the predicted value of the third model.

[0048] In some embodiments, the sum of a and b may be 1.

[0049] According to embodiments, only the weights of a model generated through learning based on monitoring data collected from multiple stores are transmitted to the server, and a model with weights reflecting the general characteristics of each store is distributed based on this, thereby improving the performance of analyzing the current status of each store. Specifically, since the server does not collect data collected from the actual store but only processes the weight information, the data collected from the store is not shared with other devices, including the server, but only the weights are shared. This protects personal information that may be included in the collected data. In addition, since only the weights are transmitted instead of the entire data, the communication load is reduced and communication resources are used efficiently. Furthermore, by synthesizing the unique characteristics of multiple stores to generate weights reflecting the general characteristics of these stores and generating a model applying these weights and distributing this model to each store, the problem of data sparsity or data imbalance can be resolved, and the performance of each store monitoring device in analyzing the current status of the store can be improved.

[0050] Furthermore, according to embodiments, the performance of analyzing the status of stores can be improved by reflecting the individual characteristics of each store while basing the monitoring data collected from multiple stores on a common characteristic basis. Specifically, the unique characteristics of multiple stores are synthesized to generate weights that reflect the common characteristics of these stores, a model applying these weights is created, and the model is distributed to each store. This weight is then added to the model distributed to each store, thereby reflecting the unique characteristics. This prevents the unique characteristics of the store monitoring devices operating at each store from being overly generalized.

[0051] Figure 1 is a block diagram illustrating a store analysis system according to one embodiment.

[0052] Figures 2 and 3 are block diagrams showing the operation of a store analysis system according to one embodiment.

[0053] Figure 4 is a flowchart illustrating the operation of a store analysis system according to one embodiment.

[0054] Figures 5 to 7 are block diagrams showing the operation of a store analysis system according to one embodiment.

[0055] Figure 8 is a flowchart illustrating the operation of a store analysis system according to one embodiment.

[0056] Figure 9 is a flowchart illustrating a store monitoring method according to one embodiment.

[0057] Figure 10 is a block diagram illustrating a store analysis system according to one embodiment.

[0058] FIG. 11 is a drawing showing an example of a screen displayed on a customer terminal according to one embodiment.

[0059] Figure 12 is a block diagram illustrating a computing device according to one embodiment.

[0060] Below, with reference to the attached drawings, embodiments of the present invention are described in detail so that those skilled in the art can easily implement the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in the drawings, parts irrelevant to the description are omitted for clarity of description, and similar parts are designated with similar reference numerals throughout the specification.

[0061] Throughout the specification and claims, whenever a part is referred to as "comprising" a component, this does not exclude other components, but rather includes other components, unless otherwise stated. Terms including ordinal numbers, such as "first," "second," etc., may be used to describe various components, but these components are not limited by these terms. These terms are used solely to distinguish one component from another.

[0062] Terms such as "...part," "...device," and "module" described in the specification may refer to a unit capable of processing at least one function or operation described in the specification, which may be implemented by hardware or a circuit, software, or a combination of hardware or a circuit and software. In addition, at least some of the components or functions of the store monitoring method and store monitoring device according to the embodiments described below may be implemented by a program or software, and the program or software may be stored on a computer-readable medium.

[0063] Figure 1 is a block diagram illustrating a store analysis system according to one embodiment.

[0064] Referring to FIG. 1, a store analysis system according to one embodiment may include a plurality of store monitoring devices (10, 12, 14) and a server (20). The store monitoring devices (10, 12, 14) may be implemented as IoT devices that monitor the real-time status inside a store (e.g., a cafe). The server (20) may perform processing that takes into account the universal characteristics of monitoring data collected for multiple stores in conjunction with the plurality of store monitoring devices (10, 12, 14). Meanwhile, the plurality of monitoring devices (10, 12, 14) may, in addition to the processing of the server (20), perform processing that reflects the individual characteristics of each store. To this end, the plurality of store monitoring devices (10, 12, 14) and the server (20) may exchange data with each other via a network. Networks may include, for example, wireless networks, including Wi-Fi networks, Bluetooth networks, and cellular networks, wired networks, including local area networks (LANs), and mixed wireless and wired networks.

[0065] In some embodiments, the store monitoring device (10), the store monitoring device (12), and the store monitoring device (14) may all be installed in different stores. Specifically, the store monitoring device (10) may be installed in a store located at a certain location, the store monitoring device (12) may be installed in a store located at a different location, and the store monitoring device (14) may be installed in a store located at yet another location.

[0066] In some other embodiments, some of the store monitoring device (10), the store monitoring device (12), and the store monitoring device (14) may be installed in different stores, and others may be installed in the same store but at distinct locations within the store (different locations within the same store). Specifically, for example, the store monitoring device (10) and the store monitoring device (12) may be installed in different stores, and the store monitoring device (10) and the store monitoring device (14) may be installed in the same store but at distinct locations within the store.

[0067] It should be noted that the term "different stores" is used herein to encompass not only different stores having different locations, but also locations within the same store that are recognized as the same store but have different locations.

[0068] The store monitoring device (10) may include a sensor for collecting the surrounding environment within the store, one or more processors for processing the collected data, and a communication interface for communicating with other devices. The store monitoring device (10) may communicate with the server (20) via the communication interface to transmit or receive data, and may process the store monitoring data collected within the store to perform store analysis on the store. Specifically, the store monitoring device (10) may perform store analysis of the store based on the store monitoring data collected within the store based on an artificial intelligence neural network model (also referred to as a “model”). The store analysis results are provided to the store curation server (32), which will be described later with reference to FIG. 10, and the store curation server (32) may predict the future status of the store based on the store analysis results and provide store curation services to the user via user terminals (34, 36).

