A dynamic adaptation method, device and medium for display content of an LED screen

Abstract

The application discloses a display content dynamic adaptation method and device of an LED screen and a medium. The method comprises the following steps: collecting original feature data of audiences in the radiation range of the LED screen through a multi-modal biological sensor, preprocessing the original feature data, and generating a desensitization feature vector through local anonymization processing; using a multi-modal fusion model in an edge computing node to perform real-time analysis on the desensitization feature vector and generate a primary emotion label; uploading the primary emotion label and associated audience historical behavior data to a cloud platform to predict potential interests of the audiences through a preset emotion depth analysis model, and generating a dynamic content adaptation strategy in combination with the potential interests of the audiences; based on the dynamic content adaptation strategy, calling display content matched with the emotion label from a pre-stored material library to perform low-delay rendering on the display content, and dynamically adjusting color parameters and content display sequences of the LED display screen according to the emotion analysis result.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, device and medium for dynamically adapting the display content of an LED screen. Background Technology

[0002] Traditional LED display systems primarily adjust display parameters through preset programs or environmental sensors. Their optimization goals are limited to basic physical indicators such as brightness and refresh rate, lacking the ability to perceive and respond to the subjective emotional state of the audience. Such systems cannot dynamically adapt content based on the audience's emotions, resulting in low advertising conversion rates or insufficient public information delivery.

[0003] Existing AI-based LED control solutions attempt to incorporate single-modal data, such as human posture detection, to optimize the interactive experience. However, these solutions suffer from significant limitations in data dimensionality, making it difficult to accurately identify complex emotional states. Furthermore, traditional solutions often employ a centralized cloud processing model, requiring biosensor data to be remotely transmitted to the cloud for analysis. This not only introduces significant latency but also poses a risk of sensitive biometric information leakage, failing to meet the dual requirements of real-time interaction and privacy protection in public places.

[0004] Furthermore, existing technologies have significant shortcomings in multimodal data fusion and dynamic decision-making. Most systems employ fixed-weight data fusion methods, failing to dynamically adjust the contribution of different sensor data based on scene changes or emotional states, resulting in significant deviations between analysis results and actual emotions. Simultaneously, the mapping logic between displayed content and emotional states is simplistic, lacking refined control over color space and content scheduling strategies, making it difficult to effectively guide audience attention. Moreover, existing solutions have weak detection and response mechanisms for special states such as audience fatigue, failing to alleviate fatigue or enhance interaction willingness through display parameter optimization and content strategy adjustments. Summary of the Invention

[0005] This application provides a method, device, and medium for dynamically adapting the display content of an LED screen to solve the above-mentioned technical problems.

[0006] On the one hand, embodiments of this application provide a method for dynamically adapting the display content of an LED screen, including: The raw feature data of the audience within the radiation range of the LED screen is collected by a multimodal biosensor. The raw feature data is preprocessed and generated into a desensitized feature vector through local anonymization. The raw feature data includes facial expression data, voice feature data and physiological signal data. The desensitized feature vectors are analyzed in real time using a multimodal fusion model in edge computing nodes to generate primary sentiment labels; The primary emotion tags and associated audience historical behavior data are uploaded to the cloud platform to predict potential audience interests through a preset emotion depth analysis model, and dynamic content adaptation strategies are generated based on the potential audience interests. Based on the dynamic content adaptation strategy, display content matching the emotion tag is called from the pre-stored material library to perform low-latency rendering of the display content, and the color parameters and content display order of the LED display screen are dynamically adjusted according to the emotion analysis results.

[0007] In one implementation of this application, raw characteristic data of the audience within the radiation range of the LED screen are collected using a multimodal biosensor, specifically including: Based on the high frame rate infrared camera and 3D structured light sensor embedded in the LED screen bezel, data on facial expressions, micro-expressions and head postures of the audience within the radiation range of the LED screen are collected. The system collects voice data of the audience within the radiation range of the LED screen using a directional microphone array deployed on the LED screen, and extracts the intonation and speech rate features from the voice data using voiceprint separation technology. The system uses non-contact millimeter-wave radar to detect the heart rate and respiratory rate of spectators within the radiation range of the LED screen, and combines this with a thermal imaging camera to obtain data on the distribution of body surface temperature of spectators within the radiation range of the LED screen.

