An LED window content control method, system, terminal and medium
By acquiring pedestrian feature vectors and estimated arrival times, and combining them with viewing duration records to dynamically update the consumption coefficient, the problem of LED window display systems being unable to update in real time has been solved, enabling adaptive adjustment of content and highly matched display.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG INSPUR ULTRA HD INTELLIGENT TECH CO LTD
- Filing Date
- 2025-12-10
- Publication Date
- 2026-07-31
AI Technical Summary
Existing LED window display systems cannot be updated in real time according to dynamic changes in the crowd, resulting in a low degree of matching between the displayed content and the target audience.
By acquiring pedestrian feature vectors and calculating estimated arrival times, content prediction is performed based on the matching degree between the weighted fusion target pedestrian feature vector and the displayed content feature vector. Furthermore, the consumption coefficient is dynamically updated by recording viewing time, thereby achieving adaptive adjustment of the content.
It significantly improved the timeliness and relevance of content playback, enhanced the attractiveness and conversion rate of showcase content, and achieved real-time adaptive capability under dynamic changes in the audience.
Smart Images
Figure CN121304256B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of LED window display content technology, specifically relating to an LED window display content control method, system, terminal, and medium. Background Technology
[0002] The development of LED display technology has driven the digital upgrade of commercial windows, which have become an important promotional medium for offline retail, commercial complexes, and brand stores. With advancements in the Internet of Things, artificial intelligence, and visual perception technologies, commercial displays are gradually shifting from static advertising to intelligent, interactive playback modes. More and more commercial scenarios are looking to leverage pedestrian recognition, behavioral analysis, and audience insights to achieve refined content delivery, thereby improving the match between displayed content and the target audience.
[0003] In existing technologies, some LED shop windows incorporate cameras, remote management platforms, or simple interactive terminals to support basic data collection and content display. However, these systems primarily focus on device networking, remote monitoring, or passive interaction via touch or QR code scanning. For content recommendation, they employ fixed rules or static tag matching schemes.
[0004] Existing LED window display systems lack the ability to update in real time based on dynamic changes in the crowd, making it impossible for the displayed content to adapt to changes in time, crowd composition, and behavioral patterns, resulting in low matching degree of the displayed content. Summary of the Invention
[0005] This invention addresses the problems in the prior art by providing an LED shop window content control method, system, terminal, and medium. It solves the problem that LED shop window systems lack the ability to update based on dynamic changes in the crowd, resulting in the inability of the displayed content to adapt to changes in time, crowd composition, and behavioral patterns, leading to low content matching.
[0006] The technical solution adopted in this invention is as follows: In a first aspect, this application provides a method for controlling the content of an LED shop window, the method comprising the following steps: Get the remaining playback time of the currently playing content. When the remaining playback time falls into a preset time threshold, start the prediction process for the next content to be displayed. Acquire target data of multiple pedestrians located in the target area in front of the LED shop window, extract features from the target data, and obtain the pedestrian feature vector, position, direction of movement, and speed of movement for each pedestrian; Based on the location, direction of movement and speed of each pedestrian, the estimated arrival time of each pedestrian entering the effective display area of the LED window is calculated, and pedestrians who are expected to enter the effective display area within the content switching prediction window are identified as the target group. Based on the pedestrian feature vectors of each pedestrian in the target population and their corresponding weights, the pedestrian feature vectors are weighted and fused to obtain the target pedestrian feature vectors for content prediction. Based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-built display content, the target display content for the next display cycle is predicted; After the current content has finished playing, output the predicted target content to be displayed.
[0007] Furthermore, in the process of identifying the target audience, the relative positions of each pedestrian and the effective display area are considered. and the speed of movement of each pedestrian The angle between the direction of movement and the direction of the line connecting the effective display area ,according to Calculate the estimated arrival time of each pedestrian, when If a pedestrian is determined to have crossed the LED window, they are removed from the list. The remaining pedestrians with a positive estimated arrival time are considered as the target group.
[0008] Furthermore, target data of multiple pedestrians is obtained, and feature extraction is performed on the target data to obtain the pedestrian feature vectors of each pedestrian, including: Image data of pedestrians is collected, and age and gender features of pedestrians are extracted from the image data based on deep learning feature extraction algorithms to construct pedestrian feature vectors.
