An autonomous execution method and system of a digital signage terminal

Abstract

This application discloses a method for autonomous execution of a digital signage terminal, a method for cloud-based generation and distribution of digital signage content, a digital signage terminal, a cloud server, and a computer-readable storage medium, relating to the field of digital information publishing technology. The terminal-side method includes: receiving a task instruction message containing a content package address, playback time period, priority, and validity period; writing the task into a local task queue according to the priority; obtaining and caching the content package according to the content package address; identifying the content type of the content to be played during task execution and invoking the corresponding image quality optimization strategy based on the identification result; playing the content according to the playback time period and validity period; and transmitting the execution status back to the cloud. The cloud-side method includes: receiving user input instructions and parsing them into executable tasks; generating digital signage content and performing review, improving the consistency of task distribution, the terminal's autonomous execution capability, and offline business continuity of the digital signage system.

Description

Technical Field

[0001] This application relates to the field of digital information publishing technology, and in particular to an AI-based end-to-cloud collaborative digital signage autonomous execution method, system, and storage medium. Background Technology

[0002] Digital signage is a system that publishes information through display terminal devices. It is widely used in retail supermarkets, transportation hubs, government and enterprise spaces, educational institutions, and other scenarios for purposes such as advertising display, information dissemination, and navigation guidance. With the rapid changes in the business environment and the diversification of user needs, higher requirements are being placed on the content update efficiency, intelligence level, and operational flexibility of digital signage systems.

[0003] Currently, existing digital signage systems primarily employ centralized content management platforms for content creation and distribution. Patent document CN102447672A discloses a multimedia networked digital signage system that communicates with clients via a network. Under the control of playback logic on the network end, the client plays content according to the network's playback logic, supporting various modes including local content, online content, and live streaming. This system achieves remote management and centralized control of content. However, in this technical solution, operators need to manually create materials using professional design tools, upload them through the platform, and configure playback plans. Terminal devices then receive and play the content according to preset rules. While this solution achieves remote content distribution, content creation relies on manual operation, typically taking hours or even days from conception to release, making it difficult to meet the real-time requirements of application scenarios.

[0004] In another technical solution, patent document CN101751819A discloses a digital signage system and method for dynamically updating and remotely monitoring terminals. The playback terminal periodically reads and parses the media content playlist from the platform, comparing the parsed media content information with local media content information. If they differ, the local playlist and media content are updated. This solution employs a client-initiated retrieval communication mode, with the terminal checking the server for new content at fixed time intervals. While this solution achieves automatic content updates to some extent, the polling mechanism causes delays in content delivery, and the terminal device lacks task priority management capabilities, making it unable to handle urgent information dissemination needs. Furthermore, in this solution, the terminal device only acts as a passive execution unit for content reception and playback, lacking local intelligent processing capabilities. All decisions are made in the cloud, resulting in high end-to-end latency and complete failure during network outages.

[0005] Patent document CN110581898A discloses an IoT data terminal system based on 5G and edge computing. An edge computing server located between the IoT data terminal and cloud services builds an AI front-end service platform responsible for pre-recognition, image sampling, deduplication, and data slicing of monitoring data images. This solution offloads image recognition capabilities to the IoT terminal through edge computing, reducing the load on the cloud server and improving the response speed and efficiency of the business front end. However, this solution mainly focuses on protocol parsing at the data transmission layer and does not address technical issues at the business layer such as content generation, review, and intelligent scheduling.

[0006] Existing technologies have the following shortcomings: First, content production efficiency is low, relying on manual production, resulting in a long cycle from demand submission to content launch; second, terminal devices lack intelligent execution capabilities, acting only as passive displays, with all decisions made in the cloud, leading to high end-to-end latency; third, edge-cloud communication mostly uses proprietary protocols, making it difficult for different vendors' systems to interoperate, resulting in a closed ecosystem; furthermore, terminal devices lack local AI processing capabilities, with all intelligent functions relying on the cloud, posing privacy risks and network dependency issues; finally, there is a lack of a complete closed-loop mechanism from content generation to effect feedback, making continuous optimization impossible.

[0007] Therefore, a new technical solution is needed to achieve AI-driven automated content generation, standardized edge-cloud collaborative communication, intelligent execution capabilities on the terminal side, and end-to-end data closed-loop optimization, thereby improving the content production efficiency, intelligence level, and business continuity of digital signage systems. Summary of the Invention

[0008] To address the problems of insufficient real-time task delivery, weak terminal-side execution capabilities, poor business continuity during network outages, and incomplete end-to-end cloud collaboration in existing digital signage systems, this application provides an autonomous execution method for a digital signage terminal, a cloud-based method for generating and delivering digital signage content, a digital signage terminal, a cloud server, and a computer-readable storage medium.

[0009] On the one hand, a method for autonomous execution of a digital signage terminal is provided, applied to a digital signage terminal, the method comprising: The task instruction message is received from the cloud through the end-to-cloud communication interface. The task instruction message includes at least the content package address, playback time period, priority and validity period. The task instruction message is parsed, and the corresponding task is written to the local task queue according to the priority. The corresponding content package is obtained based on the content package address and stored in the local cache; During the execution of the task, content type identification is performed on the content to be played, and the corresponding image quality optimization strategy is invoked based on the identified content type; Play the content according to the stated playback period and validity period; The execution status of the task is transmitted back to the cloud via the end-to-cloud communication interface.

