Outdoor intelligent art display system based on multimedia interaction

Abstract

The invention discloses an outdoor intelligent art display system based on multimedia interaction, and relates to the technical field of digital art display and intelligent control, and the system comprises an audience behavior collection module which is used for collecting audience staying time and behavior response in real time, obtaining a content attraction initial evaluation result, calculating an attraction index of each content unit, and outputting the attraction index of each content unit; the content analysis and screening module is used for extracting alternative multimedia content units from a preset content library, judging the theme relevance between alternative contents and the currently displayed contents, and obtaining a candidate content set meeting replacement conditions; according to the outdoor intelligent art display system based on multimedia interaction, intelligent replacement and updating of multimedia content are realized on the premise of not interrupting display, content combination is automatically optimized through audience feedback data, content with low attraction is eliminated, and welcome elements are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of digital art display and intelligent control, and particularly to an outdoor intelligent art display system based on multimedia interaction. Background Art

[0002] As a frontier field of the integration of digital art and public space, the outdoor intelligent art display system is increasingly becoming a key technical carrier for enhancing urban cultural experience and strengthening audience interaction. The importance of this field lies in that it not only promotes the innovation of art expression forms, but also realizes the dynamic dissemination and personalized presentation of cultural content through technical means, giving new vitality to urban public spaces.

[0003] However, existing methods are still insufficient in realizing the intelligent replacement and update of multimedia content. Traditional systems mostly rely on manual intervention or preset programs. When updating content, it is often necessary to pause the display, resulting in the interruption of the audience experience. At the same time, they lack the flexibility to dynamically adjust content according to real-time feedback. These limitations make the system unable to fully meet the diverse needs of the audience, and also limit the attractiveness and sustainability of art displays. The defects of current solutions are mainly reflected in the inefficiency of the update process and the lack of adaptability. Specifically, how to seamlessly replace multimedia content without interrupting the display, and how to intelligently adjust the content combination according to the audience feedback data have become the core challenges in technical implementation. First of all, seamless update involves the balance between real-time content switching and system stability. If not handled properly, it may lead to technical problems such as display interruption or content disorder. Secondly, intelligent adjustment based on audience feedback requires the linkage of accurate data analysis and content optimization. Existing systems often have difficulty accurately identifying content with low attractiveness and quickly strengthening popular elements, resulting in insufficient pertinence of content update and even deviation from audience preferences. These unsolved technical factors not only affect the coherence of the display effect, but also limit the system's adaptability to dynamic environments, thus giving rise to unique implementation difficulties. Summary of the Invention

[0004] The purpose of the present invention is to provide an outdoor intelligent art display system based on multimedia interaction, which can realize the intelligent replacement and update of multimedia content without interrupting the display, and at the same time automatically optimize the content combination through audience feedback data, eliminate content with low attractiveness and strengthen popular elements.

[0005] To achieve the above object, the present invention provides the following technical solution: An outdoor intelligent art display system based on multimedia interaction, comprising:

[0006] An audience behavior collection module, which is used to collect the audience's stay time and behavioral reactions in real time, obtain the initial evaluation result of content attractiveness, and calculate the attractiveness index of each content unit;

[0007] A content analysis and screening module, which is used to extract alternative multimedia content units from a preset content library, judge the relevance between the alternative content and the theme of the currently displayed content, and obtain a set of candidate content that meets the replacement conditions;

[0008] A content scheduling module, which is used to sort the set of candidate content when the replacement condition is met, transmit the candidate content with the highest priority to the display terminal, and determine the replacement sequence and switching time point;

[0009] A switching control module, which is used to seamlessly connect the currently displayed content with the candidate content, judge whether the system load exceeds a preset threshold during the switching process, and obtain a stable switching execution plan;

[0010] A rendering optimization module, which is used to dynamically adjust the rendering parameters of the display terminal, extract candidate content from the buffer, and obtain an updated display content combination;

[0011] A feedback evaluation module, which is used to collect a new round of audience feedback data, compare the change trends of the attraction indexes before and after, and judge whether the content optimization reaches the expected threshold;

[0012] An adaptive optimization module, which is used to extract external variables from the dynamic environment if the change trend does not reach the expected threshold, recalculate the attraction index, and obtain the adjusted content optimization direction and replacement sequence;

[0013] A content loop control module, which is used to repeatedly execute content replacement and parameter adjustment by using a real-time switching algorithm and a buffering mechanism to obtain a final self-adaptive optimized display content combination.

[0014] Preferably, the audience behavior collection module collects the audience's stay time and behavioral responses in real time to obtain an initial evaluation result of content attraction. Calculating the attraction index of each content unit includes:

[0015] Collect the audience's stay time and behavioral response data in real time through sensors and interaction devices to obtain the display status and feedback information of the multimedia content, extract the stay time and behavioral response characteristics from the collected sensor data, obtain the audience interaction data of each content unit, use data processing methods to quantitatively analyze the stay time and behavioral responses, calculate the attraction index of each content unit. If the attraction index is lower than the preset threshold, identify the abnormal presentation of the content unit through the display status, determine the adjustment direction, optimize the display status of the content unit according to the adjustment direction to obtain updated multimedia content data, collect the audience feedback information again through the updated multimedia content data, judge the change trend of the attraction index, and use the random forest algorithm to predict and analyze the change trend to obtain the long-term attraction evaluation result of each content unit.

[0016] Preferably, the content analysis and screening module extracts alternative multimedia content units from a preset content library, judges the relevance of the alternative content to the theme of the currently displayed content, and the candidate content set that meets the replacement conditions includes:

[0017] Obtain multimedia data from a preset content library, use content extraction technology to separate alternative units to obtain an initial list of alternative content. For the initial list of alternative content, calculate the attraction index, sort through the index distribution to obtain a sorted sequence of alternative content. If there is a theme association in the sorted sequence of alternative content that matches the currently displayed content, retain the relevant alternative units to obtain a subset of theme-related content. According to the replacement conditions, judge whether the alternative units in the subset of theme-related content meet the requirements to obtain a preliminary candidate set. Process the preliminary candidate set through a clustering algorithm, merge similar alternative units to obtain an optimized candidate content set. Obtain the optimized candidate content set, conduct a secondary verification on the theme relevance to determine the final candidate content set, use the final candidate content set, adjust the sorting according to the attraction index distribution, and output a candidate content sequence that meets the requirements.

[0018] Preferably, when the replacement conditions of the content scheduling module are met, sort the candidate content set, and transmit the candidate content with the highest priority to the display terminal. The determination of the replacement sequence and the switching time point includes:

[0019] Sort the candidate content set through a real-time switching algorithm to obtain a candidate content sequence, use preloading technology to load the candidate content. When the conditions are met, obtain the transmission data from the preloading technology and transmit it to the display terminal. If the transmission data reaches the display terminal, determine the replacement sequence through the switching time point. According to the replacement sequence, obtain the currently displayed content from the display terminal to determine the switching time point. If the switching time point arrives, update the content of the display terminal through the real-time switching algorithm, obtain the updated content from the display terminal, and judge whether the candidate content has been transmitted completely.

[0020] Preferably, the switching control module seamlessly connects the currently displayed content with the candidate content, and judges whether the system load exceeds a preset threshold during the switching process to obtain a stable switching execution plan, including:

[0021] Obtain the replacement sequence, process the connection between the displayed content and the candidate content through the content buffering mechanism, determine whether the switch triggers a load change, and obtain the preliminary switch data. Analyze the preliminary switch data through the buffering mechanism, judge whether the system load exceeds the preset threshold, and determine the load status. If the load status exceeds the preset threshold, adjust the replacement sequence to obtain an optimized candidate content connection plan. According to the optimized candidate content connection plan, use the load judgment result to determine the smooth transition parameters during the switch. Process the switch process through the smooth transition parameters to obtain the real-time monitoring data of the system load. According to the real-time monitoring data, use the stable scheme generation algorithm to obtain the final execution plan. Judge the smoothness of the switch process through the final execution plan to obtain a stable switch execution result.

