Intelligent commodity recommendation and display system

Abstract

The invention relates to the technical field of commodity sales optimization, and discloses an intelligent commodity recommendation and display system, which comprises an identification module, and is characterized in that the identification module records shoppers and articles through a first camera and a second camera, and generates shopper data and article data; and the analysis module calculates the purchase demand and purchase tendency of the shopper according to the data of the shopper so as to output an adjustment instruction. High-intention commodities are actively pushed to the front of customers, searching and hesitation time is shortened, the shopping experience of the customers is optimized, the commodities are automatically close to and directly face the sight line, light guide is assisted, taking is more convenient and visual, especially convenience is brought to old people, children or customers carrying objects, the satisfaction degree is improved, intervention is based on real behaviors instead of guessing, and the method is suitable for large-scale popularization and application. And invalid sales promotion is avoided, resources are concentrated for customers with real demands, the marketing efficiency is improved, tasks such as commodity display optimization, sales promotion display adjustment and the like are automatically completed, and the workload of frequently and manually arranging goods shelves by shop assistants is reduced.

Description

Technical Field

[0001] This invention relates to the field of product sales optimization technology, specifically to an intelligent product recommendation and display system. Background Technology

[0002] Merchandise display principles are standardized guidelines for product display and placement established by supermarkets to optimize the customer shopping experience and improve sales efficiency. They emphasize customer needs as the core, using scientific classification and layout to enhance product visibility and ease of selection. The classification system is divided into three levels: large, medium, and small. Large categories are based on product attributes and typically do not exceed 10 items. Medium categories focus on function, use, or manufacturing process, such as breakfast-related products or processed meat products. Small categories are further refined to dimensions such as specifications, ingredients, or flavors. Display design follows a dynamic adjustment principle, ensuring products are displayed frontally and avoiding obstruction. A golden display area of ​​60-150 cm is set according to customer height. Special locations utilize angled or repeated displays to enhance visibility. In practice, first-in, first-out (FIFO) management is emphasized to ensure product shelf-life safety. Related displays are strengthened, and themed combinations stimulate cross-selling. Products are arranged vertically or horizontally according to price gradients, with best-selling items occupying prime shelf positions. Meanwhile, it is required that the shelves have reasonable load-bearing capacity, the floor stack height not exceed 1.4 meters, and special areas be equipped with handwashing facilities or tools to eliminate shopping concerns. The display standards require uniform product location labeling, prohibit multiple displays, and create product vitality through full displays to avoid inventory exposure affecting the aesthetics of the store.

[0003] Existing product guides or product displays can only obtain data after the business hours have ended, and then adjust the product display accordingly, such as adding or removing goods. However, this method cannot be changed in real time, thus missing out on a large number of sales. Summary of the Invention

[0004] This invention provides an intelligent product recommendation and display system, which has the beneficial effects of improving marketing efficiency and automatically completing tasks such as optimizing product display and adjusting promotional displays.

[0005] This invention provides the following technical solution: an intelligent product recommendation and display system, wherein the intelligent product recommendation and display system includes:

[0006] The identification module records shoppers and items through a first camera and a second camera, and generates shopper data and item data;

[0007] The analysis module calculates shoppers' purchasing needs and preferences based on shopper data, and outputs adjustment instructions accordingly.

[0008] The shelving device includes a slide rail and a rotating base, which, according to adjustment instructions, display items closer to shoppers for easy access.

[0009] As an optional solution to the intelligent product recommendation and display system described in this invention, it further includes:

[0010] An electronic price tag, comprising a display screen for displaying item prices and promotional information;

[0011] The shopper data includes shopper visual dwell time, area dwell time, and tactile behavior data.

[0012] The analysis module sequentially calculates shoppers' visual dwell time, area dwell time, and picking behavior data to predict purchase demand, and displays suitable promotional plans based on the purchase demand and electronic price tags.

[0013] As an optional solution of the intelligent product recommendation and display system of the present invention, when a customer confirms the purchase of an item, the analysis module analyzes and pushes related accessory items based on the area division and associated attributes of the item, locates the coordinates of the accessory items and sends them to the electronic price tag of the purchased item.

[0014] The electronic price tag displays information about the accessory items;

[0015] The analysis module further analyzes the shopper data to determine purchasing needs. When a shopper shows a purchase intention, the electronic price tag's indicator light dynamically guides the shopper based on the coordinates of the accompanying items.

[0016] As an optional solution to the intelligent product recommendation and display system of the present invention, the item information includes product appearance, date and promotional information;

[0017] Filter items that fall within the middle of a given date range and obtain their coordinates;

[0018] The analysis module selects the item located at a later coordinate based on shopping psychology, controls the slide rail to move accordingly, illuminates the selected item with a guide light, and moves the item's date to face the shopper by rotating the base.