[0009] Furthermore, the weight of each pedestrian in the target population is determined sequentially by gender and age group. Gender identification is based on the proportion of pedestrians of different genders in the target population, selecting the gender that matches the gender bias of the displayed content, and using that gender as the target gender. After determining the target gender, pedestrians within that gender are divided into preset age ranges, and a corresponding consumption coefficient is assigned to each age group. Age weighting under target gender is based on The calculation yielded, where The number of rows in the target gender that fall within the j-th age range; The target pedestrian feature vector is determined based on the target's gender and age weights.
[0010] Furthermore, the target display content for the next display cycle is predicted based on the matching degree between the target pedestrian feature vector and the content feature vectors of each pre-constructed display content. Let the feature vector of the target pedestrian be denoted as Let the content feature vector of the kth displayed content be denoted as ,according to Calculate the matching degree between the feature vector of the target pedestrian and the feature vector of each displayed content, and select the displayed content with the highest matching degree as the target displayed content for the next display cycle.
[0011] Furthermore, after outputting the target display content for the next display cycle, the viewing time of each target pedestrian is recorded, and a corresponding consumption coefficient is set for each age group based on the viewing time and the corresponding age of the target pedestrian. It is dynamically updated.
[0012] Furthermore, the consumption coefficient for each age group was calculated based on viewing time and the corresponding age of the target pedestrians. Dynamic updates include: Let the sum of the viewing time of the target pedestrians in the j-th age group be denoted as . The sum of the viewing time of all target pedestrians of all age groups is recorded as The percentage of viewing time for the j-th age group is calculated as follows:
[0013] Calculate the average percentage of viewing time across all age groups. The consumption coefficient for each age group is calculated using the following formula. Update:
[0014] in, For the updated consumption coefficient, This is the preset update step size parameter.
[0015] Secondly, this application provides an LED shop window content control system for implementing the LED shop window content control method as described in the first aspect, the system comprising: The pedestrian target data acquisition unit is used to acquire target data of multiple pedestrians located in the target area in front of the LED window, extract the age and gender characteristics of pedestrians based on pedestrian image data, and obtain the pedestrian feature vector, position, direction of movement and speed of each pedestrian; The estimated arrival time calculation unit is used to calculate the estimated arrival time of each pedestrian entering the effective display area based on the relative position, moving speed, and angle between the moving direction and the direction of the line connecting to the effective display area. Pedestrians with negative estimated arrival times are eliminated to determine the target group. The target pedestrian feature vector generation unit is used to generate the target pedestrian feature vector by weighted fusion of the pedestrian feature vectors of each pedestrian in the target crowd and their corresponding weights determined by gender and age group in turn. The content prediction unit is used to predict the target display content for the next display cycle based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-built display content. The content output unit is used to output the predicted target content for the next display cycle after the current content has finished playing. The viewing time recording unit is used to record the viewing time of the target pedestrian during the viewing process after the target display content of the next display cycle is output; The consumption coefficient update unit is used to dynamically update the consumption coefficient set for each age group based on the viewing time and the age of the corresponding target pedestrian.
[0016] Thirdly, this application provides a terminal, including: The memory is used to store the LED window display content control program; A processor is configured to perform the steps of the LED window content control method as described in the first aspect when executing the LED window content control device.
[0017] Fourthly, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the LED window content control method as described in the first aspect.
[0018] As can be seen from the above technical solutions, the advantages of the present invention are: By introducing a real-time perception mechanism for pedestrian location, direction of movement, and speed, and combining this with the remaining playback time of the displayed content for trigger judgment, this method can predict the arrival time of pedestrians before they enter the effective display area, enabling predictive content switching and significantly improving the timeliness and targeting of content playback. This method uses a weighted profile of the target audience as the basis for content decision-making, significantly improving the matching degree between displayed content and the audience, enhancing the attractiveness and conversion effect of the showcase content, and simultaneously achieving real-time adaptive capabilities under dynamic changes in the audience.
[0019] By calculating the estimated time of arrival (ETA) of pedestrians entering the effective display area based on their location, speed, and direction of movement, and removing pedestrians with negative ETAs, it is possible to effectively filter out pedestrian targets that have already passed the window or are unrelated to the window. This ensures that the prediction logic is based only on pedestrians who are truly likely to appear in the effective display area, thereby improving the accuracy and efficiency of the prediction results and avoiding accidental triggering of content switching.
[0020] By extracting age and gender features from pedestrian images using deep learning-based feature extraction algorithms, stable, quantifiable, and content-matchable basic profile data of the population can be obtained. This is more objective and reliable than relying on manual observation or simple rule judgment, which helps improve the accuracy of population profile construction and supports the accuracy of subsequent weighted fusion and content prediction.
