A Smart Retail Method and System Based on Facial Recognition

Abstract

This invention relates to the field of intelligent retail management technology. Specifically, it relates to an intelligent retail method and system based on facial recognition. The system includes: a movement trajectory analysis module, a region division module, a product association analysis module, and a product reorganization module. The movement trajectory analysis module analyzes and judges abnormal behavior trajectory data by tracking the movement trajectory of a shopping basket, and generates a behavior heatmap based on the discrete movement trajectory data. The region division module divides the retail space into three regions: hot, warm, and cold. The product association analysis module analyzes the association values ​​of association combinations and constructs a product association graph based on the association combinations and corresponding association values. The product reorganization module reorganizes products in the hot and cold regions of the retail space through strongly associated product groups. This invention significantly improves retail operational efficiency and avoids overcrowding by analyzing the association degree of high-frequency and low-frequency interactive products in the retail space and scientifically reorganizing the products.

Description

Technical Field

[0001] This invention relates to the field of intelligent retail management technology, specifically to an intelligent retail method and system based on facial recognition. Background Technology

[0002] Traditional retail stores face multiple challenges, including changing consumer behavior, the impact of e-commerce, and rising operating costs. To address these challenges, smart retail stores have emerged, integrating advanced technologies such as the Internet of Things (IoT), big data, artificial intelligence (AI), and machine learning. This transforms traditional offline buying and selling activities into smart retail based on internet, IoT, and AI technologies. With the rapid development of technology, the retail industry is undergoing unprecedented transformation. This involves improving the shopping experience, optimizing operational efficiency, and enhancing data analytics capabilities to meet the growing personalized needs of consumers and improve retailers' competitiveness.

[0003] User tracking technology, as a core component of smart retail stores, provides retailers with in-depth insights by collecting and analyzing data such as consumers' movement paths, dwell time, and interactive behaviors in real time. This supports precision marketing, optimizes merchandise display, and improves customer satisfaction and operational efficiency.

[0004] With existing technology, user tracking technology can only show the physical movement of consumers in the store, and cannot directly reflect their purchase intentions or actual interest in the products, nor can it reveal the correlation between consumers and different products. Relying solely on user tracking technology for analysis will lead to errors in the behavior heatmap of retail stores, thereby reducing customer satisfaction and operational efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a smart retail method and system based on facial recognition to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, the present invention provides the following technical solution: a smart retail method based on facial recognition, applied to smart physical retail spaces;

[0007] Based on the RFID tags on the shopping basket, the movement trajectory of the shopping basket is analyzed, and a unique interaction library is generated based on the movement trajectory of the shopping basket to identify high-frequency interaction hotspots. Based on the data matching model, the purchased products are matched with the products in the high-frequency interaction hotspot areas. Based on the matching results, abnormal behavior trajectories are identified and data is removed. The density of discrete movement trajectory data is estimated to generate a behavior heatmap.

[0008] Based on behavioral heatmaps, retail spaces are divided into three zones: hot, warm, and cold. These zones correspond to retail areas with progressively decreasing foot traffic. The hot zone is the retail area with the highest customer dwell density and significant interaction frequency; the warm zone is the retail area with moderate foot traffic and transitional dwell time; and the cold zone is the retail area with significantly lower customer visit frequency and dwell time.

[0009] Based on user purchase data, a correlation analysis is performed on products in the hot and cold areas of the retail space to generate a product correlation diagram.

[0010] Based on the product association diagram, we analyze the strongly related product groups in the retail space. By reorganizing the products in the hot and cold areas of the retail space through the strongly related product groups, we can promote the sales of products in the cold area while avoiding stampedes caused by crowds.

[0011] Furthermore, specific methods for generating continuous behavioral heatmaps based on the shopping basket movement trajectory include:

[0012] The retail space is mapped onto a coordinate system, and multiple movement trajectories are tracked through RFID tags on shopping baskets to generate shopping basket movement trajectories and record the trajectory information of different users in the retail space: ut(x, y), where ut(x, y) represents the horizontal and vertical coordinate position information (x, y) of shopping basket u at the t-th time node after entering the retail space; u = 1, 2, 3...U, where U is the number of shopping baskets, and the horizontal and vertical coordinate points represent the position information of the shelves in the retail space;

[0013] By using a spatial trajectory density visualization method, the density of discrete shopping basket movement trajectory data is estimated to generate a behavior heatmap. The frequency of shopping basket movement trajectory coverage is positively correlated with the color brightness of the behavior heatmap; specifically, the more times a retail space unit is covered by the trajectory, the higher the corresponding heat value, and the darker the hue in the color scale mapping.