[0010] In some implementations, the task instruction message includes a message header and a message body. The message header includes at least a message type, a message identifier, a timestamp, and a device identifier. The message body includes at least a content package address, a playback period, a priority, and a validity period.

[0011] In some implementations, writing the corresponding task into the local task queue according to the priority includes: writing the task into the corresponding queues in the emergency task queue, high-priority task queue, medium-priority task queue, and low-priority task queue respectively; when an emergency task is received, interrupting the current normal playback and executing the emergency task; and resuming the interrupted normal playback after the emergency task is executed.

[0012] In some implementations, obtaining the corresponding content package based on the content package address and storing it in the local cache includes: initiating a content package download request to the content delivery network to obtain a compressed content package; decompressing the downloaded content package and storing the decompressed content files and configuration files in the local cache directory.

[0013] In some implementations, the local cache employs a cache eviction mechanism based on a least recently used strategy; when the cache capacity exceeds a preset threshold, cached content is evicted according to the least recently used order; and, content corresponding to tasks that have expired is either not cached or deleted from the cache.

[0014] In some implementations, the method further includes: preloading content to be played within a future preset time window into the local cache, so as to continue offline playback based on the local cache when the network is abnormal; after the network is restored, sending a synchronization request to the cloud, receiving the latest task list returned by the cloud, comparing the local task list with the latest task list, downloading the content package corresponding to the newly added task, and deleting the content package corresponding to the invalid task.

[0015] In some implementations, the step of performing content type identification on the content to be played and calling the corresponding image quality optimization strategy according to the identified content type includes: inputting a thumbnail or video frame of the content to be played into the content type identification model to obtain a content type label; and selecting a corresponding strategy from a variety of image quality optimization strategies to optimize the content to be played based on the content type label.

[0016] On the other hand, a method for cloud-based generation and distribution of digital signage content is provided, applied to a cloud server, the method comprising: Receive user input instructions and parse the user input instructions into executable tasks; Digital signage content is generated based on the executable task; The generated digital signage content will be reviewed; After the review is approved, a task instruction message is generated, which includes at least the content package address, playback time period, priority, and validity period. The task instruction message is sent to the target digital signage terminal through the end-to-cloud communication interface; Receive the execution status returned by the target digital signage terminal; Adjust the strategy for subsequent tasks based on the execution status.

[0017] In some implementations, parsing the user input instruction into an executable task includes: performing intent recognition and entity extraction on the user input instruction through a basic business capability unit to obtain a structured intent representation; mapping the structured intent representation into a scenario task through a scenario-based combination capability unit; and decomposing the scenario task through an intelligent decision-making unit to generate task parameters corresponding to the task instruction message.

[0018] In some implementations, the user input instructions include at least one of natural language input, image input, and voice input; for different input modalities, corresponding recognition processing is performed respectively and then converted into a unified task parameter representation.

[0019] In some implementations, the review of the generated digital signage content includes: performing a brand compliance check; performing a content security review; and determining the review result based on the brand compliance check result and the content security review result.

[0020] In some implementations, performing the brand compliance check includes: reading pre-configured brand parameters; detecting color parameters, logo position parameters, and font parameters in the digital signage content; and determining the brand compliance check result based on the comparison results of the color parameters, logo position parameters, and font parameters with the pre-configured brand parameters.

[0021] In some implementations, the content security audit includes: performing sensitive word detection on the text content based on a sensitive word library; performing advertising law-restricted word detection on the text content based on a restricted word library; performing compliance detection on the image content based on an image recognition model; and when preset risky content is detected, transferring the corresponding content to a manual review process and recording the review log.

[0022] In some implementations, the end-to-cloud communication interface adopts a plug-in protocol adaptation architecture; by dynamically loading protocol adapter plug-ins, external protocol messages are converted into task instruction messages or the execution status returned by the terminal is converted into a data format that can be processed by the cloud.

[0023] In some implementations, adjusting the subsequent task strategy based on the execution status includes: adjusting content generation parameters or publishing strategies according to the playback completion rate; adjusting the content preloading strategy of the target digital signage terminal according to the network connection status; and / or adjusting the model publishing scope according to the terminal-side model execution status.

[0024] On another front, a digital signage terminal is provided, including a communication interface, a memory, and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the digital signage terminal performs the aforementioned autonomous execution method.

[0025] On another front, a cloud server is provided, including a communication interface, a memory, and a processor. The memory stores a computer program, which, when executed by the processor, causes the cloud server to perform the aforementioned method for generating and distributing digital signage content in the cloud.

[0026] On another front, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the above-mentioned autonomous execution method of the digital signage terminal or the cloud-based generation and distribution method of digital signage content.

[0027] Beneficial effects Compared with the prior art, one or more embodiments of the above technical solutions have at least the following beneficial effects: By including the content package address, playback time period, priority, and validity period in the task instruction message, the terminal can complete task parsing, queuing, and execution based on unified task parameters, thereby improving the consistency between task issuance and terminal execution.