[0021] By decomposing pedestrian weights into a two-tiered judgment mechanism of gender and age group, and introducing an age group consumption coefficient for weighting, a quantitative estimate of the overall preferences of the target population can be achieved. This allows the target pedestrian feature vector to accurately represent the true consumption tendencies of the current population structure. This mechanism effectively solves the problem that a single pedestrian feature cannot represent the entire population, enabling content prediction to move beyond single-point profiles and instead rely on a comprehensive judgment based on group preferences, thereby improving the rationality and attractiveness of displayed content.
[0022] By employing a matching degree calculation mechanism between target pedestrian feature vectors and content feature vectors, a quantitative correspondence between audience preferences and content attributes can be achieved within a unified feature space. This allows for the automatic selection of the most suitable content for display based on vector similarity. This approach not only supports different types of content tagging systems but also allows for multi-dimensional preference expression through the expansion of content feature vectors, improving the accuracy and scalability of content recommendations.
[0023] By recording the actual viewing time of target pedestrians, the attractiveness of the currently displayed content to the audience can be accurately reflected. Incorporating the viewing time as a feedback signal into the model enables the system to have real-time perception of the content's effectiveness, providing a reliable basis for the adaptive optimization of subsequent content display logic, thereby achieving continuous iteration and dynamic evolution of the preference model.
[0024] By dynamically adjusting the consumption coefficient for each age group based on the difference between the percentage of viewing time and the average percentage, the consumption coefficient can reflect changes in the viewing preferences of the actual population, enabling preference parameters to be updated over time and adaptively adjusted according to population changes. This update method can strengthen the influence of the age group with high viewing time and weaken the influence of the age group with low viewing time, thereby continuously improving the accuracy of the content prediction model. Attached Figure Description
[0025] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 This is a flowchart illustrating the steps of the LED shop window content control method in this embodiment. Detailed Implementation
[0027] 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.
[0028] Please see Figure 1 As shown, this application provides a method for controlling the content of an LED shop window, including: Step S1: Obtain the remaining playback duration of the currently playing content. When the remaining playback duration falls within a preset time threshold, start the prediction process for the next content to be displayed. In one feasible approach, the control unit continuously tracks the played duration and total duration of any displayed content while playing, and calculates the remaining playback time in real time. When the remaining playback time falls below a preset threshold, such as a few seconds before the end of playback, the system automatically determines whether to predict the next displayed content in advance. This preset threshold can be configured based on the pedestrian flow characteristics of the scene where the shop window is located. For example, in scenarios with fast-moving pedestrians, prediction can be made in advance to prevent content switching delays. In this way, the system can complete the analysis of pedestrian status data and the decision on the next content before the playback ends, thereby maintaining the continuity and smoothness of the playback logic and avoiding short pauses after playback ends.
[0029] Step S2: Obtain target data of multiple pedestrians located in the target area in front of the LED window, extract features from the target data, and obtain the pedestrian feature vector, position, direction of movement and speed of each pedestrian. In practical applications, the target area can be defined by a preset spatial range, such as an area within a certain distance and width in front of a shop window. The system captures video streams within this area using a camera and identifies the positions of each pedestrian using a pedestrian detection model. By combining coordinate changes between consecutive frames, the system calculates the pedestrian's speed and direction. Simultaneously, based on an image analysis model, it detects the pedestrian's appearance area to extract features such as age and gender, which are used to construct pedestrian feature vectors. The system can generate a unique identifier for each identified pedestrian to continuously track the same pedestrian within a short period and ensure the stability and continuity of the feature extraction process.
[0030] Step S3: Based on the position, direction of movement and speed of each pedestrian, calculate the estimated arrival time of each pedestrian entering the effective display area of the LED window, and identify the pedestrians who are expected to enter the effective display area within the content switching prediction window as the target group. In the application, the effective display area can be defined by the projection surface in front of the shop window, used to determine when pedestrians actually enter a position where they can clearly view the content. The system predicts the estimated arrival time based on factors such as the distance between the pedestrian and the display area, whether their walking direction is towards the display area, and their walking speed. When the predicted time falls within the content switching window, the pedestrian is identified as a potential target for viewing the content in the next cycle. This filtering method ignores irrelevant pedestrians, such as those far from the shop window or walking in the opposite direction, thereby reducing computational resource consumption and improving the accuracy of the content matching logic.