[0008] In one implementation of this application, a preset deep emotion analysis model is used to predict the audience's potential interests, and combined with the audience's potential interests, a dynamic content adaptation strategy is generated, specifically including: A deep emotion analysis model is built based on the Transformer architecture, and historical behavioral data within the radiation range of the LED screen is obtained; the historical behavioral data includes dwell time and interaction frequency. By using a trained deep emotion analysis model, combined with the historical behavioral data and the primary emotion tags, the potential interests of the audience are predicted, and the emotion type corresponding to the potential interests of the audience is determined. When the emotional type is fatigue, interactive games or dynamic advertisements are pushed to increase the audience's attention; When the emotion type is excitement, the display time of the corresponding advertisement is extended, and a preset high-saturation color is matched for the corresponding advertisement.

[0009] In one implementation of this application, the original feature data is preprocessed, and a desensitized feature vector is generated through local anonymization, specifically including: The original feature data is subjected to spatiotemporal alignment and noise filtering; Using a federated learning framework, sensitive information in the preprocessed raw feature data is locally anonymized, retaining only desensitized feature vectors related to emotion recognition; the sensitive information includes facial features. The desensitized feature vector is uploaded to the cloud platform, and the original feature data is automatically destroyed after being stored locally for a preset period of time.

[0010] In one implementation of this application, a multimodal fusion model in an edge computing node is used to perform real-time analysis on the desensitized feature vector to generate a preliminary sentiment label, specifically including: Based on a multimodal fusion model pre-deployed in edge computing nodes, an attention mechanism is adopted, and the input weights of facial expression data and physiological signal data are adjusted according to the current ambient lighting conditions or audience behavior. In scenarios involving emotional fluctuations, the feature fusion between voice feature data and heart rate data in physiological signal data is prioritized to generate primary emotion labels in real time and trigger a rapid response strategy; the primary emotion labels include excitement, fatigue, and neutrality.

[0011] In one implementation of this application, the color parameters and content display order of the LED display screen are dynamically adjusted based on the sentiment analysis results, specifically including: The primary emotion labels are mapped to the HSL color space, and the corresponding hue, saturation, and brightness combinations are selected according to the emotion type in the primary emotion labels. The order of ad playback is optimized by using reinforcement learning algorithms to maximize viewer dwell time and interaction probability.

[0012] In one implementation of this application, low-latency rendering of the displayed content specifically includes: Hardware accelerators are used to render the displayed content in parallel to ensure that the rendering latency is below a preset threshold; During the rendering process, the color engine parameters corresponding to the primary emotion tags are matched in real time to dynamically adjust the brightness and color temperature of the image.

[0013] In one implementation of this application, it further includes: The audience's eye closure frequency, head posture deviation angle, and heart rate variability were analyzed to determine the audience's fatigue level. The eye protection mode is triggered based on the fatigue level, dynamically reducing the proportion of blue light on the screen and simplifying interface information, while pushing voice navigation prompts or emergency warnings.

[0014] On the other hand, embodiments of this application also provide a dynamic content adaptation device for an LED screen, the device comprising: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a dynamic adaptation method for LED screen display content as described above.

[0015] On the other hand, this application embodiment also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, implement the above-described method for dynamic adaptation of LED screen display content.

[0016] This application provides a method, device, and medium for dynamically adapting the display content of an LED screen, which has at least the following beneficial effects: By integrating multi-dimensional data such as facial expressions, voice features, and physiological signals, and dynamically adjusting sensor weights using an attention mechanism, the system effectively overcomes the vulnerability of single-modal data to environmental interference, significantly improving the accuracy and robustness of emotion recognition. Secondly, edge computing nodes perform localized data processing and generate preliminary emotion labels, greatly reducing decision-making delays caused by traditional centralized cloud processing. Simultaneously, a federated learning framework anonymizes biometric data, uploading only desensitized feature vectors to the cloud, mitigating the risk of sensitive information leakage and meeting the real-time interaction needs and privacy compliance requirements of public places. Furthermore, based on dynamic content optimization strategies and hardware-accelerated rendering technology, the system can adjust color space mapping and ad playback order in real time according to emotion analysis results, thereby accurately guiding viewer attention and improving ad conversion rates and information delivery efficiency. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for dynamically adapting the display content of an LED screen, as provided in an embodiment of this application; Figure 2 This is a schematic diagram of the internal structure of an LED screen display content dynamic adaptation device provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart illustrating a method for dynamically adapting the display content of an LED screen, as provided in an embodiment of this application.