[0019] As an optional solution of the intelligent product recommendation and display system of the present invention, the analysis module is further used to determine whether the item needs to be removed from the shelves based on the item's appearance data and date data. When the item needs to be removed from the shelves, the analysis module sends the coordinates of the item to the administrator.

[0020] The analysis module is also used to calculate the value of items to be removed from shelves, thereby generating promotions for managers to determine.

[0021] Once the manager confirms, the analysis module prioritizes controlling the shelving unit and electronic price tags to push items to be removed from the shelves.

[0022] As an optional solution to the intelligent product recommendation and display system described in this invention, it further includes:

[0023] Calculate the shopper’s first average movement speed and the items that their eyes linger on based on shopper data;

[0024] After the shopper receives information about the accessory items, the identification module calculates the shopper's second average moving speed again;

[0025] When the second average moving speed is greater than the second average speed, and the moving path coincides with the path of the indicator light, the indicator light continues to guide the shopper;

[0026] When the second average moving speed is equal to or less than the second average speed, and the moving path does not completely coincide with the indicator light path, the indicator light stops guiding the shopper.

[0027] As an optional solution to the intelligent product recommendation and display system of the present invention, the identification module is further used to divide several shelves into several areas;

[0028] The analysis module dynamically adjusts the placement of items on several shelves based on recent sales data and visual dwell time data.

[0029] As an optional solution to the intelligent product recommendation and display system of the present invention, the analysis module calculates the shopper's body data based on the shopper data;

[0030] Models are built based on shoppers' physical data to simulate their activity range;

[0031] The analysis module calculates the optimal location of the shelf unit for easy access to items based on the shopper's activity range.

[0032] The present invention has the following beneficial effects:

[0033] 1. This intelligent product recommendation and display system proactively pushes highly-intended products to customers' front lines, reducing search and hesitation time. By optimizing the customer shopping experience, products automatically approach and are positioned directly in line of sight, supplemented by lighting guidance, making retrieval more convenient and intuitive. This is especially beneficial for the elderly, children, or customers carrying items, thereby increasing satisfaction. Intervention is based on real behavior rather than guesswork, avoiding ineffective promotions. Resources are concentrated on customers with genuine needs, improving marketing efficiency. The system automatically completes tasks such as optimizing product display and adjusting promotional displays, reducing the workload of store staff in frequently manually organizing shelves.

[0034] 2. This intelligent product recommendation and display system triggers precise related recommendations based on real purchasing behavior. Customers do not need to recall information themselves; the system proactively suggests reasonable combinations and uses lighting to intuitively guide locations, reducing search time and cognitive burden. Dynamic lighting guidance can divert customer flow, avoid congestion, and naturally guide customers to low-traffic areas, balancing the distribution of heat within the store. The light flow guidance has strong visual appeal and a futuristic feel, enhancing customers' perception of the store's level of intelligence and improving brand favorability.

[0035] 3. This intelligent product recommendation and display system proactively promotes the sales of products in the middle of their shelf life, preventing them from accumulating until near their expiration date and significantly reducing waste caused by expiration. It is especially suitable for fresh and short-shelf-life products. By leveraging the cognitive bias that products at the end of the shelf life are fresher, it places products that meet the date strategy in a psychologically advantageous position, enhancing customers' willingness to take them. It proactively directs date labels towards customers, conveying an image of integrity with no hidden information, and reducing purchase abandonment due to date doubts. The coordinated operation of the slide rails, rotating bases, and indicator lights gives static shelves the ability to proactively communicate, creating a differentiated shopping experience. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall structure of the present invention. Detailed Implementation

[0037] 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.

[0038] Example 1

[0039] Please see Figure 1 Furthermore, a smart product recommendation and display system was disclosed, including:

[0040] The identification module records shoppers and items through a first camera and a second camera, generating shopper data and item data.

[0041] The analysis module calculates shoppers' purchasing needs and preferences based on shopper data, and then outputs adjustment instructions accordingly.

[0042] The shelving unit includes slide rails and a rotating base. The slide rails and rotating base, according to adjustment instructions, display items closer to shoppers for easy access.

[0043] The first camera (overhead / aisle view) is mounted on the ceiling or top of the shelves to track shopper movement, areas they linger in, and their body orientation. It outputs the shopper's ID, location coordinates, dwell time, and approximate direction of their gaze.

[0044] The second camera (close-up view of the shelf) is installed on or in front of the shelf, focusing on the actions of picking up / placing goods, hand interactions, and the duration of eye contact with the goods. It outputs the product SKU, interaction type (pick up, put back, and view only), and interaction duration.

[0045] By aligning timestamps with locations in a spatiotemporal manner, for example, it can determine who did what and when on which product. This generates shopper and item data.

[0046] The analysis module then infers intent from shopper and item data to generate marketing and display adjustment instructions. Demand intensity is determined based on rules or machine learning models: high demand is characterized by prolonged eye contact, multiple picking up of items, and comparison with similar products; low interest is characterized by quick browsing and no interaction.