[0031] Step S4: Based on the pedestrian feature vectors of each pedestrian in the target crowd and their corresponding weights, perform weighted fusion of the pedestrian feature vectors to obtain the target pedestrian feature vectors used for content prediction. In one embodiment, the system first constructs a basic feature vector based on the pedestrian's gender and age, and then calculates corresponding weights according to the gender and age group proportions within the target group. Subsequently, the system multiplies each pedestrian's feature vector by its corresponding weight and then fuses them to generate a target pedestrian feature vector representing the overall preferences of the target group. This fusion result reflects the common characteristics of the target group in terms of gender and age, avoiding bias caused by the features of individual pedestrians, and making the predicted content more closely aligned with the preferences of most potential viewers.
[0032] Step S5: Based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-constructed display content, predict the target display content for the next display cycle; During implementation, features can be extracted from each piece of content before system deployment to form content feature vectors. For example, high-dimensional vectors can be generated based on the content's topic type and target audience attribute tags. During system runtime, the target pedestrian feature vector is sequentially compared with each content vector to calculate similarity. The content with the highest similarity is the one most likely to interest the current audience. This matching method allows the system to automatically select the most suitable content from the content pool without manual intervention, making it particularly suitable for scenarios with rich and frequently changing content.
[0033] Step S6: After the current content is finished playing, output the predicted target content to be displayed. In one implementation, the system prepares the predicted next content before the current content finishes playing and switches content the instant playback ends to maintain display continuity. If the current content has not finished playing but new pedestrian prediction data appears and the estimated arrival time is short, the system can also initiate content switching in advance according to preset rules to ensure that the content is more attractive to the actual audience. Through this playback logic, the system can significantly improve the real-time responsiveness of the shop window and make the display effect more interactive.
[0034] In some embodiments, during the process of identifying the target audience, the relative positions of each pedestrian and the effective display area are considered. and the speed of movement of each pedestrian The angle between the direction of movement and the direction of the line connecting the effective display area ,according to Calculate the estimated arrival time of each pedestrian, when If a pedestrian is determined to have crossed the LED window, they are removed from the list. The remaining pedestrians with a positive estimated arrival time are considered as the target group.
[0035] In practice, the system calculates the relative coordinates between pedestrians and the effective display area in real time, acquiring changes in the horizontal and vertical distances of pedestrians through continuous frame positioning. After measuring speed and direction of travel, the system can quickly estimate the pedestrian's movement trend towards the window, thus determining whether they will enter the effective display area in a short time. If the calculation result is negative, it indicates that the pedestrian is already in front of the window but is leaving; such targets are not meaningful for display. Through this predictive mechanism, the system can effectively filter meaningless pedestrian data, significantly improving resource utilization.
[0036] In some embodiments, target data of multiple pedestrians is acquired, and feature extraction is performed on the target data to obtain pedestrian feature vectors for each pedestrian, including: Image data of pedestrians is collected, and age and gender features of pedestrians are extracted from the image data based on deep learning feature extraction algorithms to construct pedestrian feature vectors.
[0037] During implementation, the system continuously captures video data using high-definition cameras deployed above or on both sides of the shop window, and locates the position area of each pedestrian using a real-time pedestrian detection algorithm. Based on this, a trained age recognition network and gender recognition network are used to analyze the features of this area, thereby obtaining age classification results and gender labels. The system encodes age and gender into numerical form and combines them into a pedestrian feature vector to provide the basic input for subsequent weighted fusion. To improve recognition accuracy, the system can employ a multi-frame feature fusion strategy, that is, statistically evaluating the recognition results of several consecutive frames to reduce the impact of occasional recognition errors and ensure the stability of the pedestrian feature vector.
[0038] In some embodiments, the weight of each pedestrian in the target population is determined sequentially by gender discrimination and age group discrimination; Gender identification is based on the proportion of pedestrians of different genders in the target population, selecting the gender that matches the gender bias of the displayed content, and using that gender as the target gender. After determining the target gender, pedestrians within that gender are divided into preset age ranges, and a corresponding consumption coefficient is assigned to each age group. Age weighting under target gender is based on The calculation yielded, where The number of rows in the target gender that fall within the j-th age range; The target pedestrian feature vector is determined based on the target's gender and age weights.