[0028] By setting up local task queues, cache management, and offline synchronization mechanisms on the terminal side, the terminal can continue playing based on the local cache even in the event of network anomalies, thereby improving business continuity.

[0029] By performing content type identification and invoking corresponding image quality optimization strategies on the terminal side, different types of content can be displayed using corresponding display processing methods, thereby improving the display effect of digital signage terminals.

[0030] By parsing user input commands, generating content, reviewing and issuing them through cloud servers, and receiving execution status feedback from terminals to adjust strategies, a closed-loop task system for end-to-cloud collaboration is formed.

[0031] The protocol adapter plugin enables the conversion between external protocol messages and standard task messages, allowing the system to adapt to different communication protocols and thus improve the scalability of the edge-cloud collaborative system. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall architecture of the edge-cloud collaborative AI digital signage autonomous execution system.

[0033] Figure 2 This is a schematic diagram of the three-layer architecture of the cloud-based AI decision-making center.

[0034] Figure 3 This is a diagram illustrating the three-level review and release process.

[0035] Figure 4 This is a schematic diagram of the terminal task queue scheduling process.

[0036] Figure 5 This is a schematic diagram of the entire AI closed loop. Detailed Implementation

[0037] The specific embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that, in the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. In this embodiment, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features. Therefore, features defined with "first" and "second" can explicitly or implicitly include one or more features. In the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0038] In this application embodiment, "cloud server" refers to a computing device used to perform task parsing, content generation, content review, task message generation, task distribution, and strategy adjustment; "digital signage terminal" refers to a display terminal device used to receive task instruction messages, download content packages, perform content playback, optimize local display, and return execution status; "task instruction message" refers to a structured message sent by the cloud server to the digital signage terminal; "execution status message" refers to task execution-related messages returned by the digital signage terminal to the cloud server; "content package" refers to a set of resources for terminal playback, wherein the resource set includes at least media files and configuration files. Edge-cloud collaboration refers to a system architecture where the cloud and terminal devices transmit commands and interact with data through network protocols; MCP stands for Model Context Protocol, a standardized edge-cloud communication protocol; NPU stands for Neural Processing Unit, dedicated to AI inference computation; LRU stands for Least Recently Used, used for cache eviction strategies; CDN stands for Content Delivery Network; TOPS stands for Tera Operations Per Second, a unit of measurement for the computing power of AI chips.

[0039] In one implementation, after generating a task instruction message, the cloud server sends it to the digital signage terminal via an end-to-cloud communication interface. The task instruction message includes a message header and a message body.

[0040] The message header includes at least: message type, message identifier, timestamp, and device identifier.

[0041] The message body includes at least: content package address, playback time period, priority, and validity period.

[0042] Among them, the message type is used to identify whether the current message is a task issuance message, a status query message, or a synchronization response message; the message identifier is used to uniquely identify the current message so that the terminal can reference it when receiving a receipt; the timestamp is used to indicate the message generation time; and the device identifier is used to indicate the target digital signage terminal.

[0043] The content package address is used to instruct the terminal to obtain the download address of the content to be played; the playback period is used to instruct the time range within which the content is allowed to be played; the priority is used to instruct the scheduling level when the task enters the local task queue; and the validity period is used to limit the effective time and expiration time of the task.

[0044] In one implementation, the digital signage terminal sets a task status for the received task. The task status includes at least: received, queued, downloading, cached, pending execution, executing, completed, expired, and execution failed.

[0045] When the terminal receives a task instruction message and completes message validity verification, it sets the task status to "received"; after writing the message to the corresponding task queue according to priority, it sets the task status to "enqueued"; when retrieving the content package based on the content package address, it sets the task status to "downloading"; after the content package is downloaded and decompressed, it sets the task status to "cached"; when the playback period is met and the current time is within the validity period, it sets the task status to "pending execution"; when playback starts, it sets the task status to "execution"; when playback ends, it sets the task status to "completed"; when the current time exceeds the validity period, it sets the task status to "expired"; when content download fails, content verification fails, or playback is abnormally interrupted, it sets the task status to "execution failed".

[0046] The cloud server can determine the execution progress of the corresponding task based on the task status reported by the terminal, and adjust the subsequent distribution strategy accordingly.

[0047] In one implementation, the digital signage terminal maintains a local cache directory for storing content packets obtained from the content delivery network.

[0048] After receiving the task instruction message, the terminal initiates a download request to the content delivery network based on the content package address to obtain the compressed content package; after the download is completed, the content package is decompressed to obtain media files and configuration files, and the media files and configuration files are written to the local cache directory.

[0049] To control cache capacity, the terminal performs cache eviction processing on the content in the local cache directory. This eviction process is based on a Least Recently Used (LRU) strategy; that is, when the cache capacity exceeds a preset threshold, cached content with the earliest access time is deleted first. For content corresponding to tasks that have expired, the terminal either stops caching or deletes it from the cache directory.

[0050] When the network is normal, the terminal preloads content to be played within a preset time window in the future; when the network is abnormal, the terminal continues to play content based on the cached content; after the network is restored, the terminal sends a synchronization request to the cloud server, receives the latest task list, and downloads the content corresponding to the newly added tasks and deletes the content corresponding to the invalid tasks by comparing the local task list with the latest task list.