[0047] By combining historical data or group behavior patterns with the analysis module, such as 70% of customers who pick up this yogurt eventually make a purchase, or if there are many family customers at the current time, the conversion rate of high-calcium milk is high.

[0048] Furthermore, by using a reinforcement learning framework in the analysis module to increase the conversion rate as a reward signal, the strategy is continuously optimized, thereby outputting adjustment instructions.

[0049] Among them, the slide rails include horizontal electric slide rails on each shelf, and the merchandise pallets can move back and forth along the rails to slide the target merchandise from the back row to the front row.

[0050] The rotating base supports 0-180 degree rotation, ensuring that the label / packaging faces the customer and, with the help of focused lighting, creates a highlighted display area.

[0051] For example, a customer enters the monitored area (t=0s);

[0052] As a customer (anonymous ID: V1024) enters the store, the first overhead camera (wide-angle overhead view) begins tracking, using YOLOv8 and DeepSORT algorithms for human detection and cross-frame tracking, outputting updated position coordinates (x, y), orientation angle, and movement speed every second. Simultaneously, the second camera (installed in front of the yogurt shelf) remains on standby, activating high-frame-rate recording only when someone is detected within 1.5 meters of the shelf.

[0053] The customer then interacted with the product (t=5s), stopping in front of the probiotic yogurt (SKU:YOGURT_PROBIOTIC) with their body facing the shelf. The second camera captured the following sequence of behavior:

[0054] At t=5.2s, the gaze is focused on product A for 2.1 seconds;

[0055] At t=7.5s, the right hand reaches out to product A and picks it up;

[0056] t=9.8s, check the label on the back for 3.2 seconds;

[0057] At t=13.0s, the item is returned to its original position.

[0058] At t=14.5s, turn your head to look at competitor B next to you and pause briefly for 1 second.

[0059] The analysis module aggregates the above behaviors into a single product interaction record, from which it extracts behavioral features, spatial features, contextual features, and product attribute features. Based on these features, and using a lightweight XGBoost model with predefined rules, the module outputs the demand intensity to determine whether the purchase demand is high.

[0060] Then a pre-trained logistic regression model is used, including:

[0061] ;

[0062] ;

[0063] in, The variable is whether or not the item is picked up;

[0064] This is a variable indicating whether a product is currently on sale.

[0065] The total duration of customer interaction with products;

[0066] This is a variable indicating whether or not the item should be returned to its original position.

[0067] The weighted composite score of all features represents the "original strength of the purchase intention";

[0068] for function;

[0069] Obtain the purchase propensity rate using the formula above;

[0070] Subsequently, based on the purchase propensity rate and whether the purchase demand is high, the analysis module determines whether proactive guidance is needed, and generates an instruction accordingly. The instruction is sent to the corresponding shelf controller via the MQTT protocol;

[0071] The shelf controller (embedded ARM chip) interprets instructions, activates a micro stepper motor, and the slide rail system moves the product tray forward 25cm along the guide rail, making it protrude from other products. The rotating base rotates 30 degrees clockwise to ensure the brand logo is directly facing the customer's line of sight to the LED light strip. The LEDs below the product turn to a warm white high-brightness mode, and a confirmation signal ACTION_SUCCESS is sent back upon completion. If a customer's hand is detected reaching towards the product (based on real-time posture estimation via a second camera), the movement is paused to prevent collision.

[0072] If a customer makes a purchase, the POS system reports "YOGURT_PROBIOTIC sold to V1024". The analysis module records the successful intervention and reinforces the strategy's weight. If a customer leaves without making a purchase, the recognition module detects the customer leaving the area without further interaction and triggers a negative sample: reducing the score of the strategy still recommended after `put_back`. The shelf automatically resets. All results are used for online learning to continuously optimize model parameters.

[0073] In summary, this example demonstrates how the system records customer behavior in real-time as they enter the store and approach the merchandise area using cameras mounted on the ceiling and shelves. This includes their walking path, the time spent in front of a product, whether they look at it, and actions such as picking up or putting it back. This raw video data is not saved but analyzed instantly on a local edge device. Subsequently, based on this behavioral data, combined with product information (such as whether it's on sale, category attributes), environmental factors (such as the current time of day, customer density), and the customer's real-time status (such as distance from the shelf, whether they are comparing multiple products), the system uses preset rules and a lightweight AI model to quickly calculate the intensity of purchase demand. This determines the strength of the customer's interest in the product and the probability of purchase, predicting whether the customer will ultimately buy it. Once the system determines that a customer has a high demand for a product and a high probability of purchase (e.g., a probability exceeding 70%), it immediately generates an adjustment command and sends it to the corresponding smart shelf. This shelf, with its built-in sliding rails and rotating base, automatically moves the target product forward along the rails to a more easily accessible position upon receiving the command, and slightly rotates it so that the product faces the customer, while simultaneously highlighting it with a light, thus proactively welcoming the customer.