[0039] In one implementation, the system first calculates the ratio of males to females within the target audience and, based on the gender bias of the displayed content, prioritizes the gender that aligns with the target audience. Then, for pedestrians within that gender group, the system divides them into multiple preset age ranges, such as children, youth, middle-aged, or other suitable ranges. Each age group is pre-configured with different consumption weight parameters, representing the potential interest level of that age group in specific content types. The system combines the number of pedestrians falling into the corresponding age group with the consumption parameters to obtain the overall age preference weight for that gender. Finally, based on the gender selection and age group calculation results, it generates a target pedestrian feature vector that accurately reflects the dominant audience characteristics of the content they are likely to view.
[0040] In some embodiments, predicting the target display content for the next display cycle based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-constructed display content includes: Let the feature vector of the target pedestrian be denoted as Let the content feature vector of the kth displayed content be denoted as ,according to Calculate the matching degree between the feature vector of the target pedestrian and the feature vector of each displayed content, and select the displayed content with the highest matching degree as the target displayed content for the next display cycle.
[0041] In implementation, the system pre-constructs a set of content feature vectors for each displayed content. This is achieved by encoding the content's style attributes, topic category, and target audience tags to form a structured vector. During runtime, the system sequentially evaluates the similarity between the target pedestrian feature vector and the content vector, improving the relevance between the content and the audience. Higher similarity indicates a closer match between the displayed content and the current target audience's focus, increasing the chance of attracting viewers to stay. Therefore, the system automatically selects the content with the highest similarity as the content for the next playback cycle. This decision-making process requires no manual intervention and can be dynamically completed before the end of each playback, ensuring a high degree of adaptability in the showcase display.
[0042] In some embodiments, after outputting the target display content for the next display cycle, the viewing time of each target pedestrian during the viewing process is recorded, and a corresponding consumption coefficient is set for each age group based on the viewing time and the corresponding age of the target pedestrian. It is dynamically updated.
[0043] In a practical system, viewing time can be determined by continuously tracking the dwell time of target pedestrians within the effective display area. The system determines whether a pedestrian is viewing the window content based on changes in their position and facing direction, and starts timing from that moment until they leave the effective display area. The recorded viewing time reflects the true appeal of the displayed content to pedestrians of different age groups. The system dynamically adjusts the age-group consumption coefficient based on the proportion of viewing time for each age group, gradually bringing the consumption coefficient closer to actual user preferences. Through this feedback-driven update mechanism, the system can continuously optimize the age preference weight, making future content predictions more aligned with the actual needs of the target audience.
[0044] In some embodiments, the consumption coefficient for each age group is based on the viewing duration and the age of the corresponding target pedestrian. Dynamic updates include: Let the sum of the viewing time of the target pedestrians in the j-th age group be denoted as . The sum of the viewing time of all target pedestrians of all age groups is recorded as The percentage of viewing time for the j-th age group is calculated as follows:
[0045] Calculate the average percentage of viewing time across all age groups. The consumption coefficient for each age group is calculated using the following formula. Update:
[0046] in, For the updated consumption coefficient, This is the preset update step size parameter.
[0047] In practical applications, after the displayed content has finished playing, the system will statistically analyze the viewing time for all age groups and calculate the percentage of each age group's total viewing time. If the viewing time percentage for a certain age group is higher than the average, it indicates that that age group prefers the currently displayed content, and the system will appropriately increase its corresponding consumption coefficient. Conversely, if the viewing time percentage for a certain age group is lower, its consumption coefficient will be decreased accordingly, making the overall preference prediction closer to real user behavior. This process is adjusted through a step size parameter to avoid drastic updates that could cause model fluctuations. With continuous operation, the model's expression of user preferences will become increasingly accurate, gradually forming a stable content recommendation capability.
[0048] In some embodiments, this application provides an LED shop window content control system for implementing an LED shop window content control method, the system comprising: The pedestrian target data acquisition unit is used to acquire target data of multiple pedestrians located in the target area in front of the LED window, extract the age and gender characteristics of pedestrians based on pedestrian image data, and obtain the pedestrian feature vector, position, direction of movement and speed of each pedestrian; In one possible implementation, the pedestrian target data acquisition unit may include a camera acquisition component and an image processing component. The camera acquisition component is positioned above or to the sides of the shop window for continuous imaging of the target area. The image processing component detects pedestrian positions from video frames using real-time video analysis algorithms and estimates the pedestrian's direction and speed of movement by combining coordinate changes across several frames. Subsequently, based on age and gender recognition models, age and gender features are extracted from the pedestrian's head image area to form a pedestrian feature vector. The system can assign a unique pedestrian number to each pedestrian entering the target area for continuous tracking, preventing recognition interruptions due to brief occlusion or posture changes. This unit ensures that the system can obtain stable, continuous, and predictable basic crowd data.