[0022] Preferably, the rendering optimization module dynamically adjusts the rendering parameters of the display terminal, extracts candidate content from the buffer, and obtains the updated display content combination, including:

[0023] Obtain candidate content from the buffer, use a preset threshold to judge low-attraction content, and obtain the content set to be replaced. For the content set to be replaced, dynamically adjust the rendering parameters. By comparing the characteristics of the candidate content and the low-attraction content, determine the replacement priority. If the replacement priority is higher than the preset threshold, extract high-attraction segments from the candidate content and gradually replace the low-attraction content to obtain the preliminarily adjusted display content. According to the preliminarily adjusted display content, obtain the current state of the display terminal, judge whether the rendering parameters match, and obtain the optimized parameter configuration. Through the optimized parameter configuration, adjust the display content combination, use the K-means algorithm to cluster and analyze the content characteristics, and determine the distribution of the final display content. For the distribution of the final display content, extract supplementary content from the buffer. If the supplementary content matches the display content combination, perform fusion to obtain the updated display content combination. According to the updated display content combination, judge the load condition of the display terminal, and adjust the rendering parameters through the load balancing strategy to obtain a stable output display plan.

[0024] Preferably, the feedback evaluation module collects a new round of audience feedback data, compares the change trends of the attraction indices before and after, and judges whether the content optimization reaches the expected threshold, including:

[0025] Obtain the updated display content, extract a new round of audience feedback data through data collection technology to obtain a feedback data set, extract the attraction index from the feedback data set, use an iterative algorithm to calculate the change in the index before and after to obtain the change trend data. For the change trend data, obtain a preset threshold, and use a comparison algorithm to determine whether the change trend reaches the preset threshold to obtain a preliminary judgment result. If the preliminary judgment result does not reach the preset threshold, adjust the display content to generate an optimized content combination. Through the optimized content combination, collect a new round of feedback data to obtain an updated feedback data set, extract the attraction index from the updated feedback data set, use an iterative algorithm to calculate the new change trend to obtain the optimized trend data, and for the optimized trend data, use a comparison algorithm to judge with the preset threshold to obtain the final judgment result.

[0026] Preferably, if the change trend does not reach the expected threshold, the adaptive optimization module extracts external variables from the dynamic environment and recalculates the attraction index. The obtained adjusted content optimization direction and replacement sequence include:

[0027] If the change trend does not reach the preset threshold, the environmental extraction module obtains external variables from the dynamic environment to obtain a preliminary data set. Through the preliminary data set, the data analysis module calculates the attraction index to obtain the initial index value. If the initial index value is lower than the preset threshold, the external variables are adjusted through regression analysis to obtain an optimized variable set. According to the optimized variable set, the attraction index is recalculated to obtain the adjusted index value. Through the adjusted index value, content adjustment parameters are generated to obtain an optimization direction sequence. According to the optimization direction sequence, a replacement sequence is generated using a sorting algorithm to obtain the final content adjustment plan. Through the final content adjustment plan, the variable configuration in the dynamic environment is updated to obtain new change trend data.

[0028] Preferably, the content loop control module uses a real-time switching algorithm and a buffering mechanism to repeatedly execute content replacement and parameter adjustment. The obtained final self-adaptive optimized display content combination includes:

[0029] Obtain the initial replacement sequence and display content through the real-time switching algorithm to obtain a preliminary content combination, store the preliminary content combination using the buffering mechanism, and determine whether the self-adaptability requirement is met. If the self-adaptability requirement is not met, execute the adjustment logic for the replacement sequence to obtain an updated sequence.

[0030] Preferably, the content loop control module uses a real-time switching algorithm and a buffering mechanism to repeatedly execute content replacement and parameter adjustment. The obtained final self-adaptive optimized display content combination further includes:

[0031] By repeatedly performing content replacement through an updated sequence, obtaining optimized display content, processing the optimized display content using a parameter optimization algorithm, determining the final combination form, performing dynamic switching according to the final combination form to obtain self-adaptive optimized display content, and verifying the self-adaptive optimized display content through a buffering mechanism to determine whether the expected standard is met.

[0032] As can be seen from the above technical solutions, the present invention has the following beneficial effects:

[0033] The outdoor intelligent art display system based on multimedia interaction evaluates the content attractiveness by real-time collecting the audience behavior data, selects candidate content related to the theme from a preset content library, performs seamless replacement using a real-time switching algorithm and a buffering mechanism, dynamically adjusts the display parameters and gradually updates the low-attractiveness content to form a new display combination. The present invention repeatedly compares the change trends of the attractiveness before and after through iterative cycles, recomputes the optimization direction in combination with external environmental variables, repeatedly performs content replacement and parameter adjustment, and finally obtains a self-adaptive optimized display content combination. This method can intelligently optimize the multimedia display content according to the real-time feedback of the audience, improve the content attractiveness and display effect, realize the dynamic adjustment and personalized recommendation of the display content, enhance the audience interaction experience, and improve the information dissemination effect. Brief Description of the Drawings

[0034] Figure 1 It is a connection diagram of the system modules of the present invention. Detailed Embodiments

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0036] As Figure 1 shown, the present invention provides a technical solution: an outdoor intelligent art display system based on multimedia interaction, including:

[0037] An audience behavior collection module, configured to collect the audience stay time and behavior responses in real time, obtain an initial evaluation result of the content attractiveness, and calculate the attractiveness index of each content unit;

[0038] A content analysis and screening module, configured to extract alternative multimedia content units from a preset content library, determine the relevance between the alternative content and the current display content theme, and obtain a set of candidate content that meets the replacement conditions;

[0039] A content scheduling module, which is used to sort the candidate content set when the replacement condition is met, transmit the candidate content with the highest priority to the display terminal, and determine the replacement sequence and switching time point;

[0040] A switching control module, which is used to seamlessly connect the current displayed content with the candidate content, judge whether the system load exceeds the preset threshold during the switching process, and obtain a stable switching execution plan;

[0041] A rendering optimization module, which is used to dynamically adjust the rendering parameters of the display terminal, extract the candidate content from the buffer, and obtain an updated display content combination;

[0042] A feedback evaluation module, which is used to collect a new round of audience feedback data, compare the change trends of the attraction indexes before and after, and judge whether the content optimization reaches the expected threshold;

[0043] An adaptive optimization module, which is used to extract external variables from the dynamic environment and recalculate the attraction index if the change trend does not reach the expected threshold, and obtain the adjusted content optimization direction and replacement sequence;

[0044] A content loop control module, which is used to repeatedly execute content replacement and parameter adjustment by using a real-time switching algorithm and a buffering mechanism to obtain a final self-adaptive optimized display content combination.

[0045] The present invention realizes the intelligent dynamic update of outdoor art display content by constructing a multi-level interactive mechanism. The audience behavior acquisition module includes a high-definition infrared camera and a visual recognition system, and uses a 120Hz frame rate video stream to capture the audience's stay trajectory, stay duration, facial expressions, etc. in the display area. The system extracts audience behavior characteristics in real time through a convolutional neural network (CNN) model, and calculates the behavioral response index value (for example: average gaze time ≥ 5 seconds is determined as high attention). The content attractiveness index is evaluated by a linear weighted model that comprehensively evaluates indicators such as the number of viewers, average stay time, and facial pleasure. The index range is set to 0-100, and a content update process is triggered when it is lower than 60. The content analysis and screening module uses a semantic matching model BERT to perform content theme similarity analysis, and screens content units with a match degree of more than 0.85 with the current display theme to enter the candidate set. The content scheduling module is sorted based on a priority scoring system, and the weights include historical attractiveness index, relevance score, and novelty score, and the optimal switching time point is determined in combination with the system clock and the crowd density fluctuation curve. The switching control module has a load monitoring logic unit built in, which analyzes parameters such as CPU occupancy (threshold is 85%) and GPU temperature (threshold is 80℃) in real time to avoid switching operations when system resources are tight. The rendering optimization module relies on the Vulkan API or OpenGL engine to dynamically adjust rendering parameters based on the display resolution (such as 1920×1080), refresh rate (60Hz or higher) and current cache ratio (the cache area keeps at least 30% content preloaded). The feedback evaluation module collects behavioral data again and uses a time window algorithm (such as Tumbling Window 10 seconds) to analyze the trend of audience behavior changes. If the attractiveness index after the update is less than 5% higher than before, the adaptive optimization module is started. This module combines external environmental information such as temperature and humidity, light intensity (obtained through integrated sensors), background noise (>70dB) and other variables, and uses the support vector regression (SVR) algorithm to predict the content optimization direction. Finally, the content cycle control module continuously iterates the strategy based on the reinforcement learning optimization model to achieve a sustainable self-evolving content scheduling system.