[0014] To avoid errors in the color gradation display of the behavior heatmap due to abnormal behavior, the discrete shopping basket movement trajectory data does not include abnormal behavior trajectory data. The determination of abnormal behavior trajectory data includes:

[0015] Obtain the movement trajectory u'-t(x, y) of any shopping basket. When the position information of the shopping basket remains unchanged at consecutive time points, it is initially determined that the user is interacting with the product. Record the current position information of the shopping basket and generate a dedicated interaction library. Use an indicator function to count the number of times the same position information appears in the dedicated interaction library, and record the position information with the highest occurrence as the high-frequency interaction hotspot. When the maximum number of counts exceeds the total number of position information recorded in the movement trajectory u'-t(x, y) of the shopping basket, it is determined that the high-frequency interaction hotspot indicates potential user purchase behavior.

[0016] The data matching model determines whether the items purchased by the user in the shopping basket u' belong to the high-frequency interaction hotspot area. When the items purchased by user u' do not belong to the high-frequency interaction hotspot area, the movement trajectory of any shopping basket is determined to be abnormal behavior trajectory data; otherwise, the movement trajectory is determined to be normal. The data matching model scans the items during the user's payment process and matches the item data of high-frequency interaction hotspots with the item location database.

[0017] Furthermore, specific methods for conducting correlation analysis on goods in the hot and cold zones of a retail space include:

[0018] The products in the hot and cold zones of the retail space are sorted and generated into sets R and L, respectively, where R = {ri} and L = {lj}, ri is the i-th product in the hot zone and lj is the j-th product in the cold zone. Both ri and lj are the product numbers in the hot and cold zones of the retail space; i = 1, 2, 3...I, j = 1, 2, 3...J, and I and J are both constants.

[0019] For any user's historical purchase data in a retail store, products belonging to sets R and L that appear simultaneously are filtered to generate associated combinations (ri', lj'). Collaborative filtering is used to count the number of times the associated combinations appear in the historical purchase data to obtain M(ri', lj'). The association value of the associated combinations (ri', lj') of hot and cold zones in the retail space is proportional to M(ri', lj'), where i'∈{1, 2, 3...I}, j'∈{1, 2, 3...J}.

[0020] A product association graph is constructed using the association combinations and their corresponding association values, where products in the association combinations serve as nodes in the graph, and association values ​​serve as weights for the product connection links.

[0021] Furthermore, specific methods for reorganizing merchandise in the hot and cold zones of a retail space include:

[0022] Select any product ri' in the hot zone, and iterate through products lj' in set L that have a connection link with product ri'. Determine the association value between products ri' and lj' using the product association graph. This includes: when there is only one connection link between products ri' and lj', the association value between the two products is the weight of the link; when there are multiple connection links between products ri' and lj', a weighted summation method is used to calculate the association value between the two products. Specifically, if there is a set of link products between the two products, the corresponding link weight sequence is V = {v1, v2, v3...vN}, where vN is the weight of the last connection link in the corresponding link weight sequence, and N is the number of connection links between products ri' and lj'. Then the association value between products ri' and lj' is...

[0023] When the correlation value between products ri' and lj' exceeds a preset correlation threshold, products ri' and lj' are set as a strongly correlated product group. By analyzing the replacement scores of hot zone products, the replacement method for the strongly correlated product group is determined.

[0024] Based on the product association diagram, determine the number of products ri' and lj' that are associated with sets L and R, respectively, and the corresponding association values. The replacement scores of the two products are calculated by the weighted product of the number of associated products, the sum of the association values ​​of the associated products, and the corresponding hot and cold zone areas of the two products.

[0025] When the replacement score of a product in a cold zone exceeds that of a product in a hot zone, product lj' is moved to the hot zone corresponding to ri', potentially stimulating related purchase potential of product lj' within a unit area. When the replacement score of a product in a cold zone is lower than that of a product in a hot zone, product ri' is copied to the cold zone corresponding to lj', potentially increasing the exposure intensity of related products in the cold zone ri' within a unit area. Strongly related product groups in the retail store are traversed, and products in the hot and cold zones of the retail space are reorganized based on these groups.

[0026] Furthermore, specific methods for dividing retail spaces into zones based on behavioral heatmaps include:

[0027] The behavior heatmap is initially divided into regions using an edge algorithm to determine the boundaries of different regions. The overall activity level of different regions in the retail space is calculated by weighting and accumulating the brightness values ​​of each pixel in the region. The higher the pixel brightness value, the lower the overall activity level of different regions in the retail space. All regions are sorted from high to low according to their overall activity level, and the different regions in the retail space are divided into three regions: hot, warm, and cold using the quantile threshold method.