[0051] In one implementation, the digital signage terminal performs content type identification on the content to be played before or during playback.

[0052] For static image content, the terminal extracts a thumbnail of the static image content as input to the content type recognition model; for video content, the terminal extracts video frames as input to the content type recognition model.

[0053] The content type recognition model outputs content type labels, which include at least one of the following: plain text, mixed text and images, photos, videos, and charts.

[0054] The terminal selects the corresponding image quality optimization strategy from a variety of pre-configured image quality optimization strategies based on the content type tag.

[0055] For example, when the content type tag is plain text, the sharpening intensity and contrast parameters are increased; when the content type tag is a mix of text and images, optimized parameters that balance text edge clarity and image color reproduction are used; when the content type tag is video, dynamic display optimization parameters for video content are invoked.

[0056] The above methods enable the terminal to use differentiated display processing for different content types.

[0057] In one implementation, the model files in the digital signage terminal employ a dual-slot storage mechanism. The dual slots include a first slot and a second slot, where one slot stores the currently running model and the other slot stores the model to be updated.

[0058] When the cloud server issues a model update task, the terminal writes the new model into a slot not occupied by the currently running model. After the model file is written, integrity and executability checks are performed on the new model. If the checks pass, the inference call entry is switched to the slot storing the new model. If the checks fail, the currently running model in the original slot continues to perform inference.

[0059] Therefore, during the model update process, the terminal's content recognition and image quality optimization functions continue to operate continuously.

[0060] In one implementation, the cloud server generates audit log records during the manual review process.

[0061] Each audit log entry should include at least the audit time, auditor ID, audit target ID, audit result, and risk hits.

[0062] Audit logs are appended to the log storage area, and a verification digest is generated for each log record after it is written.

[0063] When reading the audit log, the integrity of the log content is verified based on the verification summary.

[0064] By using the methods described above, the risk of log records being unrecognizable after being abnormally modified can be reduced.

[0065] In one implementation, the cloud server receives an execution status message from the digital signage terminal. The execution status message includes at least one of the following: device identifier, current playback task identifier, playback start time, playback end time, playback completion rate, inference latency, remaining cache capacity, and network connection status.

[0066] The strategy optimization unit performs a strategy update based on the execution status message.

[0067] For example, when the playback completion rate is lower than a preset threshold, adjust the release time or content complexity parameters of subsequent tasks; when the network connection status indicates that the network environment of the corresponding terminal is unstable, increase the preloading time window of the corresponding task for that terminal; when the terminal reports an abnormal execution status of a new model, reduce or suspend the distribution range of the corresponding model version.

[0068] The updated strategy parameters are used in subsequent task generation, content publishing, or model publishing processes.

[0069] See Figure 1 As shown, Figure 1 This is a schematic diagram of the overall architecture of an edge-cloud collaborative AI digital signage autonomous execution system. The system includes a cloud-based intelligent decision-making module, a standardized edge-cloud communication interface, a terminal intelligent execution module, an edge-side AI processing unit, a feedback data acquisition unit, and a strategy optimization unit. (See also...) Figure 2 As shown, Figure 2 This is a schematic diagram of the three-layer architecture of the cloud-based AI decision-making center, from bottom to top: basic business capability unit, scenario-based combined capability unit, and intelligent decision-making unit. (See also...) Figure 3 As shown, Figure 3 This is a diagram illustrating the three-tiered review and release process, including brand compliance checks, content security reviews, and approval level determination. (See also...) Figure 4 As shown, Figure 4 This is a schematic diagram of the terminal task queue scheduling process, illustrating the scheduling mechanism for MCP instruction reception, priority queuing, emergency interruption, and execution. See also... Figure 5 As shown, Figure 5 This is a diagram illustrating the entire AI closed loop, showing the complete data flow from user command input, AI content generation, three-level review, MCP issuance, terminal execution, status feedback to strategy optimization.

[0070] Example 1: Chain Retail Supermarket Scenario This embodiment provides an edge-cloud collaborative AI-native digital signage autonomous execution system applied to chain retail supermarket scenarios. The terminal hardware environment consists of a 55-inch commercial signage unit equipped with an MTK520 chip. The MTK520 chip integrates a 9 TOPS NPU, operates at 5V, supports the Android AOSP operating system, is equipped with an ambient light sensor for adaptive brightness adjustment, has 32GB of storage capacity, and supports both Ethernet and Wi-Fi connectivity. The cloud-based OpenClaw intelligent architecture is deployed on a Linux server cluster, integrating a large language model for natural language understanding, an image generation API for multimedia content creation, and a brand parameter management database storing brand specifications such as primary and secondary colors, logo, and fonts. A three-level review engine and an effect analysis module are also deployed.