[0069] Since the store monitoring device (12, 14) has substantially the same operating configuration as the store monitoring device (10), the description thereof may be omitted by referring to the description of the store monitoring device (10). Therefore, the store monitoring device (10) will be described in detail herein.

[0070] The store monitoring device (10) can perform analysis on the store to monitor real-time conditions inside the store and predict future conditions and provide them to a user terminal by executing program codes loaded on one or more memory devices via one or more processors. Here, the one or more processors and the one or more memory devices may have configurations corresponding to the processor (501) and memory (502) of the computing device (50) described below with reference to FIG. 12, and the corresponding program can be executed to implement the functions or operations of the store monitoring device (10) described below.

[0071] Specifically, the store monitoring device (10) can receive a first model from the server (20). Here, the first model may refer to an initial model created to perform store analysis on the store. That is, the server (20) can create an initial model implemented with a predetermined neural network type and distribute the created initial model to the store monitoring device (10). In addition, the server (20) can also distribute the created initial model to the store monitoring device (12) and the store monitoring device (14). The store monitoring device (10) can collect store monitoring data by monitoring the real-time status inside the store through IoT technology that embeds sensing and communication functions in objects and connects them to the Internet.

[0072] The store monitoring device (10) can perform learning based on store monitoring data collected from the store for the first model distributed from the server (20). In some embodiments, the store monitoring data may include image or video data collected from the store via a camera, sound data collected from the store via a microphone, etc. The store monitoring device (10) can transmit the weights of the model corresponding to the results of such learning to the server (20) as weight updates. At this time, the weight updates transmitted from the store monitoring device (10) to the server (20) may include weights that reflect the unique characteristics of the store monitored by the store monitoring device (10).

[0073] Similarly, the store monitoring device (12) may receive a first model from the server (20) and perform learning on the first model based on store monitoring data from other stores or other locations that the store monitoring device (12) monitors. The store monitoring device (12) may transmit a weight update including weights reflecting the unique characteristics of the store monitored by the store monitoring device (12) to the server (20).

[0074] Similarly, the store monitoring device (14) may receive a first model from the server (20) and perform learning on the first model based on store monitoring data from other stores or other locations that the store monitoring device (14) monitors. The store monitoring device (14) may transmit a weight update including weights reflecting the unique characteristics of the store monitored by the store monitoring device (14) to the server (20).

[0075] The server (20) can generate a new weight update by synthesizing the weight update provided from the store monitoring device (10), the weight update provided from the store monitoring device (12), and the weight update provided from the store monitoring device (14). The server (20) can newly generate a second model with the new weight update applied and distribute it to a plurality of store monitoring devices (10, 12, 14). Here, the new weight update can include weights that reflect the general characteristics of the store monitored by the store monitoring device (10), the store monitored by the store monitoring device (12), and the store monitored by the store monitoring device (14).

[0076] Thereafter, the store monitoring device (10) can receive a second model to which weights reflecting the general characteristics of multiple stores are applied from the server (20), and can perform store analysis on the store related to the store monitoring device (10) based on the second model. Specifically, the store monitoring device (10) can generate a store analysis result including data on at least one of table or seat arrangement status, table or seat occupancy status, noise status, status of music played in the store, status of store music playback history, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay of visiting customers, order status by menu, and interior status through the store analysis, and can provide the store analysis result to a store curation server (32) to be described later with reference to FIG. 10, and the store curation server (32) can predict the future status of the store based on the store analysis result. In some embodiments, the result of the store curation server (32) predicting the future status of the store may include data regarding at least one of: prediction of changes in table or seat occupancy, prediction of seat atmosphere, prediction of waiting time, prediction of changes in noise, prediction of music to be played in the store, prediction of gender of visiting customers, prediction of age group of visiting customers, prediction of behavioral patterns of visiting customers, prediction of purpose of visit, prediction of menu ranking, prediction of recommended menu, sales prediction, inventory prediction, and prediction of atmosphere inside the store.

[0077] Likewise, the store monitoring device (12) and the store monitoring device (14) also receive a second model with weights applied that reflect the general characteristics of multiple stores from the server (20), and based on this, store analysis can be performed on stores related to the store monitoring device (12) and the store monitoring device (14), respectively.

[0078] The store analysis results of the store monitoring device (10), the store analysis results of the store monitoring device (12), and the store analysis results of the store monitoring device (14) are provided to the store curation server (32), and the store curation server (32) can predict the future status of each store based on the respective store analysis results. In addition, the store curation server (32) can execute a store curation service that recommends stores to users based on the future status prediction results.

[0079] In the present embodiment, the store monitoring device (10), the store monitoring device (12), and the store monitoring device (14) perform learning on the first model based on their respective store monitoring data, and transmit only the weights of the learned first model to the server (20), so that the server (20) can process only the weight information without collecting data collected from the actual store. Accordingly, the data collected from each store monitoring device (10, 12, 14) is not shared with the server (20) or other devices, but only the weights are shared, so that personal information that may be included in the collected data is protected, and since only the weights are transmitted without transmitting the entire data, the communication load can be reduced and communication resources can be used efficiently. Meanwhile, the server (20) synthesizes the unique characteristics of the store monitored by the store monitoring device (10), the unique characteristics of the store monitored by the store monitoring device (12), and the unique characteristics of the store monitored by the store monitoring device (14), generates weights reflecting the universal characteristics of these stores, generates a model applying the weights, and distributes the model to each store monitoring device (10, 12, 14), thereby resolving the data scarcity problem or the data imbalance problem and improving the performance of each store monitoring device (10, 12, 14) in analyzing the current status of the store.