[0074] In summary, by proactively placing high-intent products in front of customers, the time spent searching and hesitating is reduced, effectively promoting the final purchase decision. This is especially effective for high-margin or new products. Optimizing the customer shopping experience involves automatically positioning products closer to the customer's line of sight and using lighting guidance, making them easier and more intuitive to retrieve. This is particularly beneficial for the elderly, children, or customers carrying items, increasing satisfaction. Interventions are based on real behavior rather than guesswork, avoiding ineffective promotions. Resources are concentrated on customers with genuine needs, improving marketing efficiency. The system automatically completes tasks such as optimizing product display and adjusting promotional displays, reducing the workload of store staff who frequently manually organize shelves.

[0075] Example 2

[0076] This embodiment is an improvement upon embodiment 1. For details, please refer to [link / reference]. Figure 1 It also includes:

[0077] Electronic shelf labels, which include a display screen, are used to display the price and promotional information of items;

[0078] Shopper data includes shopper visual dwell time, area dwell time, and picking behavior data;

[0079] The analysis module sequentially calculates shoppers' visual dwell time, area dwell time, and picking behavior data to predict purchase demand, and then displays suitable promotional plans based on purchase demand and electronic price tags.

[0080] The system uses the first camera deployed at the top of the store and the second camera in front of the shelves to record customer behavior in real time and generate two types of structured data:

[0081] Shopper data, including:

[0082] Visual dwell time, which is the cumulative duration that the customer's line of sight focuses on a certain product (estimated from the line of sight direction and head pose);

[0083] Area dwell time, which is the total time the customer stays in a certain product area (such as the yogurt area);

[0084] Taking behavior data, including interaction details such as whether the product is picked up, the number of pick-ups, the holding duration, and whether it is put back.

[0085] Item data includes product SKU, current price, inventory status, category attributes, and a preset promotion strategy library.

[0086] The analysis module processes the above three types of behavior data in order of priority, inferring the purchase intention layer by layer. First, it judges the initial interest based on the visual dwell time. If the customer's visual dwell on a certain product exceeds the threshold (such as 3 seconds), it is regarded as attracting attention and enters the next analysis level. Subsequently, it combines the area dwell time to evaluate the decision-making depth. If the customer stays in the area for a long time (such as >15 seconds) and makes multiple round trips between similar products, it indicates that the customer is in the comparison and decision-making stage and the demand intensity increases. Finally, it confirms the high intention through the taking behavior. Once the picking-up action occurs, it is immediately determined as a high-value signal. If it is accompanied by a long holding time or not being put back, the purchase tendency is further strengthened. Combining the above three layers of information, it outputs a purchase demand level (low, medium, high) and the corresponding purchase probability value. Subsequently, the analysis module matches the most suitable marketing strategy from the preset promotion plan library according to the predicted purchase demand and synchronously controls two types of terminals:

[0087] Electronic price tag linkage. The electronic price tag is equipped with a low-power display screen and can update the content remotely. It dynamically adjusts the display information according to the demand level. For high demand, it displays urgent copywriting such as limited-time special offer, only X pieces left, recommended for you, and the QR code for getting coupons. For medium demand, it displays regular discount information such as buy one get one half price. For low demand, it keeps the original price or displays new products and high-margin substitute information.

[0088] Shelving unit linkage. For high-demand products, it synchronously triggers the shelf slide rail and rotating base to move the product to an easily accessible position and face the customer directly. There is a feedback loop. If the customer finally purchases, it records the result of this behavior intervention for optimizing the subsequent demand prediction model and promotion strategy matching rules. If the customer does not purchase, it analyzes the reason for failure (such as not paying attention after putting it back) and dynamically adjusts the threshold or promotion intensity.

[0089] In summary, promotional information is no longer uniform across the entire store, but is personalized based on each customer's real-time behavior. This avoids interfering with those who are not interested, while strengthening the decision-making motivation of those with high interest, thus enhancing the value of electronic shelf labels. Traditional electronic shelf labels are only used for static price adjustments. This system transforms them into dynamic marketing terminals, supporting rich media content such as graphics, countdowns, and personalized slogans, significantly improving information delivery efficiency. Single behaviors (such as brief stops) are prone to misjudgment, while the three-level verification mechanism of visual, regional, and pickup greatly improves the accuracy of demand identification and reduces false triggers. For high-margin, new, or slow-moving products, exclusive offers can be proactively pushed when potential interest is detected, accelerating trial use and inventory clearance.

[0090] Example 3

[0091] This embodiment is an improvement upon embodiment 2. For details, please refer to [link / reference]. Figure 1 Once a customer confirms the purchase of an item, the analysis module analyzes and pushes related accessories based on the item's regional classification and associated attributes, locates the coordinates of the accessories, and sends them to the electronic price tag of the purchased item.