[0071] See Figure 2 As shown, the cloud-based intelligent decision-making module adopts a three-layer architecture. The basic business capability unit is located at the bottom layer, providing atomic-level AI capabilities, including a natural language understanding unit, an image generation unit, a video editing unit, a text generation unit, and a data analysis unit. The natural language understanding unit receives text commands input by the user, performs intent recognition and entity extraction through a large language model, and outputs a structured intent representation. The image generation unit calls image generation APIs such as Stable Diffusion or DALL-E to generate the main visual image based on the text description. The video editing unit supports video clip editing, transition effect addition, and subtitle overlay. The text generation unit generates marketing copy based on the GPT model. The data analysis unit performs statistical analysis on the playback data returned from the terminal. The scenario-based capability unit, located in the middle layer, combines basic capabilities into scenario-based tasks, including promotional poster generation, new product launch, emergency notification, and membership marketing tasks. The promotional poster generation task combines text and image generation units to automatically generate text and images based on the promotional theme. The new product launch task combines image and video editing units to generate new product demonstration videos. The emergency notification task uses the text generation unit to quickly generate warning information. The membership marketing task combines the data analysis unit to identify target member groups and generate personalized content. The intelligent decision-making unit, located at the top layer, is responsible for intent understanding and task decomposition. It receives natural language instructions from operations personnel, identifies user intent through the intent parsing engine, and breaks down complex instructions into multiple sub-tasks. For example, if an operations personnel inputs "Push spring promotional posters to all stores in East China, playing them daily from 9 AM to 9 PM," the intelligent decision-making unit identifies the intent as "content generation + scheduling + device targeting," breaking it down into four sub-tasks: AI-generated spring promotional posters, brand compliance check + content security review, matching East China store terminals, and generating a scheduling plan.

[0072] See Figure 3As shown, AI-generated content undergoes a three-tiered review mechanism. The brand compliance check unit reads pre-configured brand parameters from the brand parameter management database, including the primary color RGB value, secondary color RGB value, standard logo layout, and font specifications. The brand compliance check unit performs pixel-level inspection on the AI-generated graphic content, extracts the proportion of the primary color in the content, and determines whether the color difference between the primary color RGB value and the pre-configured parameters is within the threshold range (the color difference ΔE should not exceed 5). It also checks whether the logo is placed according to the standard layout and whether the font conforms to the brand specifications. If the detected color deviation exceeds the threshold, the logo position is incorrect, or the font does not conform to the specifications, the brand compliance is deemed unsatisfactory, and the content is returned for modification. The content security review unit performs sensitive word detection on text content. The sensitive word database includes politically sensitive words, violent and terrorist words, vulgar words, etc. It also performs advertising law-restricted word detection on text content. The advertising law-restricted word database includes words prohibited by advertising law such as "most", "first", "top", "ultimate", etc. It performs compliance detection on image content. It uses an image recognition model to detect whether the image contains prohibited content. If sensitive words, extreme words, or images are detected as violating regulations, the content is deemed to have failed the security review and enters the mandatory manual review process. The approval level determination unit automatically determines the approval level based on the results of brand compliance checks and content security audits. If both the brand compliance check and the content security audit pass, it is automatically approved, and the content directly enters the publication process. If the brand compliance check passes but the content security audit detects low-risk words, it is determined to require manual confirmation, and the content is pushed to the reviewer for manual review. The reviewer can choose to approve or reject. If the content security audit detects sensitive content, it is determined to require mandatory manual review, and the content must be manually reviewed before it can be published. Even the super administrator cannot bypass this review process. The review log is fully recorded and cannot be tampered with or deleted. The content security audit log is retained for no less than one year. The log records include the review time, reviewer ID, review results, detected sensitive or extreme words, review comments, etc.

[0073] The standardized edge-cloud communication interface uses the MCP protocol to implement command and data transmission. The MCP protocol client is deployed on the Android system of the terminal label, while the MCP protocol server is deployed in the cloud OpenClaw architecture. The MCP protocol uses WebSocket as the underlying transport protocol to establish a long connection to maintain real-time communication between the edge and the cloud. The MCP protocol defines a standard message format, with the message header containing the message type, message ID, timestamp, and device ID, and the message body containing specific command parameters or data content. The MCP protocol supports a plug-in architecture, with each standardized protocol corresponding to a protocol adapter plugin. The protocol adapter plugin implements protocol parsing and data conversion functions. The cloud MCP server supports dynamically loading protocol adapter plugins. When a new communication protocol needs to be supported, only the corresponding protocol adapter plugin needs to be developed and loaded through a configuration file, without modifying the core execution engine. For example, when supporting the Azure IoT Hub protocol, an Azure IoT Hub protocol adapter plugin is developed. The plugin implements message parsing for the Azure IoT Hub protocol and message conversion for the MCP protocol. After loading the plugin through the configuration file, the cloud can communicate with the terminal through the Azure IoT Hub protocol.

[0074] Approved task instructions are sent to the terminal signage via the MCP protocol. The task instruction includes metadata such as the content package URL, playback time period, priority, and validity period. The content package URL points to the content resource on the CDN. The content resource is a ZIP compressed package containing image files, video files, JSON configuration files, etc. The playback time period defines the playback time range of the content; for example, "09:00-21:00" means playback from 9:00 to 21:00 every day. The priority defines the task's execution priority, which is divided into four levels: urgent, high-priority, medium-priority, and low-priority. The validity period defines the task's effective and expiration time; for example, "2024-03-01 00:00:00 to 2024-03-07 23:59:59" means the task is effective at 00:00 on March 1, 2024, and expires at 23:59:59 on March 7, 2024. After expiration, the task is automatically removed from the playlist.