[0028] The pixel brightness value is obtained by decoding the pixels of the behavior heatmap using a color space conversion algorithm, extracting the 8-bit quantization value (0-255) of the RGB (red, green, blue) channels of each pixel, and calculating the lightness and darkness of the RGB values ​​by quantizing the lightness component in the HSV color model.

[0029] A smart retail system based on facial recognition, the smart retail system comprising: a movement trajectory analysis module, a region division module, a product association analysis module, and a product reorganization module;

[0030] The movement trajectory analysis module is used to track the movement trajectory of the shopping basket using RFID tags. Based on whether the location information of the shopping basket changes at consecutive time points, it preliminarily determines whether the user has interacted with the products. When the user interacts with the products, it records the current location information of the shopping basket and generates a dedicated interaction database. It uses an indicator function to count the frequency of occurrences of the same location information in the dedicated interaction database, recording the location information with the highest frequency as a high-frequency interaction hotspot. Based on a data matching model, for products whose purchased items do not belong to the high-frequency interaction hotspot area, the movement trajectory is determined to be abnormal behavior trajectory data. Using a spatial trajectory density visualization method, the density of the discrete shopping basket movement trajectory data (non-abnormal behavior trajectory data) is estimated to generate a behavior heatmap.

[0031] The region division module is used to initially divide the behavior heatmap using an edge algorithm, determine the boundaries of different regions, calculate the comprehensive activity level of different regions of the retail space by weighted summation of pixel brightness values ​​in each region, sort all regions from high to low comprehensive activity level, and divide the different regions of the retail space into three regions: hot, warm, and cold using the quantile threshold method.

[0032] The product association analysis module is used to generate sets by sorting products in the hot and cold zones of the retail space, and to filter products that belong to both hot and cold zones and appear simultaneously in the historical purchase data of any user in the retail store to generate association combinations; to count the number of times the association combinations appear in the historical purchase data through collaborative filtering, wherein the association value of two products in the association combination of hot and cold zone products in the retail space is proportional to the number of times the association combination appears; and to construct a product association graph through the association combinations and the corresponding association values, wherein the products in the association combination are nodes in the graph, and the association value is the weight of the product connection link;

[0033] The product reorganization module is used to set up a strongly associated product group when the correlation value between any products in the hot and cold zones exceeds a preset correlation threshold, to traverse the strongly associated product groups in the retail store, and to reorganize the products in the hot and cold zones of the retail space based on the strongly associated product groups.

[0034] Furthermore, the movement trajectory analysis module includes a movement trajectory tracking unit, an abnormal behavior trajectory data filtering unit, and a behavior heatmap generation unit;

[0035] The movement trajectory tracking unit is used to achieve multi-track tracking through RFID tags and generate movement trajectories. The abnormal behavior trajectory data filtering unit is used to preliminarily determine whether the user has interacted with the product based on whether the location information of the shopping basket changes at consecutive time points. When the user interacts with the product, the current location information of the shopping basket is recorded to generate a dedicated interaction library. The number of times the same location information appears in the dedicated interaction library is counted using an indicator function, and the location information with the highest number of occurrences is recorded as a high-frequency interaction hotspot. Based on the data matching model, the shopping basket movement trajectory is determined to be abnormal behavior trajectory data for products that do not belong to the high-frequency interaction hotspot area. The behavior heatmap generation unit is used to perform density estimation on the discrete shopping basket movement trajectory data of non-abnormal behavior trajectory data through a spatial trajectory density visualization method to generate a behavior heatmap.

[0036] Furthermore, the product association analysis module includes an association combination analysis unit, a combination association value analysis unit, and a product association graph generation unit;

[0037] The association combination analysis unit is used to generate association combinations by filtering products that belong to sets R and L and appear simultaneously in the historical purchase data of any user in the retail store; the combination association value analysis unit is used to count the number of times the association combination appears in the historical purchase data through collaborative filtering, and the association value of two products in the association combination of hot and cold zones of the retail space is proportional to the number of times the association combination appears; the product association graph generation unit is used to construct a product association graph through the association combination and the corresponding association value, wherein the products in the association combination are nodes in the graph, and the association value is the weight of the product connection link.