[0021] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.

[0022] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.

[0023] like Figure 1 As shown in the embodiment of this application, a method for dynamically adapting the display content of an LED screen includes: Step 101: Collect raw feature data of the audience within the radiation range of the LED screen using a multimodal biosensor, preprocess the raw feature data, and generate desensitized feature vectors through local anonymization.

[0024] It should be noted that the original feature data in the embodiments of this application includes facial expression data, voice feature data, and physiological signal data.

[0025] In this embodiment, the multimodal biosensor includes a vision module, a voice module, and a physiological signal module. The vision module consists of a high-frame-rate infrared camera and a 3D structured light sensor. For example, the infrared camera captures facial micro-expressions in low-light environments, such as twitching of the mouth and pupil dilation, while the 3D structured light sensor accurately calculates head posture angles, such as pitch and yaw angles, by projecting structured light spots and analyzing reflection deformation. In a shopping mall setting, when a viewer approaches the LED screen, the vision module can track their facial expression changes in real time, providing basic data for emotion recognition.

[0026] The voice module collects audience voice data through a directional microphone array, uses beamforming technology to filter out environmental noise, and extracts intonation and speech rate features, such as pitch and speed, through a voiceprint separation algorithm. For example, in a shopping mall setting, when an audience member speaks to the screen, the system can distinguish between hesitation and excitement in their voice.

[0027] The physiological signal module uses non-contact millimeter-wave radar to detect heart rate and respiratory rate. Its principle is to transmit millimeter-wave signals and receive the Doppler frequency shift caused by chest cavity movement to calculate physiological rhythms. Simultaneously, a thermal imaging camera captures infrared radiation from the body surface to generate temperature distribution maps, such as the temperature difference between the forehead and hands, to aid in assessing emotional state.

[0028] In this embodiment, the preprocessing of the raw feature data includes spatiotemporal alignment and noise filtering. It is understood that spatiotemporal alignment, through timestamp synchronization and spatial coordinate mapping, ensures the spatiotemporal consistency of data from different sensors. For example, aligning the time window of the voice data with the frame rate of facial expression capture avoids timing deviations during emotion analysis. Noise filtering employs wavelet transform algorithms to eliminate environmental interference, such as filtering out interference from light reflections from the LED screen itself on the camera data.

[0029] Local anonymization is achieved through a federated learning framework. Specifically, facial features are blurred and encoded, such as retaining only features of the eye region, and the voiceprint data undergoes spectral desensitization, such as removing frequency bands that can identify timbre, generating a desensitized vector containing only the features needed for emotion recognition. The original data is automatically destroyed after being stored locally for a preset period of time to ensure compliance with privacy protection requirements.

[0030] In this embodiment, a sensor array is embedded in the bezel of the LED screen, with a camera spacing of ≤10cm, a radar coverage angle of 120°, and an edge computing node using NVIDIA Jetson AGX Orin, achieving a computing power of 200 TOPS. Multimodal data acquisition is initiated when a viewer is detected entering within a 3-meter radius.

[0031] In this embodiment, the intelligent information screens in transportation hubs such as airports and high-speed rail stations need to achieve passenger status perception and accurate information push in complex lighting and dense crowd environments, while solving the problem of distraction caused by passenger fatigue.