[0046] This method significantly improves the interactivity and relevance between the display content and the audience, can respond to audience behavior feedback in real time to update the content, and achieves accurate adaptation between content and scenes and audience emotions. The system dynamically adjusts the content while ensuring seamless switching and controllable resources, improving the operating efficiency and visual quality of the display terminal. At the same time, by introducing the joint modeling of environmental variables and behavioral data, the display content has stronger adaptability and intelligent adjustment capabilities, effectively improving the communication value of public art installations and audience experience satisfaction. In addition, the entire system architecture is scalable and can be widely used in various types of outdoor advertising, exhibition installations and urban landscape interaction systems.

[0047] Taking a large square in a certain city as an example, three multimedia intelligent display screens are installed in this square, located in areas with dense tourists, rest corridors and night lighting areas respectively. The system automatically switches the display theme every day according to the peak distribution of the number of people and weather conditions (for example, it displays the city's historical culture during the day and automatically switches to the light and shadow art performance at night). Once the system detects that the average fixation time is less than 3 seconds, it triggers a new content replacement process, loads high-definition content and completes the switch within 2 seconds, and at the same time adjusts the screen brightness to adapt to the night environment, achieving the effect of increasing the audience's stay rate and participation enthusiasm.

[0048] The audience behavior collection module collects the audience's stay time and behavioral responses in real time, obtains the initial evaluation result of the content attractiveness, and calculates the attractiveness index of each content unit, including:

[0049] The stay time and behavioral response data of the audience are collected in real time through sensors and interactive devices to obtain the display status and feedback information of the multimedia content. The stay time and behavioral response characteristics are extracted from the collected sensor data to obtain the audience interaction data of each content unit. Data processing methods are used to quantitatively analyze the stay time and behavioral responses, calculate the attractiveness index of each content unit. If the attractiveness index is lower than the preset threshold, the abnormal presentation of the content unit is identified through the display status, the adjustment direction is determined, and the display status of the content unit is optimized according to the adjustment direction to obtain the updated multimedia content data. The audience feedback information is collected again through the updated multimedia content data, the change trend of the attractiveness index is judged, and the random forest algorithm is used to predict and analyze the change trend to obtain the long-term attractiveness evaluation result of each content unit.

[0050] The core of the system lies in building a real-time data collection and evaluation mechanism based on audience behavior feedback. The audience behavior collection module consists of an infrared thermal imaging camera, an ultrasonic rangefinder, and a pressure-sensitive floor mat, which are used to accurately record the stay path and stay time of each audience in the display area. The behavior responses of the audience are obtained through micro-expression recognition (based on object detection algorithms) and limb movement recognition (such as skeletal point tracking). The system packs and uploads the data collected every second to the local edge processing unit. The system first calculates the attraction index for each display content unit. The attraction index consists of three parts, namely: the average stay time of the audience for this content, the positive score of the audience's facial emotions, and the action interaction score of the audience (such as clapping, waving, nodding, etc.). Each part of the index is converted into a standardized value. Among them, the average stay time is in seconds, and both the emotion score and the behavior score range from zero to one. The system calculates the attraction index of each content unit through weighted average. For example: the system sets the weight of the stay time to 50%, the weight of the facial emotion score to 30%, and the weight of the action behavior score to 20%. Then, multiply the actual value of each index by its corresponding weight and add the results to obtain the final attraction index value. The system sets the threshold of the attraction index to 60 points. If the attraction index of a certain content unit is lower than this threshold, it is considered that there is an abnormality in the display state of this content. At this time, the system will identify the display state and find possible abnormal reasons, such as unstable playback frame rate, insufficient picture brightness, or content switching failure. Then, determine the optimization direction according to the identification result, such as replacing static pictures with dynamic content, enhancing color saturation, or adding sound elements. After optimization, the system generates updated multimedia content and redeploys it to the display terminal. After the optimized content is launched, the system continues to collect a new round of audience feedback information and analyzes the change trend of the attraction index. To achieve long-term prediction, the system introduces a machine learning algorithm called "random forest", which judges whether the attraction of this content is sustainable by analyzing the feedback data of multiple previous and subsequent periods. If the system judges that the long-term attraction of a certain content unit is still low, it will trigger the content optimization or replacement program again and enter the next content cycle optimization process.

[0051] Through multi-sensor fusion technology and accurate data modeling, the system can efficiently evaluate the true attraction level of each display content and provide scientific decision support for content optimization. Using the random forest algorithm enhances the prediction accuracy and system adaptability, significantly improving the audience response quality and participation of the multimedia display system. The system has a closed-loop control structure, making the content adjustment process more real-time and adaptive, providing an intelligent and data-driven upgrade approach for outdoor interactive art displays.

[0052] An interactive installation square built during a certain international art festival. Each LED screen obtains interaction data of thousands of audiences through a thermal imaging camera and a behavior collection system. The system found that the average viewing time of a "digital landscape animation" was only 3.2 seconds, far lower than the average level of 5 seconds. The system detected that the cold color and lack of background music might be the reasons. Subsequently, the optimization direction was to enhance the color saturation and add background music. After the updated content was launched, the average viewing time increased to 6.8 seconds. Random forest analysis predicted that the content would remain at the top of the ranking in terms of continuous attraction rate during the festival. Finally, the content was selected as the "audience's favorite display unit" every day.

[0053] The content analysis and screening module extracts alternative multimedia content units from the preset content library, judges the relevance of the alternative content to the current display content theme, and obtains a set of candidate content that meets the replacement conditions, including:

[0054] Obtain multimedia data from the preset content library, use content extraction technology to separate alternative units, obtain an initial list of alternative content, calculate the attraction index for the initial list of alternative content, sort through the exponential distribution, obtain a sorted sequence of alternative content. If there is a theme association that matches the current display content in the sorted sequence of alternative content, retain the relevant alternative units to obtain a subset of theme-related content. According to the replacement conditions, judge whether the alternative units in the subset of theme-related content meet the requirements to obtain a preliminary candidate set. Process the preliminary candidate set through a clustering algorithm, merge similar alternative units to obtain an optimized candidate content set. Obtain the optimized candidate content set, conduct a secondary verification for theme relevance to determine the final candidate content set. Use the final candidate content set to adjust the sorting based on the attraction index distribution and output a candidate content sequence that meets the requirements.

[0055] The content analysis and screening module in this system mainly functions to intelligently screen out candidate content that is relevant to the current display theme and has high attraction from a large-capacity preset multimedia content library for the display system to replace. The operation logic of this module includes the following steps in sequence:

[0056] First, the system extracts all available multimedia data from the content library. This data may include various types such as images, videos, texts, and audios. The system will apply content parsing technology to separate each content item, split it into basic content units, and construct a preliminary content list, called the "initial list of alternative content".