[0080] Meanwhile, in some embodiments, the store monitoring device (10) may receive a second model with weights applied that reflect the general characteristics of multiple stores from the server (20), and perform store analysis on stores related to the store monitoring device (10) based on a model that combines the first model and the second model that have been trained in the store monitoring device (10). Similarly, the store monitoring device (12) and the store monitoring device (14) may also receive a second model with weights applied that reflect the general characteristics of multiple stores from the server (20), and perform store analysis on stores related to the store monitoring device (12) and the store monitoring device (14) based on a model that combines the first model and the second model that have been trained in each of the store monitoring device (12) and the store monitoring device (14).

[0081] In this embodiment, by synthesizing the unique characteristics of the store monitored by the store monitoring device (10), the unique characteristics of the store monitored by the store monitoring device (12), and the unique characteristics of the store monitored by the store monitoring device (14), a weight reflecting the universal characteristics of these stores is generated, a model applying the weight is generated, and the model is distributed to each store monitoring device (10, 12, 14). Then, each store monitoring device (10, 12, 14) can reflect the unique characteristics of each store in addition to the distributed model. Accordingly, it is possible to prevent the unique characteristics of the store monitoring devices (10, 12, 14) from being excessively generalized by reflecting the individual characteristics of each store together based on the universal characteristics for the monitoring data collected for multiple stores.

[0082] Figures 2 and 3 are block diagrams showing the operation of a store analysis system according to one embodiment.

[0083] Referring to FIG. 2, in a store analysis system according to an embodiment, a store monitoring device (10, 12, 14) may receive a first model from a server (20). The store monitoring device (10) may perform learning on the first model based on first store monitoring data (MD1) collected from the first store. At this time, learning is performed only in the store monitoring device (10), and the server (20) does not access the first store monitoring data (MD1). Thereafter, the store monitoring device (10) may transmit a first weight update (WU1) generated by learning on the first model to the server (20). Here, the first weight update (WU1) may include weights reflecting the unique characteristics of the first store.

[0084] Meanwhile, the store monitoring device (12) can perform learning on the first model based on the second store monitoring data (MD2) collected from the second store. At this time, learning is performed only in the store monitoring device (12), and the server (20) does not access the second store monitoring data (MD2). Thereafter, the store monitoring device (12) can transmit the second weight update (WU2) generated through learning on the first model to the server (20). Here, the second weight update (WU2) can include weights that reflect the unique characteristics of the second store.

[0085] Meanwhile, the store monitoring device (14) can perform learning on the first model based on the third store monitoring data (MD3) collected from the third store. At this time, learning is performed only in the store monitoring device (14), and the server (20) does not access the third store monitoring data (MD3). Thereafter, the store monitoring device (14) can transmit the third weight update (WU3) generated through learning on the first model to the server (20). Here, the third weight update (WU3) can include weights that reflect the unique characteristics of the third store.

[0086] As described above, the first store monitoring data (MD1), the second store monitoring data (MD2), and the third store monitoring data (MD3) may include image or video data collected in the store via a camera, sound data collected in the store via a microphone, etc.

[0087] Next, referring to FIG. 3, the server (20) may generate a fourth weight update (WU4) by synthesizing the first weight update (WU1), the second weight update (WU2), and the third weight update (WU3), and may generate a second model (referred to as a “weight update (WU4)-based model” in FIG. 3) having the weights of the fourth weight update (WU4). In some embodiments, the server (20) may calculate an average of the values ​​corresponding to the first weight update (WU1), the second weight update (WU2), and the third weight update (WU3) to generate the fourth weight update (WU4). Then, the server (20) may distribute the generated second model to the store monitoring devices (10, 12, 14). Here, the fourth weight update (WU4) may include weights that reflect the general characteristics of the first store, the second store, and the third store.

[0088] The store monitoring device (10) can perform a store analysis on the first store based on the second model distributed from the server (20), and accordingly, can generate data on at least one of the following: table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of store music playback, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay of visiting customers, order status by menu, and interior status of the first store as the first analysis result (R1).

[0089] Meanwhile, the store monitoring device (12) can perform a store analysis on the second store based on the second model distributed from the server (20), and accordingly, can generate data on at least one of the table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of store music playback, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay of visiting customers, order status by menu, and interior status as a second analysis result (R2).

[0090] Meanwhile, the store monitoring device (14) can perform a store analysis on the third store based on the second model distributed from the server (20), and accordingly, can generate data on at least one of the table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of store music playback, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay of visiting customers, order status by menu, and interior status of the third store as a third analysis result (R3).

[0091] Thereafter, the first analysis result (R1), the second analysis result (R2), and the third analysis result (R3) are provided to the store curation server (32) to perform a prediction of the future status of the stores, and as a result, for each store, data on at least one of prediction of changes in table or seat occupancy, prediction of seat atmosphere, prediction of waiting time, prediction of noise change, prediction of music to be played in the store, prediction of visiting customer gender, prediction of visiting customer age group, prediction of visiting customer behavior pattern, prediction of visit purpose, prediction of menu ranking, prediction of recommended menu, sales prediction, inventory prediction, and prediction of in-store atmosphere can be obtained.