[0038] Furthermore, the product restructuring module includes a product correlation value analysis unit, a product replacement evaluation unit, and a product restructuring unit;

[0039] The product association value analysis unit is used to determine the association value between any two products in the hot and cold zones through the product association graph, including: when there is only one connection link between two products, the association value of the two products is the weight of the link; when there are multiple connection links between two products, the association value of the two products is calculated by weighted summation method; the product replacement evaluation unit is used to calculate the replacement score of two products by the number of associated products corresponding to any two products in the hot and cold zones, the sum of the association values ​​of the associated products, and the weighted product of the areas of the hot and cold zones corresponding to the two products; when the replacement score of the cold zone product exceeds the replacement score of the hot zone product, the cold zone product is promoted to the hot zone corresponding to another product; when the replacement score of the cold zone product is lower than the replacement score of the hot zone product, the hot zone product is copied to the cold zone corresponding to another product; the product reorganization unit is used to traverse the strongly associated product groups in the retail store and reorganize the products in the hot and cold areas of the retail space according to the strongly associated product groups.

[0040] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: The present invention analyzes abnormal behavior trajectory data in the movement trajectory through a data matching model to generate a behavior heat map; the retail space is divided into areas based on the behavior heat map, and the products are scientifically reorganized by the correlation values ​​of products between the divided areas, thereby avoiding crowding and significantly improving retail operation efficiency and customer satisfaction. Attached Figure Description

[0041] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0042] Figure 1 This is a flowchart of the steps of a scene design method based on virtual reality according to the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Please see Figure 1 The present invention provides a technical solution: The purpose of the present invention is to provide a smart retail method and system based on face recognition to solve the problems mentioned in the background art.

[0045] To address the aforementioned technical problems, this invention provides the following technical solution: This solution strictly adheres to relevant laws and regulations regarding personal information protection during implementation, and all account login processes involving facial recognition technology have obtained explicit user authorization. The specific process is as follows:

[0046] User consent required: When a user initiates an account login request via mobile or self-service device, the system will forcefully display the user agreement and privacy policy page, clearly informing the user of the purpose of using facial recognition technology, the scope of data collection, and the storage period. The user must actively check the "I have read and agree" option before proceeding to the next step. This step employs a double confirmation mechanism, requiring the user to swipe to verify or enter a verification code to confirm the authenticity of the operation and avoid accidental authorization.

[0047] Conditions for enabling facial recognition technology: Users select "facial login" as the authentication method to log in to their accounts.

[0048] After completing facial recognition login, users can receive a shopping basket equipped with an RFID (Radio Frequency Identification) tag. This tag serves as a unique identifier and records information such as the movement trajectory of the shopping basket, the duration of stay, and the frequency of area access during the user's shopping process in real time.

[0049] A smart retail method based on facial recognition, applied to smart physical retail spaces;

[0050] Based on the RFID tags on the shopping basket, the movement trajectory of the shopping basket is analyzed, and a unique interaction library is generated based on the movement trajectory of the shopping basket to identify high-frequency interaction hotspots. Based on the data matching model, the purchased products are matched with the products in the high-frequency interaction hotspot areas. Based on the matching results, abnormal behavior trajectories are identified and data is removed. The density of discrete movement trajectory data is estimated to generate a behavior heatmap.

[0051] Based on behavioral heatmaps, retail spaces are divided into three zones: hot, warm, and cold. These zones correspond to retail areas with progressively decreasing foot traffic. The hot zone is the retail area with the highest customer dwell density and significant interaction frequency; the warm zone is the retail area with moderate foot traffic and transitional dwell time; and the cold zone is the retail area with significantly lower customer visit frequency and dwell time.

[0052] Based on user purchase data, a correlation analysis is performed on products in the hot and cold areas of the retail space to generate a product correlation diagram.

[0053] Based on the product association diagram, we analyze the strongly related product groups in the retail space. By reorganizing the products in the hot and cold areas of the retail space through the strongly related product groups, we can promote the sales of products in the cold area while avoiding stampedes caused by crowds.

[0054] Furthermore, specific methods for generating continuous behavioral heatmaps based on the shopping basket movement trajectory include:

[0055] The retail space is mapped onto a coordinate system, and multiple movement trajectories are tracked through RFID tags on shopping baskets to generate shopping basket movement trajectories and record the trajectory information of different users in the retail space: ut(x, y), where ut(x, y) represents the horizontal and vertical coordinate position information (x, y) of shopping basket u at the t-th time node after entering the retail space; u = 1, 2, 3...U, where U is the number of shopping baskets, and the horizontal and vertical coordinate points represent the position information of the shelves in the retail space;

[0056] By using a spatial trajectory density visualization method, the density of discrete shopping basket movement trajectory data is estimated to generate a behavior heatmap. The frequency of shopping basket movement trajectory coverage is positively correlated with the color brightness of the behavior heatmap; specifically, the more times a retail space unit is covered by the trajectory, the higher the corresponding heat value, and the darker the hue in the color scale mapping.