[0057] Subsequently, the system calculates an "attractiveness index" for each content unit in the initial alternative list. This attractiveness index is calculated by weighting three key indicators: average dwell time, which is the average duration that viewers stayed when watching this content in the past, usually measured in seconds; facial emotion score, which is obtained by identifying viewers' facial expressions and quantifying them into emotion values ranging from 0 to 1, where 1 represents extremely high positive emotion; and behavior response score, such as whether viewers have operations like clicking, scanning codes, or liking, which is also standardized to be between 0 and 1. The system multiplies these three indicators by their corresponding weight values respectively, and then sums up the results to obtain the comprehensive attractiveness score. For example, if the system sets: the weight of average dwell time is 50%; the weight of emotion score is 30%; the weight of behavior score is 20%. And assume a content unit has the following indicators: average dwell time is 8 seconds; facial emotion score is 0.7; behavior score is 0.5. Then the calculation process of the attractiveness index is as follows: multiply 8 by 0.5 to get 4; multiply 0.7 by 0.3 to get 0.21; multiply 0.5 by 0.2 to get 0.10. Add the above three results together to get the final attractiveness index of 4.31. At this time, the system sorts all content units in descending order of the attractiveness index to form a sorted alternative content sequence. Then, the system enters the "theme matching judgment" link. The system generates a "theme vector" for the currently displayed content, and also generates its corresponding "theme vector" for each alternative content. Then, it uses a semantic comparison algorithm to calculate the similarity between the two vectors. The most commonly used method is the "cosine similarity algorithm", which outputs a value between 0 and 1. The closer it is to 1, the closer the themes are. The system generally sets a similarity threshold, such as 0.75. Only the content units that reach or exceed this value are retained to form a "theme-related content subset". Next, the system further filters the units in the "theme-related content subset" according to the "replacement conditions" set by the actual requirements or strategies of the display terminal. The conditions may include: whether the playback length meets the requirements, whether the resolution matches the terminal, whether the file size is within the system's bearable range, etc. The units that meet all conditions form a "preliminary candidate set". The system uses a "clustering algorithm" to merge the content in the preliminary candidate set. Taking the K-means algorithm as an example, the system will vectorize and encode the content features (such as color distribution, theme tags, emotion tone, etc.). By clustering similar content, removing redundancy, and retaining representative content, an "optimized candidate content set" is obtained. After that, the system will conduct a second round of semantic comparison between the content in this set and the currently displayed theme for "secondary verification" to ensure a high degree of theme matching, and finally generate a "final candidate content set". Finally, the system re-orders these contents based on the latest round of attractiveness index to form a "candidate content sequence" with a clear structure and predictable effects, as the candidate source for the next round of switching of the display system.Parameter Description and Determination Method: Residence Time (seconds): Automatically obtained through historical data collection, serving as an indirect measure of user attention intensity; Emotion Score (0 - 1): Scoring the audience's face through the facial expression recognition module, quantifying features such as the amplitude of smiling and the degree of eyebrow raising; Behavior Score (0 - 1): Normalized based on the frequency of the audience's operation behaviors (such as click-through rate, number of scans); Weight Setting: The optimal allocation can be obtained through machine learning training using actual research or historical audience data, and the initial values are usually set to 0.5, 0.3, 0.2; Similarity Threshold (such as 0.75): Set the best interval that can ensure the balance between content relevance and diversity through simulation experiments or A / B tests; Clustering Parameter (such as the value of K): Dynamically calculated or set according to the actual quantity and content distribution characteristics of the alternative content set.

[0058] This module effectively improves the pertinence of content switching and the matching degree with the audience response through a multi-stage content analysis and refined screening mechanism. By using structured feature extraction and clustering technologies, it realizes content diversity while reducing redundancy. The introduction of the attractiveness index and the dual constraint mechanism of the theme enhances the quality control of the candidate content, ensuring that each display switch is based on data support and driven by audience preferences. This module is highly modular, facilitating subsequent integrated deployment with various terminal systems.

[0059] In the intelligent commercial block of a certain city, the system is deployed on multiple touch interactive screens. Every morning, the system automatically screens the content related to the current solar term and the themes of surrounding stores from the background content library. By analyzing the audience interaction records of the previous day, the content with an average residence time exceeding 6 seconds is screened out, and the video materials with the theme tags containing "coffee" and "life aesthetics" are automatically included in the candidates. After system clustering and optimization, a group of high-definition dynamic videos with unified styles, similar colors, and lengths within 15 seconds are screened out and pushed to the large screen at the intersection to guide pedestrians into the block. The display is accurate and effective, greatly enhancing the regional business atmosphere and the participation rate of the pedestrian flow.

[0060] When the replacement conditions are met, the content scheduling module sorts the candidate content set and transmits the candidate content with the highest priority to the display terminal. The determination of the replacement sequence and the switching time point includes:

[0061] Sort the candidate content set through a real-time switching algorithm to obtain the candidate content sequence, use preloading technology to load the candidate content, and when the conditions are met, obtain the transmission data from the preloading technology and transmit it to the display terminal. If the transmission data reaches the display terminal, determine the replacement sequence through the switching time point. According to the replacement sequence, obtain the current display content from the display terminal, determine the switching time point, and if the switching time point arrives, update the content of the display terminal through the real-time switching algorithm, obtain the updated content from the display terminal, and determine whether the candidate content has been transmitted completely.

[0062] The content scheduling module in this system is activated when replacement conditions such as a decline in the attraction index or abnormal display status are met, and is responsible for intelligent scheduling and content replacement. The key to the module lies in content priority sorting, preloading mechanism, and precise control of the switching time point. First, the system applies the "real-time switching algorithm" to sort the candidate content set. The sorting function combines the following three weight factors: the attraction index of the candidate content, the semantic relevance to the current theme, and the system adaptability of the content. The priority score can be expressed as: P = α×A + β×S + γ×C; where: P represents the comprehensive priority score of the candidate content; A is the attraction index, derived from the aforementioned audience behavior model (range from 0 to 100); S is the theme similarity score between the candidate content and the currently displayed content (range from 0 to 1); C is the system adaptability score, which measures the matching degree of parameters such as the format, size, and resolution of the candidate content with the display terminal (range from 0 to 1). α, β, and γ are weight parameters, generally set as α = 0.5, β = 0.3, γ = 0.2, and can be adjusted according to the actual scenario and display strategy. After sorting, the system uses the "preloading mechanism" to pre-cache the candidate content with higher scores in the display terminal or edge node in advance. The preloading methods include local caching and continuous loading to ensure that the content can be quickly loaded when a switch is needed. After the content is loaded, the system will determine the specific execution time of content replacement based on the control logic of the "switching time point". The determination of the switching time point considers the following two parameters: the principle of minimum interference: avoiding the highlight moment or interaction peak of the current content playback; system resource monitoring: the system occupancy rate does not exceed the set threshold (such as CPU usage less than 80%, GPU temperature below 75°C); Once the switching time point arrives, the system retrieves the current content identifier from the display terminal, calls the real-time switching algorithm, executes the replacement action, and seamlessly displays the updated content on the screen. The system also verifies whether the candidate content is successfully transmitted and rendered. If not, it maintains the current display state to avoid display interruption.

[0063] The content scheduling module accurately selects the optimal display content through a multi-dimensional sorting strategy, and combines the preloading and switching mechanisms to ensure the efficiency, stability, and low latency of the replacement process. The intelligent control of the switching time point improves the coherence and visual fluency of the system operation, effectively avoiding the discomfort of the audience caused by abrupt switching. Through the priority management of candidate content, the system can dynamically adapt to various external environments and display strategies, realizing the intelligent and scenario-based deployment of multimedia content.

[0064] In a large art park, 10 multimedia display terminals are deployed. Every morning, the system automatically screens out the most popular art content through the scheduling module for preloading. One day at noon, due to a sharp increase in the number of visitors, the system detected that the attraction index of the content "Interactive Fountain Animation" dropped to 52 points, below the threshold of 60 points. The scheduling module triggered the replacement process. The preloaded content "Dynamic Light and Shadow Bridge" ranked first and had been cached. The system detected that the GPU load was 67% and there were no critical interaction behaviors, determining it to be an appropriate time point for switching. Thus, the display content was quickly replaced. The replacement process took less than 1 second, and the average dwell time after switching increased to 8.4 seconds, achieving good interaction feedback.

[0065] The switching control module seamlessly connects the current display content with the candidate content, and determines whether the system load exceeds the preset threshold during the switching process, obtaining a stable switching execution plan including:

[0066] Obtain the replacement sequence, process the connection between the display content and the candidate content through the content buffering mechanism, determine whether the switching triggers a load change, obtain preliminary switching data, analyze the preliminary switching data through the buffering mechanism, determine whether the system load exceeds the preset threshold, and determine the load status. If the load status exceeds the preset threshold, adjust the replacement sequence to obtain an optimized candidate content connection plan. According to the optimized candidate content connection plan, use the load judgment result to determine the smooth transition parameters during the switching process. Process the switching process through the smooth transition parameters to obtain real-time monitoring data of the system load. According to the real-time monitoring data, use the stable scheme generation algorithm to obtain the final execution plan. Through the final execution plan, determine the smoothness of the switching process to obtain a stable switching execution result.