[0092] In the present embodiment, the store monitoring device (10), the store monitoring device (12), and the store monitoring device (14) perform learning on the first model based on their respective store monitoring data, and transmit only the weights of the learned first model to the server (20), so that the server (20) can process only the weight information without collecting data collected from the actual store. Accordingly, the data collected from each store monitoring device (10, 12, 14) is not shared with the server (20) or other devices, but only the weights are shared, so that personal information that may be included in the collected data is protected, and since only the weights are transmitted without transmitting the entire data, the communication load can be reduced and communication resources can be used efficiently. Meanwhile, the server (20) synthesizes the unique characteristics of the store monitored by the store monitoring device (10), the unique characteristics of the store monitored by the store monitoring device (12), and the unique characteristics of the store monitored by the store monitoring device (14), generates weights reflecting the universal characteristics of these stores, generates a model applying the weights, and distributes the model to each store monitoring device (10, 12, 14), thereby resolving the data scarcity problem or the data imbalance problem and improving the performance of predicting the future status of the store of each store monitoring device (10, 12, 14).

[0093] Figure 4 is a flowchart illustrating the operation of a store analysis system according to one embodiment.

[0094] Referring to FIG. 4, based on the store monitoring device (10), in step (S401), the store monitoring device (10) can receive a first model from the server (20), and in step (S402), the store monitoring device (10) can obtain first store monitoring data. In step (S403), the store monitoring device (10) can perform learning on the first model based on the first store monitoring data.

[0095] In step (S404), the store monitoring device (10) can transmit the first weight update generated through learning for the first model to the server (20). In step (S405), the server (20) can perform learning for a second model having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores. In step (S406), the store monitoring device (10) can receive the second model from the server (20).

[0096] In step (S407), the store monitoring device (10) may perform a store analysis on the first store based on the second model, and may generate data on at least one of the following as an analysis result: a table or seat arrangement status, a table or seat occupancy status, a noise status, a status of music played in the store, a status of store music playback history, a status of visiting customer gender, a status of visiting customer age, a status of visiting customer behavior, a status of visiting customer stay time, a status of orders by menu, and a status of interior design. Thereafter, in step (S408), the store monitoring device (10) may provide the analysis result to the store curation server (32).

[0097] The store curation server (32) may, based on the analysis results, obtain data on at least one of table or seat occupancy rate change prediction, seat atmosphere prediction, waiting time prediction, noise change prediction, music to be played in the store, gender prediction of visiting customers, age prediction of visiting customers, behavioral pattern prediction of visiting customers, purpose of visit prediction, menu ranking prediction, recommended menu prediction, sales prediction, inventory prediction, and interior atmosphere prediction for each store, and may present to the customer, through the user terminal (34, 36), stores that match the values ​​that the customer considers important when selecting a store based on these data. That is, the store curation server (32), based on the internal status data of the store, may discover and recommend stores that can actually provide customer satisfaction, rather than allowing the customer to simply select a store close to their current location or a famous franchise store.

[0098] Figures 5 to 7 are block diagrams showing the operation of a store analysis system according to one embodiment.

[0099] Referring to FIG. 5, in a store analysis system according to an embodiment, a store monitoring device (10, 12, 14) may receive a first model from a server (20). The store monitoring device (10) may perform learning on the first model based on first store monitoring data (MD1) collected from the first store. At this time, learning is performed only in the store monitoring device (10), and the server (20) does not access the first store monitoring data (MD1). Thereafter, the store monitoring device (10) may transmit a first weight update (WU1) generated by learning on the first model to the server (20). Here, the first weight update (WU1) may include weights reflecting the unique characteristics of the first store. In addition, first fusion ratio data (FD1) indicating a ratio set for the first store may be loaded into the memory of the store monitoring device (10).

[0100] Meanwhile, the store monitoring device (12) can perform learning for the first model based on the second store monitoring data (MD2) collected from the second store. At this time, learning is performed only in the store monitoring device (12), and the server (20) does not access the second store monitoring data (MD2). Thereafter, the store monitoring device (12) can transmit the second weight update (WU2) generated through learning for the first model to the server (20). Here, the second weight update (WU2) can include weights reflecting the unique characteristics of the second store. In addition, the memory of the store monitoring device (12) can be loaded with second fusion ratio data (FD2) indicating a ratio set for the second store.

[0101] Meanwhile, the store monitoring device (14) can perform learning for the first model based on the third store monitoring data (MD3) collected from the third store. At this time, learning is performed only in the store monitoring device (14), and the server (20) does not access the third store monitoring data (MD3). Thereafter, the store monitoring device (14) can transmit the third weight update (WU3) generated through learning for the first model to the server (20). Here, the third weight update (WU3) can include weights reflecting the unique characteristics of the third store. In addition, the memory of the store monitoring device (14) can be loaded with third fusion ratio data (FD3) indicating a ratio set for the third store.

[0102] Next, referring to FIG. 6, the server (20) may generate a fourth weight update (WU4) by synthesizing the first weight update (WU1), the second weight update (WU2), and the third weight update (WU3), and may generate a second model (referred to as a “weight update (WU4)-based model” in FIG. 6) having the weights of the fourth weight update (WU4). In some embodiments, the server (20) may calculate an average of the values ​​corresponding to the first weight update (WU1), the second weight update (WU2), and the third weight update (WU3) to generate the fourth weight update (WU4). The server (20) may then distribute the generated second model to the store monitoring devices (10, 12, 14). Here, the fourth weight update (WU4) may include weights that reflect the general characteristics of the first store, the second store, and the third store.