[0057] To avoid errors in the color gradation display of the behavior heatmap due to abnormal behavior, the discrete shopping basket movement trajectory data does not include abnormal behavior trajectory data. The determination of abnormal behavior trajectory data includes:

[0058] Obtain the movement trajectory u'-t(x, y) of any shopping basket. When the position information of the shopping basket remains unchanged at consecutive time points, it is initially determined that the user is interacting with the product. Record the current position information of the shopping basket and generate a dedicated interaction library. Use an indicator function to count the number of times the same position information appears in the dedicated interaction library, and record the position information with the highest occurrence as the high-frequency interaction hotspot. When the maximum number of counts exceeds the total number of position information recorded in the movement trajectory u'-t(x, y) of the shopping basket, it is determined that the high-frequency interaction hotspot indicates potential user purchase behavior.

[0059] The data matching model determines whether the items purchased by the user in the shopping basket u' belong to the high-frequency interaction hotspot area. When the items purchased by user u' do not belong to the high-frequency interaction hotspot area, the movement trajectory of any shopping basket is determined to be abnormal behavior trajectory data; otherwise, the movement trajectory is determined to be normal. The data matching model scans the items during the user's payment process and matches the item data of high-frequency interaction hotspots with the item location database.

[0060] Furthermore, specific methods for conducting correlation analysis on goods in the hot and cold zones of a retail space include:

[0061] The products in the hot and cold zones of the retail space are sorted and generated into sets R and L, respectively, where R = {ri} and L = {lj}, ri is the i-th product in the hot zone and lj is the j-th product in the cold zone. Both ri and lj are the product numbers in the hot and cold zones of the retail space; i = 1, 2, 3...I, j = 1, 2, 3...J, and I and J are both constants.

[0062] For any user's historical purchase data in a retail store, products belonging to sets R and L that appear simultaneously are filtered to generate associated combinations (ri', lj'). Collaborative filtering is used to count the number of times the associated combinations appear in the historical purchase data to obtain M(ri', lj'). The association value of the associated combinations (ri', lj') of hot and cold zones in the retail space is proportional to M(ri', lj'), where i'∈{1, 2, 3...I}, j'∈{1, 2, 3...J}.

[0063] A product association graph is constructed using the association combinations and their corresponding association values, where products in the association combinations serve as nodes in the graph, and association values ​​serve as weights for the product connection links.

[0064] Furthermore, specific methods for reorganizing merchandise in the hot and cold zones of a retail space include:

[0065] Select any product ri' in the hot zone, and iterate through products lj' in set L that have a connection link with product ri'. Determine the association value between products ri' and lj' using the product association graph. This includes: when there is only one connection link between products ri' and lj', the association value between the two products is the weight of the link; when there are multiple connection links between products ri' and lj', a weighted summation method is used to calculate the association value between the two products. Specifically, if there is a set of link products between the two products, the corresponding link weight sequence is V = {v1, v2, v3...vN}, where vN is the weight of the last connection link in the corresponding link weight sequence, and N is the number of connection links between products ri' and lj'. Then the association value between products ri' and lj' is...

[0066] When the correlation value between products ri' and lj' exceeds a preset correlation threshold, products ri' and lj' are set as a strongly correlated product group. By analyzing the replacement scores of hot zone products, the replacement method for the strongly correlated product group is determined.

[0067] Based on the product association diagram, determine the number of products ri' and lj' that are associated with sets L and R, respectively, and the corresponding association values. The replacement scores of the two products are calculated by the weighted product of the number of associated products, the sum of the association values ​​of the associated products, and the corresponding hot and cold zone areas of the two products.

[0068] When the replacement score of a product in a cold zone exceeds that of a product in a hot zone, product lj' is moved to the hot zone corresponding to ri', potentially stimulating related purchase potential of product lj' within a unit area. When the replacement score of a product in a cold zone is lower than that of a product in a hot zone, product ri' is copied to the cold zone corresponding to lj', potentially increasing the exposure intensity of related products in the cold zone ri' within a unit area. Strongly related product groups in the retail store are traversed, and products in the hot and cold zones of the retail space are reorganized based on these groups.