[0067] This module aims to ensure that when the displayed content is switched, the system resources are not over-occupied, and the continuity and smoothness of the content switching process are maintained. Its core mechanism is the coordinated action of content buffering, load monitoring, and stability control algorithms. First, the system extracts the current displayed content and candidate content in sequence according to the "replacement sequence" output by the upper-level content scheduling module, and prepares to execute the switch. At this time, the new and old content frames are loaded in parallel through the dual-channel content buffering mechanism to construct a temporary shared buffer area to support seamless transition processing. The system will monitor the following resource usage situations during the content switching preparation stage to form "preliminary switching data": CPU usage rate (unit: %), GPU temperature (unit: degree Celsius), memory occupancy rate (unit: %), frame rate change value (unit: FPS difference). These parameters will be sent as inputs to the "load judgment function", and the output of this function is the current load state of the system, and the calculation method is as follows: L = a×U + b×G + c×M + d×F, where: L represents the total load value (load index, range 0 - 100), U represents the CPU occupancy rate, G represents the normalized value of the GPU temperature (the temperature is linearly mapped to 0 - 1 according to the set upper limit of 100 °C), M represents the memory ratio, F represents the average frame rate fluctuation range, and a, b, c, d are weighting coefficients, for example, set to 0.4, 0.3, 0.2, 0.1. If the calculated L value exceeds the system-set load threshold T (for example, T = 75), that is, the system is in the "overload risk area". At this time, the system does not directly execute the switch, but adjusts the replacement order, compresses the volume of the content file, or replaces it with a lightweight candidate version to form an "optimized connection plan". Then, based on the current load state and optimization results, the system determines the key control parameters for adjusting the switching smoothness, called "smooth transition parameters", including: buffer delay time (unit: millisecond), content decoding speed multiplier, screen frame rate reduction strategy, content blur or low-resolution transition configuration. These parameters act dynamically on the rendering control engine. When the content switch is officially executed, the content is loaded and displayed frame by frame, and the "system load real-time monitoring data" is continuously collected, and dynamic correction is made again through the "stable scheme generation algorithm". Finally, based on the load sampling points in multiple time periods, the system generates a "switching smoothness index". If it remains within the threshold (such as the frame rate drop does not exceed 15% and the system response delay does not exceed 200 milliseconds), it is determined as a "smooth switch"; otherwise, the switch failure is recorded and re-planned.

[0068] By precisely controlling the resource load during the switching process of the displayed content, this system can effectively avoid common problems such as "black screen", "stuttering", and "screen tearing", ensuring that the audience continuously obtains a smooth viewing experience. The real-time monitoring and dynamic adjustment mechanism provides the system with an adaptive ability, enabling it to still operate stably in a complex and changeable environment, especially suitable for outdoor display scenarios with high concurrency and heavy image calculations.

[0069] In the outdoor LED cultural and art screen system of a commercial CBD, the display control platform needs to switch content units more than 200 times a day on average. When switching to the "Light and Shadow Interactive Show" one evening after the system was deployed, it was detected that the current GPU temperature reached 82°C, the CPU usage rate was 78%, the memory occupancy rate was 89%, and the frame rate fluctuation exceeded 15 FPS. At this time, the system load index was 83, exceeding the threshold. The system immediately scheduled the low-resolution buffered content version through the switching control module, reduced the frame rate to 45 FPS, and delayed the switching time by 100 milliseconds. Finally, the switching process completion time was controlled within 800 milliseconds, without interruption perception, and the stability score reached over 95 points, obtaining good audience feedback.

[0070] The rendering optimization module dynamically adjusts the rendering parameters of the display terminal, extracts candidate content from the buffer, and obtains the updated display content combination including:

[0071] Obtain candidate content from the buffer, use a preset threshold to judge low-attraction content, and obtain the set of content to be replaced. For the set of content to be replaced, dynamically adjust the rendering parameters. By comparing the characteristics of the candidate content and the low-attraction content, determine the replacement priority. If the replacement priority is higher than the preset threshold, extract high-attraction segments from the candidate content and gradually replace the low-attraction content to obtain the preliminarily adjusted display content. According to the preliminarily adjusted display content, obtain the current state of the display terminal, judge whether the rendering parameters match, and obtain the optimized parameter configuration. Through the optimized parameter configuration, adjust the display content combination, use the K-means algorithm to cluster and analyze the content characteristics, determine the distribution of the final display content. For the distribution of the final display content, extract supplementary content from the buffer. If the supplementary content matches the display content combination, then perform fusion to obtain the updated display content combination. According to the updated display content combination, judge the load situation of the display terminal, and adjust the rendering parameters through the load balancing strategy to obtain a stable output display plan.

[0072] The core task of the rendering optimization module is to dynamically adjust the rendering parameters (such as resolution, frame rate, contrast, color temperature, etc.) of the display terminal according to the real-time attractiveness feedback of the displayed content, so as to ensure the balance between content quality and the load of terminal operation. First, the system extracts all current candidate content and existing displayed content from the content buffer, and screens the current displayed content by setting an attractiveness threshold (such as 60 points). Content below this threshold is marked as "low-attractiveness content" and enters the "content set to be replaced". Subsequently, the system compares the features of the content to be replaced with the candidate content, and uses the following replacement priority formula: R = η×ΔA + θ×ΔT + μ×ΔC, where: R is the content replacement priority score, ΔA is the difference in the attractiveness index between the candidate content and the content to be replaced, ΔT is the difference in the theme similarity between the two, ΔC is the difference in the content complexity (such as frame rate, resolution) between the two, and η, θ, μ are the weight coefficients of the above differences (for example, set to 0.5, 0.3, 0.2), which are obtained through empirical debugging or historical data regression. If the replacement priority score R is greater than the set threshold (such as 0.6), the system selects high-attractiveness segments from the candidate content for segmented replacement, thus forming the "preliminary adjusted displayed content". After that, the system obtains the current operating state of the display terminal, including: the current GPU utilization rate, memory occupancy, current output resolution and frame rate, and judges whether it matches the content complexity. If not, it regenerates the "optimized rendering parameter configuration", such as adjusting the resolution to 1280×720, limiting the frame rate to 30FPS, etc. The system binds the preliminary adjusted content with the optimized parameters to generate a combined content set, performs K-means clustering on it, extracts the content structure features (such as picture style, color tone, rhythm), determines the distribution balance, and judges whether there is a concentrated deviation in the content style. If there is a problem of uneven content group styles, the system extracts content with high similarity to the current combination from the buffer (the similarity can be calculated through image Hash or deep feature extraction) for "supplementary fusion" to form the "final updated displayed content combination". Finally, the system judges whether the average complexity of the final combination (content occupying GPU resources, frame rendering load, etc.) exceeds the controllable range of the terminal. If there is an overload risk, the system enables the "load balancing strategy" for dynamic adjustment. The strategies include but are not limited to: reducing brightness, enabling hardware acceleration, frame skipping rendering, etc., so as to achieve the "stable output solution" of the system.

[0073] Through the dual control logic of attractiveness-driven and system state feedback, the system realizes the dynamic coupling optimization of the displayed content and the rendering parameters. Compared with the fixed-parameter rendering mechanism, this method has stronger self-adaptability and visual optimization ability, and can reduce system jitter, latency and other problems while ensuring the display quality. The K-means clustering and supplementary fusion strategies further improve the integrity and rhythm coordination of the displayed content, enhancing the consistency and beauty of the audience experience.

[0074] In an outdoor projection device in a large expo park, the system dynamically adjusts the display content every day based on the interactive feedback from visitors. During a certain festival, the system detected that the attractiveness of the "Urban Traffic Interactive Layer" content continued to drop to 57 points, and the GPU rendering load reached 85%. The system immediately extracted the "Night Scene Light Track Interpretation" content with an attractiveness score of 83 points from the buffer as a candidate segment, and dynamically adjusted the frame rate to 25FPS and the resolution to 720P. After K-means analysis, some "Starry Sky Light" videos were merged to form a new combination with a unified style. After the switch was completed, the GPU load dropped to 69%, the picture rhythm was more soothing, and the audience's dwell time was increased to 6.3 seconds, achieving dual optimization of display effect and terminal pressure.

[0075] The feedback evaluation module collects a new round of audience feedback data, compares the changing trend of the attraction index before and after, and determines whether the content optimization has reached the expected threshold, including:

[0076] Obtain updated display content, extract a new round of audience feedback data through data collection technology to obtain a feedback data set, extract the attraction index from the feedback data set, use a cyclic iterative algorithm to calculate the index change before and after, obtain the change trend data, obtain a preset threshold for the change trend data, judge whether the change trend reaches the preset threshold through a comparison algorithm, and obtain a preliminary judgment result. If the preliminary judgment result does not reach the preset threshold, adjust the display content to generate an optimized content combination. Through the optimized content combination, collect a new round of feedback data to obtain an updated feedback data set, extract the attraction index from the updated feedback data set, use a cyclic iterative algorithm to calculate the new change trend to obtain the optimized trend data, and judge the optimized trend data through a comparison algorithm and the preset threshold to obtain a final judgment result.