[0103] The store monitoring device (10) can read the first fusion ratio data (FD1) loaded in the memory, and, based on the first fusion ratio data (FD1), can fuse the first model learned previously and the second model distributed from the server (20). That is, the store monitoring device (10) can obtain a third model having a new weight update (third weight update) by fusion of the first weight update (WU1) regarding the first model and the fourth weight update (WU4) regarding the second model based on the first fusion ratio data (FD1).

[0104] In some embodiments, obtaining the third model may include obtaining the third model having a weight update calculated according to the following mathematical expression 1, when the ratio of the first fusion ratio data (FD1) is a:b.

[0105] (Equation 1)

[0106] w3= a * w1+ b * w2

[0107] Here, w1 represents the first weight update (WU1), w2 represents the fourth weight update (WU4), and w3 represents the third weight update of the third model. In some embodiments, the sum of a and b may be set to 1.

[0108] In some other embodiments, the fused model may be transmitted back to the server (20), where the server (20) changes the weights and distributes the fused model back to the store monitoring device (10), where the fusion process may be further performed. Specifically, a third weight update may be transmitted to the server (20), and a fourth model having a fourth weight update generated by synthesizing the third weight update and weight updates provided from other stores may be distributed from the server (20), and store analysis may be performed on the store based on the model that fuses the third model and the fourth model in the store monitoring device (10).

[0109] At this time, performing store analysis on a store based on a model that fuses the third model and the fourth model may include generating a fifth weight update by fuses the third weight update and the fourth weight update, obtaining a fifth model having the fifth weight update, and performing store analysis on the store using the fifth model.

[0110] In this case, data representing the ratio set for each store is additionally read from the memory, and obtaining the fifth model may include obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 2 when the ratio is a:b, and obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 3 when the ratio is changed to c:d.

[0111] (Equation 2)

[0112] w5= a * w3+ b * w4

[0113] (Equation 3)

[0114] w5= c * w3+ d * w4

[0115] Here, w3 is the third weight update, w4 is the fourth weight update, and w5 is the fifth weight update.

[0116] In some other embodiments, the predicted value of the third model may be calculated by applying a ratio to the predicted value of the first model and the predicted value of the second model. That is, if the ratio of the first fusion ratio data (FD1) is a:b, analysis of the store may be performed according to the following mathematical expression 4.

[0117] (Equation 4)

[0118] y3= a * y1+ b * y2

[0119] Here, y1 is the predicted value of the trained first model, y2 is the predicted value of the second model, and y3 is the predicted value of the third model. In some embodiments, the sum of a and b may be set to 1.

[0120] Meanwhile, the store monitoring device (12) can read the second fusion ratio data (FD2) loaded in the memory, and, based on the second fusion ratio data (FD2), can fuse the first model learned previously and the second model distributed from the server (20). That is, the store monitoring device (12) can obtain a sixth model having a new weight update by fusion of the second weight update (WU2) regarding the first model and the fourth weight update (WU4) regarding the second model based on the second fusion ratio data (FD2).

[0121] Meanwhile, the store monitoring device (14) can read the third fusion ratio data (FD3) loaded in the memory, and, based on the third fusion ratio data (FD3), can fuse the first model learned previously and the second model distributed from the server (20). That is, the store monitoring device (14) can obtain a seventh model having a new weight update by fusion of the third weight update (WU3) regarding the first model and the fourth weight update (WU4) regarding the second model based on the third fusion ratio data (FD3).

[0122] Here, the first fusion ratio data (FD1), the second fusion ratio data (FD2), and the third fusion ratio data (FD3) are values ​​set for each store, and at least some of the first fusion ratio data (FD1), the second fusion ratio data (FD2), and the third fusion ratio data (FD3) may have different values.

[0123] In some embodiments, a dependency may be established between the store monitoring devices (10, 12, 14). For example, if the store monitoring devices (10) and (12) are installed in the same store, and the store monitoring device (14) is installed in a different store, a dependency setting variable including a value that can reference the store monitoring device (10) (e.g., a unique identifier or network unique address of the store monitoring device (10)) may be allocated to the memory of the store monitoring device (12). If the value of the dependency setting variable allocated on the memory of the store monitoring device (12) is a value that can reference the store monitoring device (10), the store monitoring device (12) may perform fusion according to fusion ratio data that is dependent on the value of the first fusion ratio data (FD1) loaded in the memory of the store monitoring device (10).

[0124] Specifically, after the server (20) distributes the second model to the store monitoring devices (10, 12, 14), the store monitoring device (10) can search for dependency setting variables in its memory before performing model fusion. If no dependency setting variables referencing other store monitoring devices are found in its memory, the store monitoring device (10) can read the first fusion ratio data (FD1) loaded in its memory, and, based on the first fusion ratio data (FD1), fuse the first model previously learned and the second model distributed from the server (20).

[0125] Meanwhile, the store monitoring device (12) can also search for dependency setting variables in its memory before performing model fusion. If, as a result of the search, a dependency setting variable referencing the store monitoring device (10) is found, the store monitoring device (12) can receive the first fusion ratio data (FD1) loaded in the memory of the store monitoring device (10) through the network. Instead of the second fusion ratio data (FD2) loaded in its memory, the store monitoring device (12) can fuse the first model previously learned and the second model distributed from the server (20) based on the first fusion ratio data (FD1) received from the store monitoring device (10).

[0126] Meanwhile, the store monitoring device (14) can also search for dependency setting variables in its memory before performing model fusion. If no dependency setting variables referencing other store monitoring devices are found in its memory, the store monitoring device (14) can read the third fusion ratio data (FD3) loaded in its memory and, based on the third fusion ratio data (FD3), fuse the first model previously learned and the second model distributed from the server (20).