[0092] Electronic price tags display information about accessory items;

[0093] The analysis module further analyzes the shopper data to determine purchasing needs. When a shopper shows a purchase intention, the electronic price tag's indicator light dynamically guides the shopper based on the coordinates of the accompanying items.

[0094] When a customer confirms the purchase of an item, such as adding it to their shopping cart, the intelligent recommendation and guidance process for related products is triggered. The analysis module immediately retrieves detailed information about the purchased item, including its region (e.g., dairy products or cleaning supplies), product attributes (e.g., category, purpose, brand, specifications), and a pre-defined association rule base (based on historical shopping basket data, product knowledge graphs, or manual configuration). For example, purchasing coffee capsules might associate the purchase with related items such as coffee machine cleaning tablets, milk, and sugar packets.

[0095] Select accessory products that are currently in stock, located in the store, and conform to the scenario logic from the association rules, and accurately locate the physical location coordinates of each accessory product using the store's digital map (including the shelf coordinate system) (e.g., X=4.2m, Y=6.8m, corresponding to the second layer of shelf B3).

[0096] The electronic price tag corresponding to the main product will temporarily activate its display screen even after the customer has left the shelf, and be instructed to update the content to display information on the auxiliary products, including product names and pictures, brief recommendations and promotional information, and dynamic guide light control signals;

[0097] During the display of accessory product information, customer behavior is continuously monitored using first and second cameras: whether they are looking at the accessory product information on the electronic price tag, whether they are moving towards the area where the accessory products are located, and whether they linger or take items from the accessory product area. Based on this newly generated shopper data, the analysis module recalculates the customer's purchase demand level and probability of preference for the accessory products.

[0098] If the system determines that a customer has a clear purchase intention for a particular accessory (e.g., intention ≥ 60%), it immediately activates the built-in guide light on the electronic shelf tag. Combined with the accessory's coordinates, it executes dynamic light flow guidance, with the electronic shelf tag light flashing first. Simultaneously, along the path from the customer's current location to the accessory shelf, floor-mounted sensor lights or shelf-edge light strips illuminate sequentially, forming a light path. If the electronic shelf tag itself supports directional indication, it displays an arrow icon in conjunction with the light guidance. This guidance process continues until the customer approaches the target shelf or a timeout period. If there is no response after 30 seconds, it automatically shuts off.

[0099] In summary, based on real purchasing behavior, precise related recommendations are triggered, eliminating the need for customers to recall information and providing proactive suggestions for appropriate combinations. The location is intuitively guided by lighting, reducing search time and cognitive burden. Dynamic lighting can also help manage customer flow, avoid congestion, and naturally direct customers to areas with low foot traffic, balancing the distribution of heat within the store. The light flow guidance has strong visual appeal and a futuristic feel, enhancing customers' perception of the store's level of intelligence and improving brand favorability.

[0100] Example 4

[0101] This embodiment is an improvement upon embodiment 3. For details, please refer to [link / reference]. Figure 1 Item information includes product appearance, date, and promotional information;

[0102] Filter items that fall within the middle of a given date range and obtain their coordinates;

[0103] The analysis module selects items located at a later coordinate based on shopping psychology, controls the movement of the slide rail, illuminates the selected items with guide lights, and moves the items' dates to face the shopper by rotating the base.

[0104] Based on the identification of product information, this embodiment combines date attributes, shopping psychology, and spatial coordinates to achieve intelligent filtering, physical adjustment, and visual enhancement of specific products;

[0105] Specifically, information about all products on the shelves is obtained through cameras or a product database, including product appearance (such as packaging color, shape, and brand logo).

[0106] Date information (such as production date, expiration date, and shelf-life date);

[0107] Promotional information (such as discounts, spending thresholds, and limited-time offers);

[0108] All products are associated with their precise physical coordinates on the shelf (e.g., X=2.1m, Y=3.5m, Z=2nd shelf).

[0109] For products with shelf-life management requirements (such as fresh food, dairy products, and baked goods), date filtering is performed according to preset rules. Several similar products (e.g., 10 boxes of the same yogurt) are selected, and the expiration date of each product is extracted. Products with expiration dates in the middle of the range are automatically identified and filtered (e.g., products with a total shelf life of 7 days, currently on day 3-5). This strategy prioritizes selling products that are neither the newest nor near their expiration date, balancing inventory turnover and loss control.

[0110] By incorporating principles of consumer behavior psychology into the analysis module, research shows that when faced with a row of similar products, customers tend to choose items further back (i.e., deeper in the row, requiring a little reaching), perceiving them as fresher and less likely to have been picked over by others. Therefore, from the products in the aforementioned mid-date range, prioritize selecting the item with the later physical coordinate (e.g., the largest X-value) as the target product.