[0075] See Figure 4As shown, the terminal sign receives and parses task instructions through the MCP client. The MCP client runs in the background service of the Android system, listening for WebSocket long connections. When a task instruction message is received, the MCP client parses the message header and message body, extracts the metadata of the task instruction, and arranges the task instructions into the local task queue according to priority. The task queue manager adopts a four-level priority scheduling strategy. The emergency task queue, high-priority task queue, medium-priority task queue, and low-priority task queue each maintain their own task lists. The task queue manager schedules tasks according to the following rules: an emergency task is immediately interrupted and executed upon receipt, and normal playback resumes after the emergency task finishes playing; high-priority tasks play twice per round, that is, within one playback cycle, high-priority tasks play twice, medium-priority tasks play once, and low-priority tasks are played based on the round number; medium-priority tasks play once per round; low-priority tasks play once every three rounds, that is, low-priority tasks play once every three playback cycles; expired content is automatically removed from the playlist. The task queue manager periodically checks the validity period of tasks, and if the current time exceeds the expiration time of the task, the task is removed from the queue.

[0076] The terminal signage retrieves content packages from the CDN and caches them locally. The content cache manager, based on the content package URL in the task instruction, initiates an HTTP GET request to the CDN, downloads the ZIP archive, decompresses it, and stores it in the local cache directory. The content cache manager employs an LRU eviction policy. When the cache capacity exceeds 30% of the terminal storage, older content is evicted according to the least recently used principle. Specifically, the content cache manager maintains a cache list where each entry contains a content ID, content size, last access time, and access count. When content needs to be evicted, it is sorted in ascending order of last access time, prioritizing the eviction of content with the earliest last access time. Expired content is not cached. Before downloading content, the content cache manager checks the task's expiration date; if the current time has exceeded the task's expiration time, the content is not downloaded. The terminal signage supports playback for at least 48 hours based on local caching even in offline environments. The content cache manager preloads content to be played within the next 48 hours, ensuring that the terminal can continue playing for 48 hours based on local caching even if the network is interrupted. After the network is restored, the content cache manager automatically synchronizes the latest content. The synchronization process is as follows: the MCP client sends a synchronization request to the cloud, the cloud returns the latest task list, the content cache manager compares the local task list and the cloud task list, downloads the content package for newly added tasks, and deletes the content package for expired tasks.

[0077] The terminal signage performs content type recognition and intelligent image quality optimization through the on-device NPU inference framework. The edge AI processing unit is integrated into the NPU of the MTK520 chip. The NPU provides 9 TOPS of computing power and supports INT8 quantization inference. The edge AI processing unit deploys a content type recognition model and an image quality optimization model. The content type recognition model is a lightweight convolutional neural network. The model input is a thumbnail of the content with a resolution of 224x224 pixels. The model output is a content type label, including plain text, mixed text and images, photos, videos, and charts. The model inference latency is less than 50 milliseconds. The image quality optimization model automatically switches image quality strategies according to the content type. For plain text content, the image quality optimization model improves contrast and sharpness to ensure clear and readable text. For mixed text and images, the image quality optimization model balances text clarity and image color reproduction. For photos, the image quality optimization model performs color correction and noise reduction to improve the visual effect of the photos. For videos, the image quality optimization model performs deinterlacing and frame rate compensation to ensure smooth video playback. For charts, the image quality optimization model improves contrast and color saturation to ensure clear and legible chart data. The edge AI processing unit supports hot updates and canary releases of model slots A and B. The storage space of the MTK520 chip is divided into model slot A and model slot B. The currently running model is stored in model slot A, and the new model is stored in model slot B after download. During the model update process, the edge AI processing unit continues to use the old model in model slot A for inference without affecting playback. After the new model is downloaded, the edge AI processing unit performs model verification. If the verification passes, it switches to the new model in model slot B. If the verification fails, it automatically rolls back to the old model in model slot A. The canary release strategy is as follows: the cloud first distributes the new model to some terminals, monitors the inference latency and accuracy of the new model. If the new model performs normally, the distribution range is gradually expanded. If the new model is abnormal, the distribution is stopped and rolled back.

[0078] The terminal signage reports its execution status and playback data to the cloud in real time. The status reporting module sends status reporting messages to the cloud via the MCP protocol. These messages include data such as device ID, current playback task ID, playback start time, playback end time, playback completion rate, NPU inference latency, remaining content cache capacity, and network connection status. Status reporting occurs every 5 minutes, or immediately upon switching playback tasks. See also... Figure 5As shown, the cloud-based performance analysis module receives status data from the terminal, performs performance analysis and strategy optimization. The performance analysis module statistically analyzes the playback completion rate, identifies tasks with a completion rate below 90%, and analyzes the reasons, including network interruptions, content loading failures, and terminal malfunctions. The performance analysis module compares the playback performance of different content, such as comparing the number of plays and completion rates of different promotional posters, and identifies the characteristics of well-performing content, including color schemes, copywriting style, and visual layout. Based on the performance analysis results, the strategy optimization unit dynamically optimizes content strategies. For example, if a promotional poster with a certain color scheme has a high completion rate, that color scheme will be prioritized in subsequent content generation. If an unstable network connection at a store causes frequent playback interruptions, the content preloading time for that store will be increased to ensure offline playback capabilities.