[0127] In this way, through the dependency setting variable, it is possible to manage the fusion ratio data to be maintained consistently between store monitoring devices that must be managed with the same fusion ratio, for example. That is, if a dependency is set between the store monitoring device (10) and the store monitoring device (12) through the dependency setting variable, when the value of the first fusion ratio data (FD1) of the store monitoring device (10) is changed, the changed first fusion ratio data (FD1) can be applied as is in the operation of the store monitoring device (12). Accordingly, it is possible to effectively secure the convenience of management and the consistency of operation of multiple store monitoring devices, and for example, it can be particularly useful in cases where the fusion in each device is performed only at the same ratio, taking into account the special case where the store monitoring device (10) and the store monitoring device (12) are installed in close physical proximity.

[0128] Next, referring to FIG. 7, the store monitoring device (10) can perform a store analysis on the first store based on a third model that combines a first model learned with the first store monitoring data (MD1) of the store monitoring device (10) in the past and a second model distributed from the server (20), and accordingly, data on at least one of the table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of music played in the store, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay time of visiting customers, order status by menu, and interior status of the first store can be generated as a first analysis result (R1).

[0129] Meanwhile, the store monitoring device (12) can perform a store analysis on the second store based on a sixth model that combines the first model learned from the second store monitoring data (MD2) of the store monitoring device (12) in the past and the second model distributed from the server (20), and accordingly, can generate data on at least one of the table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of music played in the store, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay time of visiting customers, order status by menu, and interior status of the second store as a second analysis result (R2).

[0130] Meanwhile, the store monitoring device (14) can perform a store analysis on the third store based on the seventh model that combines the first model learned with the third store monitoring data (MD3) of the store monitoring device (14) in the past and the second model distributed from the server (20), and accordingly, can generate data on at least one of the table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of music played in the store, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay time of visiting customers, order status by menu, and interior status of the third store as the third analysis result (R3).

[0131] In this embodiment, by synthesizing the unique characteristics of the store monitored by the store monitoring device (10), the unique characteristics of the store monitored by the store monitoring device (12), and the unique characteristics of the store monitored by the store monitoring device (14), a weight reflecting the universal characteristics of these stores is generated, a model applying the weight is generated, and the model is distributed to each store monitoring device (10, 12, 14). Then, each store monitoring device (10, 12, 14) can reflect the unique characteristics of each store in addition to the distributed model. Accordingly, it is possible to prevent the unique characteristics of the store monitoring devices (10, 12, 14) from being excessively generalized by reflecting the individual characteristics of each store together based on the universal characteristics for the monitoring data collected for multiple stores.

[0132] Figure 8 is a flowchart illustrating the operation of a store analysis system according to one embodiment.

[0133] Referring to FIG. 8, based on the store monitoring device (10), in step (S801), the store monitoring device (10) can receive a first model from the server (20), and in step (S802), the store monitoring device (10) can obtain first store monitoring data. In step (S803), the store monitoring device (10) can perform learning on the first model based on the first store monitoring data.

[0134] In step (S804), the store monitoring device (10) may transmit the first weight update generated through learning for the first model to the server (20). In step (S805), the server (20) may perform learning for a second model having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores. In step (S806), the store monitoring device (10) may receive the second model from the server (20).

[0135] In step (S807), the store monitoring device (10) can read the fusion ratio data. In step (S808), the store monitoring device (10) can generate a third model by fusion of the first model learned using the first store monitoring data and the second model distributed from the server (20), based on the fusion ratio data.

[0136] In step (S809), the store monitoring device (10) may perform a store analysis on the first store based on the third model, and may generate data on at least one of the following as an analysis result: table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of music played in the store, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay of visiting customers, order status by menu, and interior status. Thereafter, in step (S810), the store monitoring device (10) may provide the analysis result to the store curation server (32).

[0137] The store curation server (32) may, based on the analysis results, obtain data on at least one of table or seat occupancy rate change prediction, seat atmosphere prediction, waiting time prediction, noise change prediction, music to be played in the store, gender prediction of visiting customers, age prediction of visiting customers, behavioral pattern prediction of visiting customers, purpose of visit prediction, menu ranking prediction, recommended menu prediction, sales prediction, inventory prediction, and interior atmosphere prediction for each store, and may present to the customer, through the user terminal (34, 36), stores that match the values ​​that the customer considers important when selecting a store based on these data. That is, the store curation server (32), based on the internal status data of the store, may discover and recommend stores that can actually provide customer satisfaction, rather than allowing the customer to simply select a store close to their current location or a famous franchise store.

[0138] Figure 9 is a flowchart illustrating a store monitoring method according to one embodiment.

[0139] Referring to FIG. 9, a store monitoring method according to an embodiment may include a step of receiving a first model from a server (S901), a step of performing learning based on first store monitoring data collected from a store for the first model (S902), a step of transmitting a first weight update generated by learning for the first model to the server (S903), a step of receiving a second model having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores from the server (S904), a step of performing store analysis for the store based on a model that combines the learned first model and the second model (S905), and a step of providing the store analysis result to a store curation server (S906).

[0140] For more detailed information on the store monitoring method, reference may be made to the description of the embodiments described in this specification, so a redundant description is omitted here.

[0141] Figure 10 is a block diagram illustrating a store analysis system according to one embodiment.

[0142] Referring to FIG. 10, a store analysis system (1) according to one embodiment may include a store monitoring device (10, 12, 14), a store curation server (32), and a user terminal (34, 36). The store monitoring device (10, 12, 14), the store curation server (32), and the user terminal (34, 36) may exchange data with each other via a network (40).