[0111] Send a composite instruction to the smart shelf unit where the target product is located, including:

[0112] The slide rail moves, driving the target product from the back row to the front to an easily accessible position (usually 5-10cm from the front edge of the shelf), lowering the threshold for picking it up;

[0113] The guide lights illuminate and activate the bright LED guide lights (such as warm white spotlights) below or above the product, creating a visual focal point and attracting customers' attention.

[0114] Rotating base adjustment: Controls the automatic rotation angle of the product base to ensure that the side with the date information is facing the customer's line of sight, improving information transparency and trust.

[0115] If a customer views and purchases the product under the light prompt, the date-guided strategy is recorded as successful. If no one inquires for a long time, the filtering logic may be adjusted (such as widening the date range) or the target product may be changed to achieve strategy self-optimization.

[0116] In summary, this embodiment proactively promotes the sale of products in the middle of their expiration period, preventing them from accumulating until near their expiration date and significantly reducing waste caused by expiration. It is particularly suitable for fresh and short-shelf-life products. By leveraging the cognitive bias that products with later expiration dates are fresher, it places products that meet the date strategy in a psychologically advantageous position, enhancing customers' willingness to take them. It proactively directs date labels towards customers, conveying an image of integrity with no hidden information, and reducing purchase abandonment due to date doubts. The coordinated action of the slide rails, rotating bases, and indicator lights gives the static shelves proactive communication capabilities, creating a differentiated shopping experience.

[0117] Example 5

[0118] This embodiment is an improvement upon embodiment 4. For details, please refer to [link / reference]. Figure 1 The analysis module is also used to determine whether an item needs to be removed from the shelves based on its appearance and date data. When an item needs to be removed from the shelves, the analysis module sends the coordinates of the item to the administrator.

[0119] The analytics module is also used to calculate the value of items to be removed from shelves, thereby generating promotions for managers to determine.

[0120] Once management confirms, the analytics module prioritizes controlling the shelving units and electronic price tags to push items to be removed from the shelves.

[0121] Calculate the shopper’s first average movement speed and the items that their eyes linger on based on shopper data;

[0122] After the shopper receives information about the accessory items, the recognition module recalculates the shopper's second average moving speed.

[0123] When the second average moving speed is greater than the second average speed, and the moving path coincides with the path of the indicator light, the indicator light continues to guide the shopper.

[0124] When the second average moving speed is equal to or less than the second average speed, and the moving path does not completely coincide with the indicator light path, the indicator light stops guiding the shopper.

[0125] After customers receive recommendations for additional products, the system continuously monitors their behavior and intelligently determines whether to continue providing lighting guidance services.

[0126] Specifically,

[0127] Initial behavior modeling: Calculate the first average movement speed and the location data of the items of interest as the customer browses the main merchandise area. The recognition module continuously collects this location data, obtained by the first camera through visual tracking, and calculates:

[0128] First average moving speed v1, the average walking speed of the customer from entering the current area to receiving information about the accompanying merchandise;

[0129] The list of items whose visual attention lingers for more than a threshold during this period is used as the initial interest profile.

[0130] Once the supplementary recommendation is triggered, secondary behavior monitoring is initiated. When the electronic price tag displays supplementary product information (such as "better with milk"), a new round of behavior analysis window is immediately opened to track the customer's subsequent movements in real time.

[0131] The second average moving speed v2 is calculated. Within a preset time window (e.g., 10–30 seconds) after the recommendation information is displayed, the identification module continuously acquires the customer's location and calculates their second average moving speed v2, which reflects the urgency or hesitation of their current action intention.

[0132] The analysis module compares v2 with v1 and matches the actual movement path with the system's preset indicator light guidance path (the optimal route from the current location to the auxiliary merchandise shelf) to determine the best course of action, and then executes the following strategy:

[0133] In the first scenario, if the following two conditions are met, continue guiding:

[0134] v2>v1, the customer moves faster, indicating that they are interested in the recommendation and actively go there;

[0135] The actual movement path and the indicator light path are highly coincident (e.g., the directional angle is <30 degrees and the deviation distance is <0.5m).

[0136] Once it is determined that the customer is responding to the guidance, the indicator light continues to illuminate and dynamically advances along the path (such as the floor light strip lighting up segment by segment) until the customer approaches the target shelf.

[0137] In the second scenario, if any of the following situations occur, the guidance will be stopped:

[0138] If v2≤v1, the customer's speed has not increased or has even slowed down, indicating hesitation or lack of interest;

[0139] The movement path deviates significantly from the indicator light path (e.g., turning to other areas, staying in place, or walking in the opposite direction);

[0140] If the guidance is deemed ineffective, immediately turn off the indicator light to avoid light pollution, energy waste, and disturbance to customers.

[0141] In summary, continued guidance should only be provided when customers demonstrate a clear willingness to respond, avoiding pushy sales tactics and respecting customer autonomy. The dual verification of speed variations and path matching is more accurate in determining true intentions than a single behavioral signal, such as eye contact.