[0079] Through the above technical solution, this embodiment achieves end-to-end AI automation from natural language commands to signage playback, significantly improving content production efficiency. Operators only need to input natural language commands, and the cloud-based AI decision-making center automatically completes intent parsing, content generation, three-level review, and task assignment, eliminating the need for manual content creation. The three-level automated review mechanism ensures that AI-generated content complies with brand guidelines and laws and regulations. Brand compliance checks ensure that color schemes, logos, and fonts conform to pre-configured parameters through pixel-level detection. Content security reviews ensure content legality and compliance through sensitive word detection and extreme word detection. A mandatory manual review mechanism ensures zero omission of sensitive content. Full and tamper-proof audit logs meet the enterprise's compliance traceability needs. The standardized MCP protocol enables standardized end-to-cloud command transmission and data interaction. The plug-in protocol adaptation architecture supports dynamic loading of multiple communication protocols, effectively solving the system closure problem caused by proprietary protocols and reducing system integration costs. The on-device NPU inference framework performs content type recognition and image quality optimization on the device side, effectively reducing processing latency. User privacy data is processed locally without uploading to the cloud, meeting privacy compliance requirements. The LRU cache management strategy and offline playback capability support continued playback based on local cache in offline environments, effectively ensuring business continuity. The latest content is automatically synchronized after the network is restored, solving the problem of black screen when the network is disconnected in existing solutions. The full-link AI closed loop constructs a complete data flow from instruction reception, content generation, review and release, terminal execution to status feedback. The cloud continuously optimizes decision-making strategies based on the backhauled data, realizing the system's continuous self-evolution capability.

[0080] Example 2: Emergency Information Distribution Scenario for Government and Enterprises The difference from Example 1 is that this example is applied to a government-enterprise emergency information dissemination scenario to verify the emergency task interruption mechanism. The operator inputs "distribute typhoon warning emergency information to all locations in the government-supporting commercial space." The cloud-based intelligent decision-making unit identifies the intent as "emergency information release + targeted delivery," breaking it down into three sub-tasks: AI-generated typhoon warning text and image content, brand compliance check + content security audit, and matching with government-supporting commercial space terminals. The AI ​​content generation module calls the text generation unit to generate the typhoon warning text, which includes information such as the typhoon name, warning level, affected area, and preventative measures. It also calls the image generation unit to generate a typhoon path map and prevention guidance map. The brand compliance check and content security audit pass. The task instruction is distributed to all terminals in the government-supporting commercial space via the MCP protocol. The task instruction is prioritized as an emergency task and includes an effective time, expiration time, and a loop playback marker.

[0081] Upon receiving an emergency task instruction, the terminal MCP client immediately interrupts the current regular playback, inserts the emergency task at the top of the queue, and executes it immediately. The currently playing promotional poster is interrupted, and the signage switches to a loop of typhoon warning content. All terminals complete the switch within 60 seconds of receiving the emergency task instruction, achieving rapid delivery of emergency information. The status reporting module reports the playback status to the cloud in real time. The cloud-based effect analysis module monitors the playback status of all terminals. If a terminal fails to complete the switch within 60 seconds, it proactively issues a warning, allowing operators to remotely troubleshoot terminal malfunctions, including network interruptions, device offline issues, and system anomalies. After the typhoon warning is lifted, the task instruction's expiration time expires, and the task queue manager automatically removes the typhoon warning task from the queue. The signage resumes its regular playback schedule, continuing to play promotional posters and other regular content.

[0082] This embodiment verifies the effectiveness of the emergency task interruption mechanism. Emergency tasks can reach the entire network within 60 seconds, meeting the real-time requirements for emergency information dissemination. Compared with existing solutions that rely on manual switching of content on each terminal, which usually takes several hours, this embodiment significantly improves the efficiency of emergency task issuance and effectively ensures the timely transmission of public safety information.

[0083] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and not by the embodiments of this application.

Claims

1. A method for autonomous execution of a digital signage terminal, characterized in that, Applied to digital signage terminals, the method includes: The task instruction message is received from the cloud through the end-to-cloud communication interface. The task instruction message includes at least the content package address, playback time period, priority and validity period. The task instruction message is parsed, and the corresponding task is written to the local task queue according to the priority. The corresponding content package is obtained based on the content package address and stored in the local cache; During the execution of the task, content type identification is performed on the content to be played, and the corresponding image quality optimization strategy is invoked based on the identified content type; Play the content according to the stated playback period and validity period; The execution status of the task is transmitted back to the cloud via the end-to-cloud communication interface.

2. The method according to claim 1, characterized in that, The task instruction message includes a message header and a message body. The message header includes at least the message type, message identifier, timestamp, and device identifier. The message body includes at least the content package address, playback period, priority, and validity period.