[0077] This system determines the optimization effect of the displayed content through a feedback evaluation module. Its core mechanism is to continuously monitor the changes in the audience's responses before and after the replacement of the displayed content, and based on the change trend of the "attraction index", judge whether the expected optimization effect has been achieved. After each update of the displayed content, the system will reactivate the audience behavior data collection module to collect a new round of behavioral response data of the audience from channels such as video monitoring devices, action recognition systems, and interactive interfaces. These data include: average dwell time (i.e., the duration each audience stays in front of the current content), facial expression analysis results (used to evaluate emotional tendencies), and behavioral responses such as clicks, scans, nods, etc. The system standardizes these raw feedback data and converts them into three key indicators: average dwell time (in seconds), representing the duration the audience is attracted by the content; emotional score (ranging from 0 to 1), output according to the facial expression recognition model, such as pleasure, surprise, etc.; behavioral score (also between 0 and 1), calculated from the recognized interaction frequency (e.g., if 3 out of 10 people interact, the score is 0.3). The system assigns weights to these three indicators. For example: the weight of the average dwell time is 50%; the weight of the emotional score is 30%; the weight of the behavioral score is 20%. Multiply the value of each indicator by its corresponding weight, and then add the three results to obtain the "attraction index". For example: Suppose the average dwell time of a certain round of feedback data is 7 seconds, the emotional score is 0.6, and the behavioral score is 0.4. Then the attraction index is calculated as follows: Dwell time: 7 multiplied by 0.5, which is 3.5; Emotional score: 0.6 multiplied by 0.3, which is 0.18; Behavioral score: 0.4 multiplied by 0.2, which is 0.08; The sum is 3.5 + 0.18 + 0.08 = 3.76. The system calculates the difference between the attraction indices before and after optimization. For example: Before optimization is 3.1, after optimization is 3.76; The difference is 0.66. The system saves the difference every other content update and continuously records the average value of the differences in the recent n rounds (such as 3 rounds), that is, forms a "change trend". If the average increase amplitude is greater than the set "expected optimization threshold" (for example, 0.5), the system determines that the content optimization is effective. In addition, the system will also calculate the fluctuation amplitude of these increase values to judge whether the optimization has stabilized. If the change in the increase differences in several consecutive rounds is very small (such as all within ±0.2), it means that the content optimization has reached a stable state, and the system can terminate the optimization process and fix the displayed content.

[0078] Among them, dwell time: Calculate the average residence time of the audience in front of the content through the camera and the personnel recognition model. The higher this value, the more attractive the content is;

[0079] Emotional score: The result output by the facial recognition algorithm, converted into a standardized floating point number, representing the proportion of the audience's positive emotions. The closer the value is to 1, the more positive the emotion is;

[0080] Behavior score: Obtained by converting the frequency of viewer behaviors (such as clicks, scanning codes, etc.) detected by the system, standardized to 0 - 1;

[0081] Weight parameter: Usually set according to the display type. For educational displays, the weight of the dwell time can be appropriately increased; for commercial promotion displays, the weight of the behavior score can be increased. The initial value is set based on the results of viewer behavior research and can be optimized through data training later;

[0082] Optimization threshold: Generally determined by testing historical data, which is the benchmark value for judging the effectiveness of optimization (such as the average attractiveness improvement is at least 0.5 points);

[0083] Fluctuation threshold: Used to determine whether the system has reached a stable optimization state (such as the change fluctuation is within 0.2).

[0084] This module realizes the adaptive upgrade of content through a closed-loop feedback mechanism of "collection - evaluation - optimization - verification". Taking the attractiveness index as the core evaluation index, the system can quantitatively judge the impact of content changes on viewers and automatically decide whether to enter the next round of optimization based on trend stability. This mechanism has high robustness and decision-making intelligence, effectively improving the iteration efficiency of display content and the accuracy of viewer response.

[0085] In a cultural exhibition square in a certain city, the system attempts to display a piece of content of "Digital Interpretation of Intangible Cultural Heritage Paper-cut Art". The average attractiveness index in the first round of collection is 61.3. After optimization (adding voice explanations and background dynamic textures), the system collects the second-round feedback, and the index increases to 67.5, with ΔA being 6.2. Tmean is 6.2, which is greater than the preset threshold of 5 points, and the system determines that the optimization is effective. If the subsequent detection shows that the fluctuation of Tmean becomes smaller (for example, within ±0.3 in three rounds), the system triggers the termination of the optimization mechanism, and the content enters the long-term display queue to achieve stable output.

[0086] If the change trend of the adaptive optimization module does not reach the expected threshold, external variables are extracted from the dynamic environment, and the attractiveness index is recalculated. The obtained adjusted content optimization direction and replacement sequence include:

[0087] If the change trend does not reach the preset threshold, the external variable extraction module obtains external variables from the dynamic environment to get a preliminary data set. Through the preliminary data set, the data analysis module calculates the attractiveness index to obtain the initial index value. If the initial index value is lower than the preset threshold, the external variables are adjusted through regression analysis to obtain an optimized variable set. Based on the optimized variable set, the attractiveness index is recalculated to obtain an adjusted index value. Through the adjusted index value, content adjustment parameters are generated to obtain an optimization direction sequence. Based on the optimization direction sequence, a sorting algorithm is used to generate a replacement sequence to obtain the final content adjustment plan. Through the final content adjustment plan, the variable configuration in the dynamic environment is updated to obtain new change trend data.

[0088] When the system discovers through the feedback evaluation module that the changing trend of the attraction index does not reach the set expected threshold (for example, the expected increase is less than 5 points), it indicates that the content optimization has not achieved the desired effect. At this time, the adaptive optimization module will activate the environmental response mechanism and attempt to adjust the content performance through external variables to achieve the optimization goal.

[0089] Phase 1: Environmental variable extraction and initial index analysis:

[0090] The system first obtains external variables from the environmental extraction module. These variables include but are not limited to: real-time weather (such as temperature, humidity, light intensity); time factors (such as day / night, weekday / weekend); crowd density; background noise level; air quality index, etc.

[0091] The above data constitutes the "preliminary environmental variable dataset". The system inputs this dataset into the analysis model, combines it with the current content, and uses the following formula to re-estimate the "content attraction index": A = f(T, E, B, V1, V2,..., Vn), where T is the average viewing time of the current content by the audience; E is the audience emotion score, B is the behavior score, V1 to Vn are each external environmental variable (such as temperature, light, noise, etc.), f is the comprehensive weighting function, which can be a linear model or a machine learning regression model, and the output A is the re-estimated attraction index after integrating environmental variables. If the re-estimated index value is lower than the system-set threshold (such as 60 points), the system enters the second phase.

[0092] Phase 2: Regression analysis and external variable adjustment:

[0093] The system uses a multivariable linear regression model or a random forest regression to analyze the sensitivity and weight between environmental variables and the attraction index, and identify the variables with the greatest impact. For example, it is found that "too strong light" and "too high background noise" have a negative correlation with the attraction index. By adjusting these external variables (such as automatically reducing the display brightness, switching to quiet sound effect content), the system forms an "optimized variable set", and substitutes it into the above function again to calculate the "adjusted attraction index". If the new index value is higher than the preset threshold, "content adjustment parameters" are generated, such as: switching to warm-tone content; adding music elements; simplifying the visual structure, etc.

[0094] Phase 3: Generating replacement sequences and final adjustment plans:

[0095] The system sorts multiple content adjustment parameters to form an "optimization direction sequence", and uses a priority sorting algorithm (such as weight-weighted sorting, heuristic rule-based sorting) to generate a "candidate replacement sequence". Finally, a "content adjustment plan" that can be used for real-time display is formed.

[0096] Phase 4: Variable Configuration Update and Trend Tracking:

[0097] The system updates the environmental variable configuration again according to the new display content and optimization parameters. For example, it updates the time period corresponding to the recommended content type or sets a special display template for nighttime. Then it continues to collect feedback, obtains "new attraction change trend data", and returns to the feedback evaluation module to enter a new round of loop.