[0142] Example 6

[0143] This embodiment is an improvement upon embodiment 5. For details, please refer to [link / reference]. Figure 1 The analysis module calculates shoppers' physical data based on their data.

[0144] Models are built based on shoppers' physical data to simulate their activity range;

[0145] The analysis module calculates the optimal location for easily accessible shelving units based on the shopper's activity range.

[0146] The recognition module uses the first camera to collect real-time visual data of customers and extracts key body parameters through human posture estimation algorithms, such as MediaPipe Pose or OpenPose, including height (estimated from the vertical distance from the top of the head to the bottom of the feet), arm span (calculated based on shoulder width and arm length), current standing position and orientation, hand height and reachable area (calculated in real time in combination with key joint points).

[0147] Based on the aforementioned body data, the analysis module constructs a three-dimensional reachable space model, simulating the physical area that the customer can comfortably reach from their current position without having to tiptoe or bend over. This model is typically represented as a semi-ellipsoidal or trapezoidal three-dimensional space centered on the customer, with its boundaries defined by:

[0148] Horizontally, approximately 60-80cm to the left and right;

[0149] Vertically, from waist level (approximately 0.8m) to below the top of the head;

[0150] In terms of depth, 0.3-0.7m ahead.

[0151] The model can be updated dynamically. The activity range is recalculated in real time if a customer moves or turns around.

[0152] When it's determined that a product needs to be recommended to a customer (e.g., due to high purchase intent, related products, etc.), the analysis module will obtain the product's current coordinates on the shelf. Combining this with a real-time customer activity range model, it will determine an optimal display location. This means moving the product to a point closest to the center of the customer's activity range and within easy viewing and access areas, while ensuring product safety and shelf structure constraints. This location must simultaneously satisfy the following conditions:

[0153] Located within a ±30 degree field of vision cone directly in front of the customer;

[0154] The height should be within the range where the hands can comfortably operate (e.g., height × 0.6 to height × 0.9).

[0155] The depth should not exceed the maximum safe reach distance (usually ≤0.6m).

[0156] The analysis module then translates the optimal location into control commands and sends them to the corresponding shelving unit.

[0157] The slide rail system moves the merchandise tray horizontally or vertically, bringing it into the customer's activity area;

[0158] Rotate the base to adjust the product's orientation, ensuring the label is directly in the customer's line of sight;

[0159] Further add a lifting mechanism; if the shelf supports vertical adjustment, fine-tune the shelf height to match the customer's height.

[0160] After the adjustment, the product is positioned in a prime location where customers can easily reach it.

[0161] It should be noted that the dynamic adaptation and reset feature means that if a customer leaves or does not interact for an extended period of time, the shelf will automatically return to its original position to prepare for the next customer.

[0162] In summary, this example automatically adapts to customers of different heights and body types, keeping products in a comfortable operating area at all times. This reduces actions such as bending over, tiptoeing, and pulling, improving shopping comfort and indirectly promoting browsing and purchasing. Recommended products are accurately placed in easily accessible prime areas, significantly increasing the probability of actual pickup, especially for heavy items and high-margin products.

[0163] The calculation formula in this example includes:

[0164] Height estimation formula, used to estimate the customer's height, for subsequent calculations of arm length, comfortable operating height, etc.

[0165] ;

[0166] This represents the coordinates of the top of the head in three-dimensional space, detected by a camera (such as a depth camera).

[0167] x represents the left-right direction (horizontal direction);

[0168] y represents the vertical direction (vertical, usually 0 for the ground);

[0169] z represents the front-to-back direction (depth);

[0170] This represents the 3D coordinates of the foot (usually the ankle joint or the center of the sole);

[0171] Represents Euclidean distance;

[0172] For example, the coordinates of the top of the head are (1.2, 1.75, 3.0), and the coordinates of the bottom of the feet are (1.2, 0.05, 3.0). Using the formula, we can calculate (0)² + (1.70)² + (0)² = 1.7 meters.

[0173] Arm length estimation formula: quickly estimate half of your arm span based on your height.

[0174] ;

[0175] Of these, 0.44 represents ergonomic research, indicating that the average length of an adult's arm (from shoulder to middle fingertip) is approximately 44% of their height.

[0176] For example, if a person is 1.7 meters tall, then 0.44 * 1.7 = 0.75 meters.

[0177] The reachable space ellipsoid model uses an ellipsoid to approximate the three-dimensional space that a customer can "reach without effort" from their current position, including:

[0178] ;

[0179] in, The center point of the shoulder serves as the origin of the accessible space;

[0180] The result was obtained by averaging the key points on the left and right shoulders:

[0181] ;

[0182] a, b, and c are the semi-axial lengths of the ellipsoid in three directions (i.e., the maximum reachable distance):

[0183] a (left and right directions):

[0184] ;

[0185] The factor multiplied by 0.8 is because a person cannot fully extend their arm horizontally (due to the torso's obstruction), and the effective horizontal range is usually about 80% of the arm length.