3. The method according to claim 1, characterized in that, The step of writing the corresponding task into the local task queue according to the priority includes: Write the tasks into the corresponding queues in the urgent task queue, high priority task queue, medium priority task queue, and low priority task queue, respectively. Upon receiving an emergency task, interrupt the current regular playback and execute the emergency task; Once the emergency task is completed, resume normal playback after the interruption.

4. The method according to claim 1, characterized in that, The step of obtaining the corresponding content package based on the content package address and storing it in the local cache includes: Initiate a content package download request to the content delivery network to obtain a compressed content package; The downloaded content package is decompressed, and the decompressed content files and configuration files are stored in the local cache directory.

5. The method according to claim 4, characterized in that, The local cache employs a cache eviction mechanism based on the least recently used strategy; When the cache capacity exceeds a preset threshold, cached content is evicted in the least recently used order. Furthermore, content for tasks that have expired will not be cached or will be removed from the cache.

6. The method according to claim 5, characterized in that, The method further includes: The content to be played within a future preset time window is preloaded into the local cache so that offline playback can continue based on the local cache in the event of network failure. After the network is restored, a synchronization request is sent to the cloud, the latest task list is received from the cloud, the local task list is compared with the latest task list, the content package corresponding to the newly added task is downloaded, and the content package corresponding to the invalid task is deleted.

7. The method according to claim 1, characterized in that, The process of performing content type identification on the content to be played and invoking the corresponding image quality optimization strategy based on the identified content type includes: Input the thumbnail or video frame of the content to be played into the content type recognition model to obtain the content type label; Based on the content type tags, select the corresponding strategy from a variety of image quality optimization strategies to optimize the content to be played; The content type includes at least one of the following: plain text, mixed text and images, photos, videos, and charts.

8. The method according to claim 1, characterized in that, The execution status includes at least one of the following: device identifier, current playback task identifier, playback start time, playback end time, playback completion rate, inference latency, remaining cache capacity, and network connection status. Furthermore, the backhaul is at least one of periodic backhaul and task switching triggered backhaul.

9. The method according to claim 7, characterized in that, The method further includes a model hot update step, which includes: Write the new model to a spare slot that is different from the slot where the currently running model is located; After the new model passes verification, switch to the new model in the spare slot; If the new model fails to be validated, the current running model will continue to perform inference.

10. A method for cloud-based generation and distribution of digital signage content, characterized in that, Applied to a cloud server, the method includes: Receive user input instructions and parse the user input instructions into executable tasks; Digital signage content is generated based on the executable task; The generated digital signage content will be reviewed; After the review is approved, a task instruction message is generated, which includes at least the content package address, playback time period, priority, and validity period. The task instruction message is sent to the target digital signage terminal through the end-to-cloud communication interface; Receive the execution status returned by the target digital signage terminal; Adjust the strategy for subsequent tasks based on the execution status.

11. The method according to claim 10, characterized in that, The step of parsing the user input command into an executable task includes: The basic business capability unit performs intent recognition and entity extraction on the user input command to obtain a structured intent representation; The structured intent representation is mapped to a scenario task through scenario-based combination capability units; The intelligent decision-making unit decomposes the scenario task to generate the task parameters corresponding to the task instruction message.

12. The method according to claim 11, characterized in that, The user input instructions include at least one of natural language input, image input, and voice input; For different input modalities, the corresponding recognition processing is performed and then converted into a unified task parameter representation.

13. The method according to claim 10, characterized in that, The review of the generated digital signage content includes: Conduct brand compliance checks; Perform content security audits; The audit results will be determined based on the brand compliance inspection results and the content security audit results.

14. The method according to claim 13, characterized in that, The aforementioned brand compliance checks include: Read the pre-configured brand parameters; Detect the color scheme parameters, logo position parameters, and font parameters in the content of the digital signage; The brand compliance check result is determined based on the comparison results between the color scheme parameters, logo position parameters, and font parameters and the pre-configured brand parameters.

15. The method according to claim 13, characterized in that, The security audit of the executed content includes: Perform sensitive word detection on text content based on a sensitive word database; Perform advertising law-related extreme word detection on text content based on the extreme word database; Compliance checks are performed on image content based on image recognition models; When content that triggers a pre-set risk is detected, the corresponding content will be transferred to a manual review process, and a review log will be recorded.

16. The method according to claim 10, characterized in that, The end-to-cloud communication interface adopts a plug-in protocol adaptation architecture. By dynamically loading protocol adapter plugins, external protocol messages are converted into the task instruction messages, or the execution status returned by the terminal is converted into a data format that can be processed by the cloud.

17. The method according to claim 10, characterized in that, The adjustment of subsequent task strategies based on the execution status includes: Adjust content generation parameters or publishing strategies based on playback completion rate; Adjust the content preloading strategy of the target digital signage terminal according to the network connection status; And / or adjust the model release scope based on the model execution status on the terminal side.

18. A digital signage terminal, characterized in that, It includes a communication interface, a memory, and a processor. The memory stores a computer program, which, when executed by the processor, causes the digital signage terminal to perform the method described in any one of claims 1 to 9.

19. A cloud server, characterized in that, It includes a communication interface, a memory, and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it causes the cloud server to perform the method of any one of claims 10 to 17.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 9.

21. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 10 to 17.

Images