[0049] The estimated arrival time calculation unit is used to calculate the estimated arrival time of each pedestrian entering the effective display area based on the relative position, moving speed, and angle between the moving direction and the direction of the line connecting to the effective display area. Pedestrians with negative estimated arrival times are eliminated to determine the target group. During implementation, this unit first obtains coordinate information from the pedestrian target data acquisition unit, calculates the spatial distance between the pedestrian and the effective display area in real time, and analyzes whether the pedestrian's direction of movement is towards the display area. If the pedestrian's movement trend significantly deviates from the display area, the system can automatically determine that they will not become potential viewers. For pedestrians moving towards the display area, the system further estimates their estimated arrival time based on their movement speed. If a pedestrian's estimated arrival time is less than zero, it means that they are near the display area but are leaving, and this pedestrian will not be included in the prediction targets, thus making subsequent content decisions more targeted. The introduction of this unit can effectively avoid invalid targets affecting the prediction logic and improve the accuracy of content recommendation.
[0050] The target pedestrian feature vector generation unit is used to generate the target pedestrian feature vector by weighted fusion of the pedestrian feature vectors of each pedestrian in the target crowd and their corresponding weights determined by gender and age group in turn. In one embodiment, the unit first statistically analyzes the gender distribution within the target audience, selects the target gender based on the content's suitability attributes, and uses this gender as the first-level weighting criterion. Subsequently, the unit categorizes pedestrians of the selected gender into preset age groups, sets corresponding parameters based on the consumption preferences of each age group, and assigns age weights based on the proportion of people in each group. Finally, the age and gender characteristics of the pedestrians are weighted according to the aforementioned weights to form a target feature vector representing the overall audience's preferences. This vector accurately describes the comprehensive profile of the people about to enter the effective display area, making content prediction more closely aligned with the actual audience structure.
[0051] The content prediction unit is used to predict the target display content for the next display cycle based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-built display content. In practical implementation, during the deployment phase, the system extracts content attributes for each displayed content, such as content theme, target age group, suitable gender, or other related tags, and encodes them into content feature vectors. At runtime, the content prediction unit performs similarity matching between the target pedestrian feature vector and each content feature vector, selecting the content with the highest similarity as the content to be played in the next cycle. This unit supports different dimensions of content feature combinations and can be flexibly expanded according to the product type, audience trends, and brand positioning in the actual scenario, achieving configurability and diversity of content recommendation logic.
[0052] The content output unit is used to output the predicted target content for the next display cycle after the current content has finished playing. In one implementation, the unit interfaces with the playback control components of the shop window LED display. While the current content is playing, the unit preloads the media resources for the next predicted content and seamlessly switches at the end of the playback cycle. If the content prediction result is updated during playback due to the entry of new pedestrians, the unit can also adjust the displayed content in advance according to a preset switching strategy. This strategy can be set to strictly play the entire cycle, allow mid-play switching, or delay replacement, depending on the scenario. This unit ensures the timeliness, continuity, and stability of the displayed content output, enhancing the overall interactive experience.
[0053] The viewing time recording unit is used to record the viewing time of the target pedestrian during the viewing process after the target display content of the next display cycle is output; In one embodiment, the unit continuously tracks target pedestrians entering the effective display area, using positional changes to determine if they are still viewing. If a pedestrian faces the display area and remains stationary, it is considered an actual viewing action. The system times each pedestrian individually, recording the complete duration from the start of viewing to leaving the display area. To ensure accuracy, the unit can also introduce a temporary occlusion compensation mechanism to prevent timing interruptions when a pedestrian is partially obscured. By continuously accumulating viewing data, the system can identify the preference trends of different groups for the displayed content, providing a solid behavioral data foundation for subsequent parameter updates.
[0054] The consumption coefficient update unit is used to dynamically update the consumption coefficient set for each age group based on the viewing time and the age of the corresponding target pedestrian.
[0055] In one implementation, the unit aggregates all viewing data at the end of each display cycle, calculates the total viewing time of pedestrians in different age groups, and determines the proportion of each age group. If a certain age group shows a higher viewing proportion in this round of display, the unit appropriately increases the consumption parameters corresponding to that age group, giving it a greater weight in the next round of prediction; conversely, it decreases its weight. This update process, through gradual adjustments, allows the system to continuously learn from real viewing behavior, making the content recommendation strategy more accurate and better suited to the ever-changing demographic structure, ultimately achieving self-evolution of content recommendation capabilities.