[0069] Furthermore, specific methods for dividing retail spaces into zones based on behavioral heatmaps include:

[0070] The behavior heatmap is initially divided into regions using an edge algorithm to determine the boundaries of different regions. The overall activity level of different regions in the retail space is calculated by weighting and accumulating the brightness values ​​of each pixel in the region. The higher the pixel brightness value, the lower the overall activity level of different regions in the retail space. All regions are sorted from high to low according to their overall activity level, and the different regions in the retail space are divided into three regions: hot, warm, and cold using the quantile threshold method.

[0071] The pixel brightness value is obtained by decoding the pixels of the behavior heatmap using a color space conversion algorithm, extracting the 8-bit quantization value (0-255) of the RGB (red, green, blue) channels of each pixel, and calculating the lightness and darkness of the RGB values ​​by quantizing the lightness component in the HSV color model.

[0072] A smart retail system based on facial recognition, the smart retail system comprising: a movement trajectory analysis module, a region division module, a product association analysis module, and a product reorganization module;

[0073] The movement trajectory analysis module is used to track the movement trajectory of the shopping basket using RFID tags. Based on whether the location information of the shopping basket changes at consecutive time points, it preliminarily determines whether the user has interacted with the products. When the user interacts with the products, it records the current location information of the shopping basket and generates a dedicated interaction database. It uses an indicator function to count the frequency of occurrences of the same location information in the dedicated interaction database, recording the location information with the highest frequency as a high-frequency interaction hotspot. Based on a data matching model, for products whose purchased items do not belong to the high-frequency interaction hotspot area, the movement trajectory is determined to be abnormal behavior trajectory data. Using a spatial trajectory density visualization method, the density of the discrete shopping basket movement trajectory data (non-abnormal behavior trajectory data) is estimated to generate a behavior heatmap.

[0074] The region division module is used to initially divide the behavior heatmap using an edge algorithm, determine the boundaries of different regions, calculate the comprehensive activity level of different regions of the retail space by weighted summation of pixel brightness values ​​in each region, sort all regions from high to low comprehensive activity level, and divide the different regions of the retail space into three regions: hot, warm, and cold using the quantile threshold method.

[0075] The product association analysis module is used to generate sets by sorting products in the hot and cold zones of the retail space, and to filter products that belong to both hot and cold zones and appear simultaneously in the historical purchase data of any user in the retail store to generate association combinations; to count the number of times the association combinations appear in the historical purchase data through collaborative filtering, wherein the association value of two products in the association combination of hot and cold zone products in the retail space is proportional to the number of times the association combination appears; and to construct a product association graph through the association combinations and the corresponding association values, wherein the products in the association combination are nodes in the graph, and the association value is the weight of the product connection link;

[0076] The product reorganization module is used to set up a strongly associated product group when the correlation value between any products in the hot and cold zones exceeds a preset correlation threshold, to traverse the strongly associated product groups in the retail store, and to reorganize the products in the hot and cold zones of the retail space based on the strongly associated product groups.

[0077] Furthermore, the movement trajectory analysis module includes a movement trajectory tracking unit, an abnormal behavior trajectory data filtering unit, and a behavior heatmap generation unit;

[0078] The movement trajectory tracking unit is used to achieve multi-track tracking through RFID tags and generate movement trajectories. The abnormal behavior trajectory data filtering unit is used to preliminarily determine whether the user has interacted with the product based on whether the location information of the shopping basket changes at consecutive time points. When the user interacts with the product, the current location information of the shopping basket is recorded to generate a dedicated interaction library. The number of times the same location information appears in the dedicated interaction library is counted using an indicator function, and the location information with the highest number of occurrences is recorded as a high-frequency interaction hotspot. Based on the data matching model, the shopping basket movement trajectory is determined to be abnormal behavior trajectory data for products that do not belong to the high-frequency interaction hotspot area. The behavior heatmap generation unit is used to perform density estimation on the discrete shopping basket movement trajectory data of non-abnormal behavior trajectory data through a spatial trajectory density visualization method to generate a behavior heatmap.

[0079] Furthermore, the product association analysis module includes an association combination analysis unit, a combination association value analysis unit, and a product association graph generation unit;

[0080] The association combination analysis unit is used to generate association combinations by filtering products that belong to sets R and L and appear simultaneously in the historical purchase data of any user in the retail store; the combination association value analysis unit is used to count the number of times the association combination appears in the historical purchase data through collaborative filtering, and the association value of two products in the association combination of hot and cold zones of the retail space is proportional to the number of times the association combination appears; the product association graph generation unit is used to construct a product association graph through the association combination and the corresponding association value, wherein the products in the association combination are nodes in the graph, and the association value is the weight of the product connection link.