[0032] The vision module employs a near-infrared wide-angle camera (120° FOV) and a ToF depth sensor, supporting simultaneous detection of multiple people at long distances (5-10 meters). It eliminates crowd obstruction interference through a binocular parallax algorithm. The physiological signal module integrates a 60GHz millimeter-wave radar array for non-contact detection of heart rate variability (HRV) and respiratory rate. Combined with a thermal imaging camera, it monitors changes in body surface temperature gradients, such as the temperature difference between the forehead and hands, for fatigue level grading, including mild, moderate, and severe. The environmental perception module deploys dust sensors (PM2.5 / PM10) and temperature and humidity sensors around the LED screen to dynamically correct display parameters, such as improving contrast in smoggy weather.

[0033] Step 102: Utilize the multimodal fusion model in the edge computing node to perform real-time analysis on the desensitized feature vector and generate primary sentiment labels.

[0034] In this embodiment, a lightweight multimodal fusion model (such as MMF-Net) is deployed on the edge computing nodes, with an attention mechanism at its core. This mechanism dynamically adjusts the weights of different sensor data based on the current environmental scene (such as light intensity and crowd density) and the audience's behavioral state (such as standing or walking). For example, in a well-lit and sparsely populated shopping mall environment, the multimodal fusion model enhances the weight of the visual module, prioritizing the analysis of facial micro-expressions and head posture. In a noisy transportation hub scene, the multimodal fusion model increases the weight of the speech module and the physiological signal module, for example, combining increased heart rate and faster speech rate to determine anxiety.

[0035] Specifically, when the model detects a viewer's rapid speech and significant heart rate fluctuations, it triggers an excitement label. If the viewer remains still for an extended period and their body temperature drops, a decrease in respiratory rate is used to determine a state of fatigue. The generated initial emotion labels include excitement, fatigue, and neutral, which trigger rapid response strategies, such as immediately switching to a high-contrast display or simplifying interface information. Understandably, edge computing's localized processing significantly reduces decision latency. For example, in an advertising screen scenario, when a viewer shows interest, edge nodes can generate emotion labels and trigger content switching within milliseconds, avoiding the response delays caused by traditional cloud processing.

[0036] In this embodiment, the Qualcomm QCS8550 chipset is used, which supports multimodal data parallel processing (computing power up to 45 TOPS). A lightweight fatigue recognition model is deployed locally, based on the MobileNetV3+Bi-LSTM architecture, to output fatigue level and emotion label in real time.

[0037] Step 103: Upload the primary emotion tags and associated audience historical behavior data to the cloud platform to predict the audience's potential interests through a preset emotion depth analysis model, and generate dynamic content adaptation strategies based on the audience's potential interests.

[0038] In this embodiment, the cloud platform constructs a deep emotion analysis model based on the Transformer architecture. Its inputs include primary emotion tags and historical behavioral data, such as dwell time and interaction frequency. The deep emotion analysis model uses a self-attention mechanism to mine the correlation between emotion tags and historical behavior, predicting potential audience interests. For example, if a viewer responds positively to technology advertisements in multiple interactions, and their current emotion is detected as excitement, the deep emotion analysis model predicts they may be interested in new product promotions, thus generating a strategy to extend the display time of technology advertisements. If a viewer repeatedly displays the fatigue tag in a transportation hub scenario, the deep emotion analysis model combines their dwell location data to generate a strategy to push navigation to nearby rest areas.

[0039] The generation of dynamic content adaptation strategies also includes contextual logic. Specifically, when the model determines that the audience is fatigued, interactive games or highly dynamic ads are pushed to increase attention; if the audience is excited, the display duration of relevant ads is extended and matched with highly saturated colors. In addition, the cloud platform continuously optimizes the strategy through reinforcement learning, for example, dynamically adjusting the content recommendation priority based on ad click-through rates.

[0040] Understandably, the collaborative computing model between the cloud and the edge ensures the accuracy of deep analysis while meeting real-time requirements through the rapid response of edge nodes. For example, in a shopping mall scenario, optimization strategies deployed from the cloud can update the content library of edge nodes in real time, ensuring that advertising content iterates in sync with audience interests.

[0041] Step 104: Based on the dynamic content adaptation strategy, retrieve the display content that matches the emotion tag from the pre-stored material library to perform low-latency rendering of the display content, and dynamically adjust the color parameters and content display order of the LED display screen according to the emotion analysis results.