[0056] In other alternative embodiments of the present invention, the LED window display content control system can also incorporate a variety of auxiliary functions to further enhance the system's integrity, stability, and commercial deployment capabilities. These functions are not essential for achieving pedestrian prediction, feature vector fusion, content matching, or preference updates, but can serve as extensions of the system in practical applications to enhance user experience and system operation and maintenance efficiency.
[0057] In one optional implementation, the system can be configured with an environmental sensing unit to collect environmental parameters such as light intensity, temperature, humidity, and noise in the area where the display window is located. The system can automatically adjust the brightness of the LED display screen, the speaker output volume, and the operating status of the cooling fans based on the collected data. For example, it automatically increases screen brightness to ensure visibility when ambient light is high, and reduces brightness at night or in low-light environments to reduce energy consumption; it increases volume when there are many people and high ambient noise, and automatically reduces volume in low-noise environments; it automatically starts the fan to dissipate heat when the temperature inside the display window rises, thereby ensuring long-term stable operation of the equipment. Through these environmental adaptive control strategies, the system can maintain optimal display effects and operating efficiency under different environmental conditions.
[0058] In another optional implementation, the system can support remote management and centralized operation and maintenance. Users can manage multiple terminals uniformly through a cloud management platform, including viewing the real-time status of each terminal, updating content playlists, adjusting brightness and volume, managing device configurations, and performing remote restarts or maintenance. This centralized management approach, especially in multi-store scenarios, can significantly reduce the workload of manual maintenance and improve content update efficiency. When the system detects an anomaly in a terminal, such as abnormal brightness, excessive temperature, or network offline, it can also generate alarm information and push it to administrators for timely handling, improving the overall operational stability of the system.
[0059] In some embodiments, the system can further support data analysis and report generation functions to statistically analyze system operating status, content playback status, changes in traffic, and interaction data. The system can automatically generate data reports daily, weekly, or monthly, including information such as content exposure frequency, changes in target audience composition, statistics on viewing time of displayed content, and device operating status, providing quantitative data support for operators. Through visualized statistical charts, operators can intuitively understand the effectiveness of content delivery, trends in showcase attractiveness, and device health, thereby better guiding subsequent content strategies and device management plans.
[0060] The system can also optionally include voice interaction capabilities. When a pedestrian approaches the shop window and utters a preset wake-up phrase, the system can analyze the pedestrian's voice content using voice recognition technology to provide services such as product information inquiry, content explanation, or interactive Q&A. For example, a pedestrian can ask about the price, material, or applicable scenarios of the displayed products. The system can then retrieve relevant information from its database by parsing the voice command and immediately display the information on the screen or speaker. Voice interaction can enhance the interactivity and fun of the shop window experience, but it is a common technology approach in the industry and therefore appears as an optional module.
[0061] At the system architecture level, the system optionally supports basic capabilities such as wireless communication modules, cloud inference services, online databases, and log storage to enable remote data storage, cloud analysis, online parameter adjustment, and system upgrades. The system can also employ various communication methods, including 4G, Wi-Fi, or Ethernet connections, to ensure stable data exchange between the terminal and the cloud platform in different deployment environments. These features contribute to improving the system's scalability and long-term maintenance capabilities, but are not considered the core inventive content of this invention.
[0062] In some embodiments, this application provides a terminal, including: The memory is used to store the LED window display content control program; A processor is used to implement the steps of the LED window content control method when executing the LED window content control system.
[0063] In some embodiments, this application provides a computer-readable storage medium that stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer executes the LED window content control method.
[0064] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.