[0098] Among them: Attraction Index A: Evaluates the ability of the content to attract the audience, with a standardized score of 0 - 100 points. T (Stay Time): The average viewing duration of the audience statistically by the system through sensors. E (Emotion Score): The positive emotion probability value output by the AI facial recognition model. B (Behavior Score): Obtained after standardizing the frequency of interaction behaviors. V1 - Vn (Environmental Variables): External state data collected by sensors or external platforms (such as weather APIs). Threshold Setting: The attraction index threshold is set based on historical tests and user response behaviors (such as 60 points), and the change trend threshold is generally set to increase by 5 - 10 points.

[0099] By introducing external environmental dynamic variables, the system can further search for external intervention paths when optimization fails, and achieve personalized content adaptation based on scenarios. Its adaptive mechanism not only depends on the adjustment of the content itself, but also integrates environmental and spatio - temporal factors, making the content strategy have stronger intelligent, precise and flexible response capabilities.

[0100] In a display system in the square of an art museum in a certain city, the system found that the attraction index of the "Future City Dynamic Projection" content displayed in the afternoon of a certain day decreased significantly (only 52 points). After the adaptive module was activated, it was analyzed that strong sunlight caused strong screen reflection and the background noise of the crowd was relatively loud, interfering with the sound effect. The system automatically reduced the display brightness to 80%, switched to a black - background content with higher contrast, and eliminated part of the noise impact through background music superposition. After the adjustment, the attraction index rose to 64.2 points, and the system determined that the optimization was effective and solidified this plan as the "Summer Afternoon Display Template".

[0101] The content loop control module uses a real - time switching algorithm and a buffering mechanism to repeatedly execute content replacement and parameter adjustment to obtain the final self - adaptive optimized display content combination including:

[0102] The initial replacement sequence and display content are obtained through a real-time switching algorithm to get a preliminary content combination. The preliminary content combination is stored using a buffering mechanism. It is judged whether the self-adaptability requirement is met. If the self-adaptability requirement is not met, an adjustment logic is executed for the replacement sequence to get an updated sequence. The content replacement is repeatedly executed through the updated sequence to obtain optimized display content. The optimized display content is processed using a parameter optimization algorithm to determine the final combination form. Dynamic switching is performed according to the final combination form to obtain self-adaptive optimized display content. The self-adaptive optimized display content is verified through the buffering mechanism to judge whether the expected standard is achieved.

[0103] The content loop control module in this system aims to achieve the final self-adaptive optimized display content combination by continuously performing content replacement and parameter adjustment. Its working process depends on the coordinated control of the "real-time switching algorithm" and the "buffering mechanism", and takes the "self-adaptability score" as the core criterion to judge whether the display content meets the current audience behavior and system load conditions.

[0104] Step 1: Generate a preliminary content combination:

[0105] The system first uses the real-time switching algorithm to generate an "initial replacement sequence" based on the currently available candidate content. The display content combination corresponding to this sequence is loaded into the display terminal and locally stored through the buffering mechanism to form a "preliminary content combination".

[0106] Step 2: Self-adaptability judgment:

[0107] The system then evaluates the self-adaptability performance of this combination. To this end, it calculates a "self-adaptability score" to measure whether the current display content is suitable for the current system operating state and audience behavior performance.

[0108] This score consists of three parts: the fluctuation degree of the audience feedback, that is, the amplitude of the fluctuation of the attraction index during the content switching in recent rounds; the gap between the current content and the target attraction, that is, the difference between the attraction score of the current content and the set standard; the deviation degree of the system resource load, that is, the gap between the current CPU / GPU load and the allowable range of the system performance. The system sums up these three indicators after weighting to form a total score value. For example, assume that the audience feedback fluctuation value is 0.8 (large fluctuation), the attraction difference is 10 points, and the system load deviation is 0.3; the weights corresponding to each indicator are set as: feedback fluctuation 0.4, attraction difference 0.4, system load deviation 0.2; then the adaptive score is calculated as follows: feedback fluctuation score: 0.8 multiplied by 0.4, equal to 0.32, attraction difference score: 10 multiplied by 0.4, equal to 4, system load score: 0.3 multiplied by 0.2, equal to 0.06, and the sum is 0.32 + 4 + 0.06 = 4.38. If the score value is greater than the set judgment threshold of the system (such as 4 points), it means that the current combination does not have good adaptability, and the system needs to enter the next optimization and adjustment process.

[0109] Step 3: Replacement sequence adjustment and parameter optimization:

[0110] At this time, the system reconstructs the replacement sequence and selects new content units to replace the mismatched parts in the current combination. After the replacement is completed, a new "optimized display content" is formed. The system performs a "parameter optimization" process on this optimized combination, such as adjusting visual parameters such as resolution, frame rate, and color temperature. The optimized value of each parameter consists of its basic default value plus an "adjustment range" determined by the feedback data.

[0111] For example: The original frame rate is set to 60 FPS. The system finds that this frame rate causes the GPU to be overloaded. After calculation, the optimal frame rate adjustment value is -15. Then the new frame rate is set to 60 - 15 = 45 FPS.

[0112] Step 4: Dynamic switching and verification:

[0113] After the parameter adjustment is completed, the system officially deploys the optimized combination content to the display terminal and realizes the content transition display through the dynamic switching algorithm. At the same time, the system continues to collect the audience feedback and the device operation status, and records the data through the buffer mechanism. If the final performance indicators of the content combination (such as the average attraction index, system load level) meet or exceed the set standards (for example: attraction index ≥ 65, GPU load ≤ 75%, average stay time ≥ 5 seconds), the system determines that this optimization is effective and the process terminates; otherwise, it re-enters the content replacement and parameter adjustment process.

[0114] Among them, the feedback fluctuation value: calculated from the standard deviation of the attraction index in several consecutive rounds, reflecting the consistency or instability of the audience response; the attraction difference: the absolute difference between the attraction score of the current combination and the target score, and the target score is based on historical average performance or manually set (such as 65 points); the system load deviation: the difference between the current GPU / CPU utilization rate and the upper limit tolerated by the system. For example, if the current GPU is 82% and the upper limit is 75%, the deviation is 7%; the weight parameter: can be initially set according to experience. If the importance of feedback fluctuation and attraction difference is equivalent, each is set to 0.4, and the system load is 0.2. Subsequently, it can be automatically adjusted through data learning according to the actual scenario; the parameter adjustment value: dynamically generated based on training models or optimizing algorithms (such as gradient descent, genetic algorithm) for feedback data to ensure that the system achieves the best balance between performance and performance.

[0115] This module constructs a dynamic, iterative, and closed-loop optimization mechanism that can continuously adjust the display content and technical parameters to achieve the maximum content adaptability and the optimal matching of system resources. Compared with traditional static display strategies, this module is more environmentally sensitive and adjustable autonomously, and can effectively respond to the changing content performance requirements at different times, with different flows of people and environments.

[0116] In a certain intelligent cultural and tourism night tour scenic area, the system needs to automatically schedule more than 30 groups of lighting and content combinations within 6 hours every night. When the system detects that the fluctuation of the tourist feedback index is too large, it automatically adjusts the playback rhythm, lowers the brightness and frame rate, and introduces static picture tweening to optimize the display effect. A certain content, "Night Landscape Scroll", caused a sudden increase in GPU pressure due to the relatively fast switching rhythm. The system detected non-compliance through a real-time scoring model (the self-adaptive score was 78), immediately triggered a parameter callback, reduced the frame rate from 60 FPS to 45 FPS, and replaced it with content of low complexity. After the switching was completed, the GPU load dropped back to 71%, and the attraction index rose to 68. The system determined that the optimization was successful.

[0117] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An outdoor intelligent art display system based on multimedia interaction, characterized in that, Including: An audience behavior collection module, which is used to collect the audience's stay time and behavioral responses in real time, obtain an initial evaluation result of content attractiveness, and calculate the attractiveness index of each content unit; A content analysis and screening module, which is used to extract alternative multimedia content units from a preset content library, judge the relevance between the alternative content and the theme of the currently displayed content, and obtain a set of candidate content that meets the replacement conditions; A content scheduling module, which is used to sort the set of candidate content when the replacement condition is met, transmit the candidate content with the highest priority to the display terminal, and determine the replacement sequence and switching time point; A switching control module, which is used to seamlessly connect the currently displayed content with the candidate content, judge whether the system load exceeds a preset threshold during the switching process, and obtain a stable switching execution plan; A rendering optimization module, which is used to dynamically adjust the rendering parameters of the display terminal, extract candidate content from the buffer, and obtain an updated display content combination; A feedback evaluation module, which is used to collect a new round of audience feedback data, compare the change trends of the attractiveness indices before and after, and judge whether the content optimization has reached the expected threshold; An adaptive optimization module, which is used to, if the change trend does not reach the expected threshold, extract external variables from the dynamic environment, recalculate the attractiveness index, and obtain the adjusted content optimization direction and replacement sequence; A content loop control module, which is used to repeatedly execute content replacement and parameter adjustment by using a real-time switching algorithm and a buffering mechanism to obtain a final self-adaptive optimized display content combination.