[0186] b (vertical direction):

[0187] ;

[0188] This indicates the height that can be easily reached upwards (approximately 30% of your height above shoulder height).

[0189] c (front-to-back direction, depth):

[0190] ;

[0191] When reaching forward, the center of gravity restricts the ability to safely extend only 60% of the arm's length.

[0192] The inequality ≤1 means that if the result of substituting a point (x, y, z) into the left side is ≤1, then that point is within the reachable range.

[0193] If the value is greater than 1, it is outside the range and requires bending over / standing on tiptoe / taking a step to reach it.

[0194] Eye alignment loss measures whether the front of the product is directly facing the customer's eyes. A smaller value indicates better alignment.

[0195] ;

[0196] in, This is the normal vector for the front of the product; by default, the product on the shelf faces forward.

[0197] ;

[0198] From the position of the eyes Pointing to product location The vector (i.e., the direction of the line of sight);

[0199] The overall optimization objective function, considering three factors, finds the most comfortable product display position overall:

[0200] ;

[0201] in, For the loss function, the smaller the value, the better;

[0202] Weight , , This indicates the importance of each item.

[0203] Determined by business requirements Usually the highest, because not being able to get it is a major drawback;

[0204] (Line of sight) is secondary, affecting information acquisition;

[0205] Height is also important, but it can be overcome by bending over briefly.

[0206] Try multiple Location, calculate each Choose the smallest one.

[0207] 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.

[0208] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent product recommendation and display system, characterized in that, include: The identification module records shoppers and items through a first camera and a second camera, and generates shopper data and item data; The analysis module calculates shoppers' purchasing needs and preferences based on shopper data, and outputs adjustment instructions accordingly. The shelving device includes a slide rail and a rotating base, which, according to adjustment instructions, display items closer to shoppers for easy access.

2. The intelligent product recommendation and display system according to claim 1, characterized in that, Also includes: An electronic price tag, comprising a display screen for displaying item prices and promotional information; The shopper data includes shopper visual dwell time, area dwell time, and tactile behavior data. The analysis module sequentially calculates shoppers' visual dwell time, area dwell time, and picking behavior data to predict purchase demand, and displays suitable promotional plans based on the purchase demand and electronic price tags.

3. The intelligent product recommendation and display system according to claim 2, characterized in that: Once a customer confirms the purchase of an item, the analysis module analyzes and pushes related accessory items based on the area classification and associated attributes of the item, locates the coordinates of the accessory items, and sends them to the electronic price tag of the purchased item. The electronic price tag displays information about the accessory items; The analysis module further analyzes the shopper data to determine purchasing needs. When a shopper shows a purchase intention, the electronic price tag's indicator light dynamically guides the shopper based on the coordinates of the accompanying items.

4. The intelligent product recommendation and display system according to claim 1, characterized in that: The item information includes product appearance, date, and promotional information; Filter items that fall within the middle of a given date range and obtain their coordinates; The analysis module selects the item located at a later coordinate based on shopping psychology, controls the slide rail to move accordingly, illuminates the selected item with a guide light, and moves the item's date to face the shopper by rotating the base.

5. The intelligent product recommendation and display system according to claim 4, characterized in that: The analysis module is also used to determine whether the item needs to be removed from the shelves based on the item's appearance data and date data. When the item needs to be removed from the shelves, the analysis module sends the coordinates of the item to the administrator. The analysis module is also used to calculate the value of items to be removed from shelves, thereby generating promotions for managers to determine. Once the manager confirms, the analysis module prioritizes controlling the shelving unit and electronic price tags to push items to be removed from the shelves.

6. The intelligent product recommendation and display system according to claim 5, characterized in that, Also includes: Calculate the shopper’s first average movement speed and the items that their eyes linger on based on shopper data; After the shopper receives information about the accessory items, the identification module calculates the shopper's second average moving speed again; When the second average moving speed is greater than the second average speed, and the moving path coincides with the path of the indicator light, the indicator light continues to guide the shopper; When the second average moving speed is equal to or less than the second average speed, and the moving path does not completely coincide with the indicator light path, the indicator light stops guiding the shopper.

7. The intelligent product recommendation and display system according to claim 6, characterized in that: The identification module is also used to divide several shelves into several areas; The analysis module dynamically adjusts the placement of items on several shelves based on recent sales data and visual dwell time data.

8. The intelligent product recommendation and display system according to claim 1, characterized in that: The analysis module calculates the shopper's physical data based on the shopper's data; Models are built based on shoppers' physical data to simulate their activity range; The analysis module calculates the optimal location of the shelf unit for easy access to items based on the shopper's activity range.

Images