Claims
1. A method for controlling the content of an LED shop window, characterized in that, Includes the following steps: Get the remaining playback time of the currently playing content. When the remaining playback time falls into a preset time threshold, start the prediction process for the next content to be displayed. Acquire target data of multiple pedestrians located in the target area in front of the LED shop window, extract features from the target data, and obtain the pedestrian feature vector, position, direction of movement, and speed of movement for each pedestrian; Based on the location, direction of movement and speed of each pedestrian, the estimated arrival time of each pedestrian entering the effective display area of the LED window is calculated, and pedestrians who are expected to enter the effective display area within the content switching prediction window are identified as the target group. Obtain target data for multiple pedestrians, extract features from the target data, and obtain pedestrian feature vectors for each pedestrian, including: Collect pedestrian image data, and extract age and gender features of pedestrians from the image data based on deep learning feature extraction algorithms to construct pedestrian feature vectors; The weight of each pedestrian in the target population is determined sequentially by gender and age group. Gender identification is based on the proportion of pedestrians of different genders in the target population, selecting the gender that matches the gender bias of the displayed content, and using that gender as the target gender. The location area of each pedestrian is determined by a real-time pedestrian detection algorithm; The region is analyzed using trained age and gender recognition networks to obtain age classification results and gender labels. Age and gender are encoded into numerical form and combined into a pedestrian feature vector; After determining the target gender, pedestrians within that gender are divided into preset age ranges, and a corresponding consumption coefficient is assigned to each age group. Age weighting under target gender is based on The calculation yielded, where The number of rows in the target gender that fall within the j-th age range; The feature vector of the target pedestrian is determined based on the weights of the target's gender and age; Based on the pedestrian feature vectors of each pedestrian in the target population and their corresponding weights, the pedestrian feature vectors are weighted and fused to obtain the target pedestrian feature vectors for content prediction. Based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-built display content, the target display content for the next display cycle is predicted; After the currently displayed content finishes playing, output the predicted target display content; After outputting the target display content for the next display cycle, the viewing time of each target pedestrian is recorded. Based on the viewing time and the corresponding age of the target pedestrian, a corresponding consumption coefficient is set for each age group. Perform dynamic updates; Consumption coefficients for each age group based on viewing time and the corresponding age of the target pedestrians. Dynamic updates include: Let the sum of the viewing time of the target pedestrians in the j-th age group be denoted as . The sum of the viewing time of all target pedestrians of all age groups is recorded as The percentage of viewing time for the j-th age group is calculated as follows: Calculate the average percentage of viewing time across all age groups. The consumption coefficient for each age group is calculated using the following formula. Update: in, For the updated consumption coefficient, This is the preset update step size parameter.
2. The LED shop window content control method according to claim 1, characterized in that, In identifying the target audience, the relative positions of each pedestrian and the effective display area are considered. and the speed of movement of each pedestrian The angle between the direction of movement and the direction of the line connecting the effective display area ,according to Calculate the estimated arrival time of each pedestrian, when If a pedestrian is determined to have crossed the LED window, they are removed from the list. The remaining pedestrians with a positive estimated arrival time are considered as the target group.
3. The LED shop window content control method according to claim 1, characterized in that, Based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-constructed display content, the target display content for the next display cycle is predicted, including: Let the feature vector of the target pedestrian be denoted as Let the content feature vector of the kth displayed content be denoted as ,according to Calculate the matching degree between the feature vector of the target pedestrian and the feature vector of each displayed content, and select the displayed content with the highest matching degree as the target displayed content for the next display cycle.
4. An LED shop window content control system, used to implement the LED shop window content control method as described in claim 1, characterized in that, The system includes: The pedestrian target data acquisition unit is used to acquire target data of multiple pedestrians located in the target area in front of the LED window, extract the age and gender characteristics of pedestrians based on pedestrian image data, and obtain the pedestrian feature vector, position, direction of movement and speed of each pedestrian; The estimated arrival time calculation unit is used to calculate the estimated arrival time of each pedestrian entering the effective display area based on the relative position, moving speed, and angle between the moving direction and the direction of the line connecting to the effective display area. Pedestrians with negative estimated arrival times are eliminated to determine the target group. The target pedestrian feature vector generation unit is used to generate the target pedestrian feature vector by weighted fusion of the pedestrian feature vectors of each pedestrian in the target crowd and their corresponding weights determined by gender and age group in turn. The content prediction unit is used to predict the target display content for the next display cycle based on the matching degree between the target pedestrian feature vector and the content feature vector of each pre-built display content. The content output unit is used to output the predicted target content for the next display cycle after the current content has finished playing. The viewing time recording unit is used to record the viewing time of the target pedestrian during the viewing process after the target display content of the next display cycle is output; The consumption coefficient update unit is used to dynamically update the consumption coefficient set for each age group based on the viewing time and the age of the corresponding target pedestrian.
5. A terminal, characterized in that, include: The memory is used to store the LED window display content control program; A processor is configured to implement the steps of the LED window content control method as described in claim 1 when executing the LED window content control device.
6. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions. When the computer reads the computer instructions from the storage medium, the computer executes the LED window content control method as described in claim 1.