2. The outdoor intelligent art display system based on multimedia interaction according to claim 1, wherein: The audience behavior collection module's real-time collection of the audience's stay time and behavioral responses, and obtaining an initial evaluation result of content attractiveness and calculating the attractiveness index of each content unit includes: Real-time collection of the audience's stay time and behavioral response data through sensors and interaction devices to obtain the display status and feedback information of the multimedia content, extraction of the stay time and behavioral response characteristics from the collected sensor data, acquisition of the audience interaction data of each content unit, quantification analysis of the stay time and behavioral responses by using a data processing method, calculation of the attractiveness index of each content unit. If the attractiveness index is lower than the preset threshold, identify the abnormal presentation of the content unit through the display status, determine the adjustment direction, optimize the display status of the content unit according to the adjustment direction to obtain updated multimedia content data, re-collect the audience feedback information through the updated multimedia content data, judge the change trend of the attractiveness index, and perform predictive analysis on the change trend by using a random forest algorithm to obtain the long-term attractiveness evaluation result of each content unit.

3. An outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: The content analysis and screening module's extraction of alternative multimedia content units from a preset content library and judging the relevance between the alternative content and the theme of the currently displayed content, and obtaining a set of candidate content that meets the replacement conditions includes: Retrieve multimedia data from a preset content library, use content extraction technology to separate alternative units, obtain an initial alternative content list. For the initial alternative content list, calculate the attraction index, perform sorting based on the exponential distribution to obtain a sorted alternative content sequence. If there is a theme association that matches the currently displayed content in the sorted alternative content sequence, retain the relevant alternative units to obtain a theme-related content subset. According to the replacement conditions, determine whether the alternative units in the theme-related content subset meet the requirements to obtain a preliminary candidate set. Process the preliminary candidate set through a clustering algorithm to merge similar alternative units and obtain an optimized candidate content set. Retrieve the optimized candidate content set, conduct a secondary verification for theme relevance to determine the final candidate content set. Use the final candidate content set to adjust the sorting based on the attraction index distribution and output a candidate content sequence that meets the requirements.

4. An outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: When the replacement conditions of the content scheduling module are met, sort the candidate content set and transmit the candidate content with the highest priority to the display terminal. Determining the replacement sequence and the switching time point includes: Sort the candidate content set through a real-time switching algorithm to obtain a candidate content sequence. Use preloading technology to load the candidate content. When the conditions are met, obtain the transmission data from the preloading technology and transmit it to the display terminal. If the transmission data reaches the display terminal, determine the replacement sequence through the switching time point. According to the replacement sequence, obtain the currently displayed content from the display terminal and determine the switching time point. If the switching time point arrives, update the content of the display terminal through the real-time switching algorithm, obtain the updated content from the display terminal, and determine whether the candidate content has been transmitted completely.

5. An outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: The switching control module seamlessly connects the currently displayed content with the candidate content and determines whether the system load exceeds a preset threshold during the switching process. The stable switching execution plan includes: Obtain the replacement sequence, process the connection between the displayed content and the candidate content through a content buffering mechanism, determine whether the switching triggers a load change to obtain preliminary switching data. Analyze the preliminary switching data through the buffering mechanism to determine whether the system load exceeds the preset threshold and determine the load status. If the load status exceeds the preset threshold, adjust the replacement sequence to obtain an optimized candidate content connection plan. According to the optimized candidate content connection plan, use the load judgment result to determine the smooth transition parameters during the switching process. Process the switching process through the smooth transition parameters to obtain real-time monitoring data of the system load. According to the real-time monitoring data, use a stable plan generation algorithm to obtain the final execution plan. Through the final execution plan, determine the smoothness of the switching process to obtain a stable switching execution result.

6. The outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: The rendering optimization module dynamically adjusts the rendering parameters of the display terminal and extracts candidate content from the buffer to obtain an updated display content combination, including: Retrieve candidate content from the buffer, use a preset threshold to judge low - attraction content, and obtain a set of content to be replaced. For the set of content to be replaced, dynamically adjust the rendering parameters. By comparing the features of the candidate content and the low - attraction content, determine the replacement priority. If the replacement priority is higher than the preset threshold, extract high - attraction fragments from the candidate content and gradually replace the low - attraction content to obtain the preliminarily adjusted display content. According to the preliminarily adjusted display content, obtain the current state of the display terminal, judge whether the rendering parameters match, and obtain the optimized parameter configuration. Through the optimized parameter configuration, adjust the display content combination, use the K - means algorithm to cluster and analyze the content features, determine the distribution of the final display content. For the distribution of the final display content, extract supplementary content from the buffer. If the supplementary content matches the display content combination, then fuse them to obtain the updated display content combination. According to the updated display content combination, judge the load situation of the display terminal, and adjust the rendering parameters through the load - balancing strategy to obtain a stable output display scheme.

7. An outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: The feedback evaluation module collects a new round of audience feedback data and compares the change trend of the attraction index before and after to judge whether the content optimization reaches the expected threshold, including: Obtain the updated display content, extract a new round of audience feedback data through data - collection technology to obtain a feedback data set, extract the attraction index from the feedback data set, use a loop - iteration algorithm to calculate the change of the index before and after to obtain the change - trend data. For the change - trend data, obtain the preset threshold, use a comparison algorithm to judge whether the change trend reaches the preset threshold to obtain a preliminary judgment result. If the preliminary judgment result does not reach the preset threshold, then adjust the display content to generate an optimized content combination. Through the optimized content combination, collect a new round of feedback data to obtain an updated feedback data set, extract the attraction index from the updated feedback data set, use a loop - iteration algorithm to calculate the new change trend to obtain the optimized trend data. For the optimized trend data, use a comparison algorithm to judge with the preset threshold to obtain the final judgment result.

8. An outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: If the change trend does not reach the expected threshold, the adaptive optimization module extracts external variables from the dynamic environment and recalculates the attraction index to obtain the adjusted content - optimization direction and replacement sequence, including: If the change trend does not reach the preset threshold, then use the environment - extraction module to obtain external variables from the dynamic environment to obtain a preliminary data set. Through the preliminary data set, use the data - analysis module to calculate the attraction index to obtain the initial index value. If the initial index value is lower than the preset threshold, then adjust the external variables through regression analysis to obtain an optimized variable set. According to the optimized variable set, recalculate the attraction index to obtain the adjusted index value. Through the adjusted index value, generate content - adjustment parameters to obtain an optimization - direction sequence. According to the optimization - direction sequence, use a sorting algorithm to generate a replacement sequence to obtain the final content - adjustment scheme. Through the final content - adjustment scheme, update the variable configuration in the dynamic environment to obtain new change - trend data.

9. An outdoor intelligent art display system based on multimedia interaction according to claim 1, characterized in that: The content loop control module uses a real-time switching algorithm and a buffering mechanism to repeatedly execute content replacement and parameter adjustment, and the final self-adaptive optimized display content combination includes: Obtain the initial replacement sequence and display content through the real-time switching algorithm to get a preliminary content combination, store the preliminary content combination using the buffering mechanism, determine whether the self-adaptability requirements are met. If the self-adaptability requirements are not met, execute the adjustment logic for the replacement sequence to obtain an updated sequence.

10. An outdoor intelligent art display system based on multimedia interaction according to claim 9, characterized in that: The content loop control module uses a real-time switching algorithm and a buffering mechanism to repeatedly execute content replacement and parameter adjustment, and the final self-adaptive optimized display content combination also includes: Repeatedly execute content replacement through the updated sequence to obtain the optimized display content, process the optimized display content using the parameter optimization algorithm to determine the final combination form, perform dynamic switching according to the final combination form to obtain the self-adaptive optimized display content, and verify the self-adaptive optimized display content through the buffering mechanism to determine whether the expected standard is achieved.

Images