[0143] In this embodiment, a store monitoring device (10, 12) is installed inside a store (A) to monitor the status inside the store (A), and a store monitoring device (14) is installed inside another store (B) to monitor the status inside the other store (B). Analysis results that can be obtained from the store monitoring device (10, 12) installed in the store (A) and analysis results that can be obtained from the monitoring device (14) installed in the store (B) can be transmitted to the store curation server (2) via a network (40).

[0144] The store curation server (32) can present stores that match the values ​​that the user (or customer) considers important when selecting a store to the user through the user terminal (34, 36), based on information about the stores collected through the store monitoring device (10, 12, 14). The information collected through the store monitoring device (10, 12, 14) can be stored and managed in a database accessible to the store curation server (32).

[0145] User terminals (34, 36) can access the store curation server (32) to receive store curation services based on the user's store selection preferences, location, movement, order time zone, order menu, and store usage history. In this way, the store curation service analyzes customer data and analyzes in-store status data captured through store monitoring devices (10, 12, 14). In this way, the customer does not simply select a store near their current location or a well-known franchise store, but rather discovers and recommends stores that can actually provide customer satisfaction. In some embodiments, the user terminals (34, 36) may be computing devices such as smartphones, tablet computers, wearable devices, laptop computers, and desktop computers. The user can utilize the store curation service through an application running on the user terminals (34, 36).

[0146] The network (40) may include a wireless network including a Wi-Fi network, a Bluetooth network, a cellular network, etc., a wired network including a LAN (Local Area Network), and a network in the form of a mixture of a wireless network and a wired network.

[0147] FIG. 11 is a drawing showing an example of a screen displayed on a customer terminal according to one embodiment.

[0148] Referring to Figure 11, a screen displayed on a customer terminal according to one embodiment may include a curated list of stores. For example, the screen may display store information for "Cafe A" and "Cafe B." Specifically, information such as a representative image for each store, sales performance, store location, distance to the store, and expected discount amount may be displayed.

[0149] The analysis results for the store and the predicted future status derived from the curation server (32) may be displayed on the screen in the form of at least one of text, icons, images, and videos. For example, referring to area (A) of the screen, current status data such as "Table occupancy: 36%" and "(approximately 30 dB)" may be displayed in text, and predicted data such as "It's quiet" and "Good for chatting" may be displayed in text. Additionally, to the right of area (A), another predicted data such as "Expected wait of approximately 12 minutes" may be displayed in text.

[0150] Figure 12 is a block diagram illustrating a computing device according to one embodiment.

[0151] Referring to FIG. 12, the store monitoring method and store monitoring device according to the embodiments can be implemented using a computing device (50).

[0152] The computing device (50) may include at least one of a processor (501), a memory (502), a storage device (503), a display device (504) communicating via a bus (509), a network interface device (505) providing a connection to a network (40) for communication with other entities, and an input / output interface device (506) providing a user input interface or a user output interface. Of course, the computing device (50) may additionally include any electronic device necessary to implement the technical concepts described herein, although not illustrated in FIG. 12.

[0153] The processor (501) may be implemented in various types such as an MCU (Micro Controller Unit), an AP (Application Processor), a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an NPU (Neural Processing Unit), etc., and may be any electronic device that executes a program or command stored in a memory (502) or a storage device (503). In particular, the processor (501) may be configured to implement the functions or methods described above with respect to FIGS. 1 to 11.

[0154] The memory (502) and the storage device (503) may include various types of volatile or non-volatile storage media. For example, the memory (502) may include a read-only memory (ROM) or a random access memory (RAM), and the memory (502) may be located inside or outside the processor (501) and may be connected to the processor (501) through various known means. Meanwhile, examples of the storage device (503) include a hard disk drive (HDD) or a solid state drive (SSD), and the scope of the present invention is not limited to the elements listed above for explanation.

[0155] In some embodiments, at least some components or functions of the store monitoring method and the store monitoring device according to the embodiments may be implemented as a program or software running on a computing device (50), and the program or software may be stored on a computer-readable medium. Specifically, the computer-readable medium according to one embodiment may be a program recorded on a computer including a processor (501) that executes a program or instructions stored in a memory (502) or a storage device (503) to execute steps included in the store monitoring method and the store monitoring device according to the embodiments.

[0156] In some embodiments, at least some components or functions of the store monitoring method and store monitoring device according to the embodiments may be implemented using hardware or circuitry of the computing device (50), or may be implemented as separate hardware or circuitry that can be electrically connected to the computing device (50).

[0157] Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by those skilled in the art to which the present invention pertains using the basic concept of the present invention defined in the following claims also fall within the scope of the present invention.

Claims

1. A store monitoring method that performs analysis on a store to monitor real-time status inside the store, predict future status, and provide it to a user terminal. Step of receiving the first model from the server; For the first model, a step of performing learning based on first store monitoring data collected from the store; A step of transmitting a first weight update generated by learning for the first model to the server; A step of distributing a second model having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores from the server; A step of performing store analysis on the store based on a model that combines the learned first model and the second model; and Including a step of providing the store analysis results to the store curation server. How to monitor your store.

2. In paragraph 1, A store monitoring method, wherein the above store analysis results include data on at least one of table or seat arrangement status, table or seat occupancy status, noise status, music played in the store, history of store music playback, gender status of visiting customers, age status of visiting customers, behavior status of visiting customers, length of stay of visiting customers, order status by menu, and interior status.