[0081] Furthermore, the product restructuring module includes a product correlation value analysis unit, a product replacement evaluation unit, and a product restructuring unit;

[0082] The product association value analysis unit is used to determine the association value between any two products in the hot and cold zones through the product association graph, including: when there is only one connection link between two products, the association value of the two products is the weight of the link; when there are multiple connection links between two products, the association value of the two products is calculated by weighted summation method; the product replacement evaluation unit is used to calculate the replacement score of two products by the number of associated products corresponding to any two products in the hot and cold zones, the sum of the association values ​​of the associated products, and the weighted product of the areas of the hot and cold zones corresponding to the two products; when the replacement score of the cold zone product exceeds the replacement score of the hot zone product, the cold zone product is promoted to the hot zone corresponding to another product; when the replacement score of the cold zone product is lower than the replacement score of the hot zone product, the hot zone product is copied to the cold zone corresponding to another product; the product reorganization unit is used to traverse the strongly associated product groups in the retail store and reorganize the products in the hot and cold areas of the retail space according to the strongly associated product groups.

[0083] In this embodiment:

[0084] Select any product ri' in the hot zone, and iterate through products lj' in set L that have a connection link with product ri'. Determine the association value between product ri' and lj' through the product association graph.

[0085] When there are 7 connection links between products ri' and lj', the association value between the two products is calculated using a weighted summation method. Based on the existence of a set of linked products between the two products, the corresponding link weight sequence is V = {1.4, 3.34, 1.65...}. Therefore, the association value between products ri' and lj' is (∑N n=1vn) / N = (∑N n = 1vn) / 7 = 0.76. The correlation value between products ri' and lj' exceeds the preset correlation threshold of 0.75. Products ri' and lj' are set as a strongly correlated product group. By analyzing the replacement scores of hot zone products, the replacement method of the strongly correlated product group is determined: Based on the product correlation diagram, the number of products associated with products ri' and lj' in sets L and R and the corresponding correlation values ​​are determined. The replacement scores of the two products are calculated by the number of associated products, the sum of the correlation values ​​of associated products, and the weighted product of the hot and cold zone areas corresponding to the two products. The replacement score of product ri' is 5 × (0.75 + 0.97 + ...) × 3 = 34.7, and the replacement score of product lj' is 7 × (0.35 + 0.67 + ...) × 10 = 108.

[0086] The replacement score of the cold zone product (108) exceeds that of the hot zone product (34.7), which elevates product lj' to the hot zone corresponding to ri', potentially stimulating related purchase potential of product lj' within a unit area.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0088] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart retail method based on facial recognition, applied to smart physical retail spaces, requiring users to log in to their accounts via facial recognition to retrieve shopping baskets, characterized in that: Based on the RFID tags on the shopping basket, the movement trajectory of the shopping basket is analyzed, and a unique interaction library is generated based on the movement trajectory of the shopping basket to identify high-frequency interaction hotspots. Based on the data matching model, the purchased products are matched with the products in the high-frequency interaction hotspot areas. Based on the matching results, abnormal behavior trajectories are identified and data is removed. The density of discrete movement trajectory data is estimated to generate a behavior heatmap. Based on behavioral heatmaps, the retail space is divided into three zones: hot, warm, and cold. The hot, warm, and cold zones correspond to retail areas with progressively decreasing foot traffic. Based on purchase data, a correlation analysis is performed on products in the hot and cold areas of the retail space to generate a product correlation diagram. Based on the product association diagram, we analyze the strongly related product groups in the retail space, and then reorganize the products in the hot and cold areas of the retail space through the strongly related product groups; Specific methods for conducting correlation analysis on goods in hot and cold zones of a retail space include: The products in the hot and cold areas of the retail space are sorted and generated into sets R and L, respectively, where R = {ri} and L = {lj}, ri is the i-th product in the hot area of ​​the retail space, and lj is the j-th product in the cold area of ​​the retail space; i = 1, 2, 3...I, j = 1, 2, 3...J; For any user's historical purchase data in a retail store, products belonging to sets R and L that appear simultaneously are filtered to generate associated combinations (ri', lj'). Collaborative filtering is used to count the number of times the associated combinations appear in the historical purchase data to obtain M(ri', lj'). The association value of the associated combinations (ri', lj') of hot and cold zones in the retail space is proportional to M(ri', lj'), where i'∈{1, 2, 3...I}, j'∈{1, 2, 3...J}. A product association graph is constructed using the association combinations and their corresponding association values, where products in the association combinations are nodes in the graph, and association values ​​are weights for the product connection links. The specific method for reorganizing goods in the hot and cold zones of a retail space includes: selecting any product ri' in the hot zone, traversing the connections between product ri' and products lj' in set L, and determining the association value between products ri' and lj' through a product association graph. This includes: when there is only one connection between products ri' and lj', the association value between the two products is the weight of the connection; when there are multiple connections between products ri' and lj', the association value between the two products is calculated using a weighted summation method, which is: if there is a set of products with a connection, the corresponding link weight sequence is V={v1, v2, v3...vN}, where vN is the weight of the last connection in the corresponding link weight sequence, and N is the number of connections between products ri' and lj', then the association value between products ri' and lj' is (∑N n=1vn) / N; When the correlation value between products ri' and lj' exceeds a preset correlation threshold, products ri' and lj' are set as a strongly correlated product group. By analyzing the replacement scores of hot zone products, the replacement method for the strongly correlated product group is determined. Based on the product association diagram, determine the number of products ri' and lj' that are associated with sets L and R, respectively, and the corresponding association values. The replacement scores of the two products are calculated by the weighted product of the number of associated products, the sum of the association values ​​of the associated products, and the corresponding hot and cold zone areas of the two products. When the replacement score of a cold zone product exceeds that of a hot zone product, product lj' is promoted to the hot zone corresponding to ri'; when the replacement score of a cold zone product is lower than that of a hot zone product, product ri' is copied to the cold zone corresponding to lj'; the strongly related product groups in the retail store are traversed, and the products in the hot and cold zones of the retail space are reorganized based on the strongly related product groups.