[0042] In this embodiment, the pre-stored content library contains multiple versions of content, such as videos, images, and text. Each type of content is associated with emotion tags and scene metadata, such as time period and geographical location. For example, for the emotion of excitement, the content library stores dynamic advertisements with high-saturation colors, such as advertisements with high-saturation red tones, and combines them with dynamic special effects to enhance visual impact. For the emotion of fatigue, soothing images with cool tones, such as blue tones, are matched. In addition, the system synthesizes personalized content in real time through Generative Adversarial Networks (GANs), such as generating virtual avatar interactive interfaces based on the viewer's age.

[0043] Dynamic adjustment of color parameters is achieved through HSL color space mapping. Specifically, mood labels are mapped to preset combinations of hue (H), saturation (S), and brightness (L). For example, a neutral mood corresponds to a natural hue and medium saturation, while a fatigue mood is mapped to a low-brightness warm color to alleviate visual fatigue.

[0044] The content display order is optimized using reinforcement learning algorithms, such as Q-learning. For example, when the system detects that a viewer is about to leave, it prioritizes displaying limited-time discount information to increase the conversion rate. If the viewer stays for a longer period, it gradually pushes more interactive content to extend the dwell time. Finally, the rendered content is output to the LED screen, and the system continuously monitors changes in viewer sentiment to create a closed-loop feedback loop.

[0045] The display controller uses FPGA or GPU hardware accelerators to achieve low-latency rendering. Understandably, the parallel computing architecture of FPGAs can complete complex rendering tasks in a very short time, such as real-time synthesis of virtual avatar interactive interfaces. During rendering, the color engine dynamically adjusts image parameters based on mood tags, for example, increasing contrast in an "excited" state or reducing the proportion of blue light in a "fatigued" state.

[0046] For example, in a shopping mall advertising screen scenario, when a viewer stops in front of the screen, the system detects an increase in their heart rate using millimeter-wave radar, while the voice module identifies a brisk tone, and edge node fusion data generates an excitement tag. The cloud model, combined with the viewer's historical browsing history (such as frequent clicks on clothing ads), generates a strategy of "extending the display time of clothing ads + high-saturation colors." The FPGA controller completes rendering within milliseconds, and the screen dynamically switches to the corresponding content, significantly improving ad click-through rates.

[0047] In this embodiment, the system determines the viewer's fatigue level and controls the eye protection mode in conjunction with it. It should be noted that PERCLOS (Percentage of Eye Closures) refers to the proportion of time the eyes are closed per unit of time, captured in real-time by a high frame rate camera in the vision module. For example, in a transportation hub scenario, the system uses the YOLOv8 object detection algorithm to locate the viewer's eye area and calculates the proportion of the pupil area covered by the eyelids in consecutive frames. When PERCLOS exceeds a preset threshold, a preliminary assessment of fatigue is made.

[0048] The head posture offset angle is calculated using a 3D structured light sensor, specifically including the pitch and yaw angles. For example, when the viewer's head is continuously lowered (pitch angle greater than the preset range) or frequently and irregularly swaying (yaw angle fluctuates wildly), combining this with the eye closing frequency can enhance the accuracy of fatigue assessment.

[0049] Heart rate variability (HRV) was acquired using non-contact millimeter-wave radar. It is understood that HRV reflects the activity level of the autonomic nervous system, and the ratio of its low-frequency power (LF) to high-frequency power (HF) (LF / HF) can be used to assess fatigue levels. For example, an elevated LF / HF ratio suggests sympathetic nervous system activity, possibly caused by stress-related fatigue. Specifically, the system extracts HRV features through spectral analysis and performs multimodal fusion with visual and head posture data. For instance, in nighttime environments, the weight of visual data is reduced, prioritizing physiological signals for fatigue grading.

[0050] The eye protection mode is triggered by adjustments to optical parameters and optimization of content strategies. Regarding optical parameter adjustments, the peak wavelength of blue light is shifted from 450nm to 460nm, and the 590nm yellow light band is enhanced, reducing the proportion of blue light below a preset threshold to decrease retinal stimulation. Simultaneously, screen brightness is non-linearly and adaptively adjusted according to ambient light, for example, reducing brightness in dim environments to avoid glare.