3. In paragraph 1, The step of performing store analysis on the store based on a model that combines the learned first model and the second model is as follows: A step of generating a third weight update by fusing the first weight update and the second weight update; obtaining a third model having the third weight update; and A store monitoring method comprising a step of performing store analysis on the store using the third model.

4. In paragraph 3, Further comprising a step of reading data representing the ratio set for each store from the memory, The step of obtaining the above third model is: A store monitoring method comprising a step of obtaining the third model having the third weight update calculated according to the following mathematical expression 1, when the above ratio is a:b: (Equation 1) w3= a * w1+ b * w2 Here, w1 is the first weight update, w2 is the second weight update, and w3 is the third weight update.

5. In paragraph 4, The sum of the above a and the above b is 1, a store monitoring method.

6. In paragraph 4, A step of transmitting the third weight update to the server; A step of distributing a fourth model having a fourth weight update generated by synthesizing the third weight update and weight updates provided from other stores from the server; and A store monitoring method further comprising a step of performing store analysis on the store based on a model that combines the third model and the fourth model.

7. In paragraph 6, The step of performing store analysis on the store based on a model that combines the third model and the fourth model is as follows: A step of generating a fifth weight update by fusing the third weight update and the fourth weight update; obtaining a fifth model having the fifth weight update; and A store monitoring method comprising a step of performing store analysis on the store using the fifth model.

8. In paragraph 7, Further comprising a step of reading data representing the ratio set for each store from the memory, The step of obtaining the above fifth model is: When the above ratio is a:b, a step of obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 2; and A store monitoring method comprising the step of obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 3 when the above ratio is changed to c:d: (Equation 2) w5= a * w3+ b * w4 (Equation 3) w5= c * w3+ d * w4 Here, w3 is the third weight update, w4 is the fourth weight update, and w5 is the fifth weight update.

9. In paragraph 1, Further comprising a step of reading data representing the ratio set for each store from the memory, The step of performing an analysis on the store by fusing the learned first model and the second model is as follows: A store monitoring method, comprising the step of performing an analysis on the store according to the following mathematical formula 4, when the above ratio is a:b: (Equation 4) y3= a * y1+ b * y2 Here, y1 is the predicted value of the learned first model, y2 is the predicted value of the second model, and y3 is the predicted value of the third model.

10. In paragraph 9, The sum of the above a and the above b is 1, a store monitoring method.

11. A store monitoring device that performs analysis on the store to monitor real-time status inside the store and predict future status and provide it to the user terminal by executing program code loaded on one or more memory devices through one or more processors. The above program code is executed, Receive the first model from the server, For the first model above, learning is performed based on the first store monitoring data collected from the store, Transmit the first weight update generated by learning the first model to the server, From the above server, a second model having a second weight update generated by synthesizing the first weight update and weight updates provided from other stores is distributed, A store analysis is performed on the store based on a model that combines the first model and the second model learned above, Provides store analysis results to the store curation server. Store monitoring device.

12. In paragraph 11, The above store analysis result is a store monitoring device that includes data on at least one of table or seat arrangement status, table or seat occupancy status, noise status, status of music played in the store, status of store music playback history, status of visiting customer gender, status of visiting customer age group, status of visiting customer behavior, status of visiting customer stay time, status of order by menu, and status of interior design.

13. In paragraph 11, Performing store analysis on the store based on a model that combines the learned first model and the second model, A third weight update is generated by fusing the first weight update and the second weight update, Obtain a third model having the third weight update, A store monitoring device comprising performing store analysis on the store using the third model.

14. In paragraph 13, The above program code is, Read data representing the ratio set for each store from the memory of the above store monitoring device, Obtaining the above third model is: A store monitoring device comprising obtaining the third model having the third weight update calculated according to the following mathematical expression 1, when the above ratio is a:b: (Equation 1) w3= a * w1+ b * w2 Here, w1 is the first weight update, w2 is the second weight update, and w3 is the third weight update.

15. In paragraph 14, The sum of the above a and the above b is 1, a store monitoring device.

16. In paragraph 14, The above program code is, Send the third weight update to the server, From the above server, a fourth model having a fourth weight update generated by synthesizing the third weight update and weight updates provided from other stores is distributed, A store monitoring device that performs store analysis on the store based on a model that combines the third model and the fourth model.

17. In paragraph 16, Performing store analysis on the store based on a model that combines the third model and the fourth model, The third weight update and the fourth weight update are fused to generate the fifth weight update, Obtain a fifth model having the fifth weight update, A store monitoring device comprising performing store analysis on the store using the fifth model.

18. In paragraph 17, The above program code is, Read data representing the ratio set for each store from the memory of the above store monitoring device, Obtaining the above fifth model is as follows: If the above ratio is a:b, obtain the fifth model having the fifth weight update calculated according to the following mathematical expression 2, A store monitoring device comprising obtaining the fifth model having the fifth weight update calculated according to the following mathematical expression 3 when the above ratio is changed to c:d: (Equation 2) w5= a * w3+ b * w4 (Equation 3) w5= c * w3+ d * w4 Here, w3 is the third weight update, w4 is the fourth weight update, and w5 is the fifth weight update.

19. In paragraph 11, The above program code is, Read data representing the ratio set for each store from the memory of the above store monitoring device, Performing an analysis on the store by fusing the learned first model and the second model, A store monitoring device comprising performing an analysis on the store according to the following mathematical formula 4, when the above ratio is a:b: (Equation 4) y3= a * y1+ b * y2 Here, y1 is the predicted value of the learned first model, y2 is the predicted value of the second model, and y3 is the predicted value of the third model.

20. In paragraph 19, The sum of the above a and the above b is 1, a store monitoring device.

Images