2. The intelligent retail method based on face recognition according to claim 1, characterized in that: Specific methods for generating continuous behavioral heatmaps based on the movement trajectory of a shopping basket include: The retail space is mapped onto a coordinate system, and multiple movement trajectories are tracked through RFID tags on shopping baskets to generate the shopping basket movement trajectory: ut(x, y), where ut(x, y) represents the horizontal and vertical coordinate position information (x, y) of shopping basket u at the t-th time node after entering the retail space; u = 1, 2, 3...U, where U is the number of shopping baskets, and the horizontal and vertical coordinate points represent the position information of the shelves in the retail space; By using a spatial trajectory density visualization method, the density of discrete shopping basket movement trajectory data is estimated to generate a behavior heatmap. The frequency of shopping basket movement trajectory coverage is positively correlated with the color brightness of the behavior heatmap; specifically, the more times a retail space unit is covered by the trajectory, the higher the corresponding heat value, and the darker the hue in the color scale mapping. The discrete shopping basket movement trajectory data does not include abnormal behavior trajectory data, wherein the determination of abnormal behavior trajectory data includes: Obtain the movement trajectory u'-t(x, y) of any shopping basket. When the position information of the shopping basket remains unchanged at consecutive time points, it is initially determined that the user is interacting with the product. Record the current position information of the shopping basket and generate a dedicated interaction library. Use an indicator function to count the number of times the same position information appears in the dedicated interaction library, and record the position information with the highest occurrence as the high-frequency interaction hotspot. When the maximum number of counts exceeds the total number of position information recorded in the movement trajectory u'-t(x, y) of the shopping basket, it is determined that the high-frequency interaction hotspot indicates potential user purchase behavior. The data matching model determines whether the items purchased by the user in the shopping basket u' belong to the high-frequency interaction hotspot area. When the purchased items do not belong to the high-frequency interaction hotspot area, the movement trajectory of any shopping basket is determined to be abnormal behavior trajectory data. The data matching model scans the items during the user's payment process and matches the item data of high-frequency interaction hotspots with the item location database.

3. The intelligent retail method based on face recognition according to claim 2, characterized in that: Specific methods for dividing retail space into zones based on behavioral heatmaps include: The behavior heatmap is initially divided into regions using an edge algorithm to determine the boundaries of different regions. The overall activity level of different regions in the retail space is calculated by weighting and accumulating the brightness values ​​of each pixel in the region. All regions are sorted from high to low according to their overall activity level, and the different regions in the retail space are divided into three regions: hot, warm, and cold using the quantile threshold method. The pixel brightness value is obtained by decoding the pixels of the behavior heatmap using a color space conversion algorithm, and extracting the 8-bit quantization value of the RGB channel of each pixel. The RGB color values ​​are calculated by quantifying the lightness and darkness variations using the lightness component in the HSV color model.

Images