[0051] Furthermore, non-critical information, such as advertising banners, is hidden, and navigation sign fonts are enlarged with increased line spacing. In airport scenarios, when moderate passenger fatigue is detected, the screen only highlights boarding gate information and a countdown timer, while the rest of the content is dynamically collapsed. Multilingual voice navigation prompts are pushed through directional speakers, such as indicating that the rest area is located on level B2, and AR projection technology is used to project laser arrows onto the sides of the LED screen to guide viewers to the target area.

[0052] In this embodiment, in the scenario of an information screen in a transportation hub, the system analyzes the viewer's eye closure frequency (PERCLOS) and heart rate variability (HRV) to determine if the viewer is moderately fatigued. Edge nodes trigger an eye protection mode, reducing the screen's blue light ratio and highlighting navigation information, while simultaneously guiding the viewer to the rest area via AR projection. If severe fatigue is detected, a red warning box is displayed across the entire screen, and the broadcast system is activated to play a voice reminder to ensure passenger safety.

[0053] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a dynamic adaptation device for the display content of an LED screen, the structure of which is as follows: Figure 2 As shown.

[0054] Figure 2 This is a schematic diagram of the internal structure of an LED screen display content dynamic adaptation device provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor; And, a memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, and the instructions, when executed by at least one processor, enable at least one processor to: Raw feature data of the audience within the radiation range of the LED screen is collected by multimodal biosensors. The raw feature data is preprocessed and desensitized feature vectors are generated through local anonymization. The raw feature data includes facial expression data, voice feature data and physiological signal data. By utilizing a multimodal fusion model in edge computing nodes, the desensitized feature vectors are analyzed in real time to generate primary sentiment labels; The initial emotion tags and associated audience historical behavior data are uploaded to the cloud platform to predict the audience's potential interests through a pre-set emotion deep analysis model, and dynamic content adaptation strategies are generated based on the audience's potential interests. Based on a dynamic content adaptation strategy, display content matching the emotion tag is called from a pre-stored material library to render the display content with low latency, and the color parameters and content display order of the LED display screen are dynamically adjusted according to the emotion analysis results.

[0055] This application also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, can: Raw feature data of the audience within the radiation range of the LED screen is collected by multimodal biosensors. The raw feature data is preprocessed and desensitized feature vectors are generated through local anonymization. The raw feature data includes facial expression data, voice feature data and physiological signal data. By utilizing a multimodal fusion model in edge computing nodes, the desensitized feature vectors are analyzed in real time to generate primary sentiment labels; The initial emotion tags and associated audience historical behavior data are uploaded to the cloud platform to predict the audience's potential interests through a pre-set emotion deep analysis model, and dynamic content adaptation strategies are generated based on the audience's potential interests. Based on a dynamic content adaptation strategy, display content matching the emotion tag is called from a pre-stored material library to render the display content with low latency, and the color parameters and content display order of the LED display screen are dynamically adjusted according to the emotion analysis results.

[0056] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0057] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0058] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0063] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0064] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0065] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0066] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for dynamically adapting the display content of an LED screen, characterized in that, The method includes: The raw feature data of the audience within the radiation range of the LED screen is collected by a multimodal biosensor. The raw feature data is preprocessed and generated into a desensitized feature vector through local anonymization. The raw feature data includes facial expression data, voice feature data and physiological signal data. The desensitized feature vectors are analyzed in real time using a multimodal fusion model in edge computing nodes to generate primary sentiment labels; The primary emotion tags and associated audience historical behavior data are uploaded to the cloud platform to predict potential audience interests through a preset emotion depth analysis model, and dynamic content adaptation strategies are generated based on the potential audience interests. Based on the dynamic content adaptation strategy, display content matching the emotion tag is called from the pre-stored material library to render the display content with low latency, and the color parameters and content display order of the LED display screen are dynamically adjusted according to the emotion analysis results. The system predicts potential audience interests using a pre-defined deep emotion analysis model and, based on these interests, generates a dynamic content adaptation strategy, specifically including: A deep emotion analysis model is built based on the Transformer architecture, and historical behavioral data within the radiation range of the LED screen is obtained; the historical behavioral data includes dwell time and interaction frequency. By using a trained deep emotion analysis model, combined with the historical behavioral data and the primary emotion tags, the potential interests of the audience are predicted, and the emotion type corresponding to the potential interests of the audience is determined. When the emotional type is fatigue, interactive games or dynamic advertisements are pushed to increase the audience's attention; When the emotion type is excitement, the display time of the corresponding advertisement is extended, and a preset high-saturation color is matched for the corresponding advertisement.

2. The method for dynamically adapting the display content of an LED screen according to claim 1, characterized in that, Raw characteristic data of the audience within the radiation range of the LED screen were collected using multimodal biosensors, specifically including: Based on the high frame rate infrared camera and 3D structured light sensor embedded in the LED screen bezel, data on facial expressions, micro-expressions and head postures of the audience within the radiation range of the LED screen are collected. The system collects voice data of the audience within the radiation range of the LED screen using a directional microphone array deployed on the LED screen, and extracts the intonation and speech rate features from the voice data using voiceprint separation technology. The system uses non-contact millimeter-wave radar to detect the heart rate and respiratory rate of spectators within the radiation range of the LED screen, and combines this with a thermal imaging camera to obtain data on the distribution of body surface temperature of spectators within the radiation range of the LED screen.

3. The method for dynamically adapting the display content of an LED screen according to claim 1, characterized in that, The original feature data is preprocessed, and a de-identified feature vector is generated through local anonymization, specifically including: The original feature data is subjected to spatiotemporal alignment and noise filtering; Using a federated learning framework, sensitive information in the preprocessed raw feature data is locally anonymized, retaining only desensitized feature vectors related to emotion recognition; the sensitive information includes facial features. The desensitized feature vector is uploaded to the cloud platform, and the original feature data is automatically destroyed after being stored locally for a preset period of time.

4. The method for dynamically adapting the display content of an LED screen according to claim 1, characterized in that, Using a multimodal fusion model in edge computing nodes, the desensitized feature vectors are analyzed in real time to generate preliminary sentiment labels, specifically including: Based on a multimodal fusion model pre-deployed in edge computing nodes, an attention mechanism is adopted, and the input weights of facial expression data and physiological signal data are adjusted according to the current ambient lighting conditions or audience behavior. In scenarios involving emotional fluctuations, the feature fusion between voice feature data and heart rate data in physiological signal data is prioritized to generate primary emotion labels in real time and trigger a rapid response strategy; the primary emotion labels include excitement, fatigue, and neutrality.

5. The method for dynamically adapting the display content of an LED screen according to claim 1, characterized in that, The LED display screen's color parameters and content display order are dynamically adjusted based on sentiment analysis results, specifically including: The primary emotion labels are mapped to the HSL color space, and the corresponding hue, saturation, and brightness combinations are selected according to the emotion type in the primary emotion labels. The order of ad playback is optimized by using reinforcement learning algorithms to maximize viewer dwell time and interaction probability.

6. The method for dynamically adapting the display content of an LED screen according to claim 1, characterized in that, Low-latency rendering of the displayed content specifically includes: Hardware accelerators are used to render the displayed content in parallel to ensure that the rendering latency is below a preset threshold; During the rendering process, the color engine parameters corresponding to the primary emotion tags are matched in real time to dynamically adjust the brightness and color temperature of the image.

7. The method for dynamically adapting the display content of an LED screen according to claim 1, characterized in that, The method further includes: The audience's eye closure frequency, head posture deviation angle, and heart rate variability were analyzed to determine the audience's fatigue level. The eye protection mode is triggered based on the fatigue level, dynamically reducing the proportion of blue light on the screen and simplifying interface information, while also pushing voice navigation prompts.

8. A device for dynamically adapting the display content of an LED screen, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a dynamic adaptation method for LED screen display content as described in any one of claims 1-7.

9. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they implement the method for dynamic adaptation of LED screen display content as described in any one of claims 1-7.

Images