AI Control Method for Multimedia Exhibition Halls Based on Distributed Device Collaboration and Task Scheduling

Abstract

This invention discloses an AI control method for multimedia exhibition halls based on distributed device collaboration and task scheduling. The method includes: collecting visitor flow data through a multimodal sensing terminal; generating predictions of visitor density, movement paths, and interaction needs using a visitor flow situation perception and prediction algorithm; allocating tasks and scheduling resources for distributed devices using a distributed device dynamic task game scheduling algorithm based on the predictions; dynamically adjusting display content parameters according to device status and visitor flow needs using a multimedia content intelligent adaptation rendering model; and monitoring and handling various anomalies during task execution through an intelligent fault-tolerant processing platform for parallel task interaction in the exhibition hall. By combining specific hardware configurations and software parameter optimization techniques, a dynamic optimization control mechanism is formed. This invention effectively improves the collaborative efficiency of distributed devices, the flexibility of content adaptation, and the system's fault tolerance, achieving intelligent and efficient AI interactive control for multimedia exhibition halls.

Description

Technical Field

[0001] This invention relates to the field of interactive control technology for multimedia exhibition halls, and in particular to an AI control method for multimedia exhibition halls based on distributed device collaboration and task scheduling. Background Technology

[0002] As multimedia exhibition halls upgrade towards intelligence and interactivity, the dynamic changes in pedestrian flow and the need for coordinated distributed equipment are becoming increasingly prominent. This necessitates precise pedestrian flow perception, efficient task scheduling, and flexible content adaptation to achieve a high-quality interactive experience. Currently, AI interactive control in multimedia exhibition halls largely relies on single sensing methods and fixed task allocation patterns, lacking dynamic prediction of pedestrian flow and global optimization of equipment resources. This makes it difficult to adapt to the complex and ever-changing operating scenarios of exhibition halls. Therefore, there is an urgent need to build a collaborative control system that integrates multiple algorithms and intelligent platforms to solve problems such as low equipment coordination efficiency and inaccurate interactive responses.

[0003] Currently, basic pedestrian flow data is typically collected using distributed sensing devices, and simple statistical models are used to analyze pedestrian flow status. Multimedia display tasks are then assigned to fixed devices based on preset rules. Content rendering often uses uniform parameter configurations without dynamically adjusting to device performance and pedestrian flow demands. Furthermore, it relies on a single fault-tolerance mechanism to handle task execution anomalies, lacking comprehensive monitoring and dynamic fault-tolerance strategies for communication links between distributed devices and task execution status. Data transmission between devices often uses generic communication protocols without optimizing latency and synchronization for exhibition hall scenarios, and the overall control process lacks a closed-loop dynamic adjustment mechanism.

[0004] Existing technologies have two significant drawbacks: First, the perception of pedestrian flow and task scheduling are disconnected. A dynamic scheduling mechanism based on the prediction of pedestrian flow patterns has not been established. Simple task allocation is based solely on real-time pedestrian flow data, resulting in a mismatch between equipment resources and pedestrian demand. Some areas have excessive equipment load while others are idle, affecting the overall interactive response efficiency. Second, content adaptation and fault tolerance lack intelligent design. Content rendering parameters are fixed and are not dynamically optimized in conjunction with device hardware performance. At the same time, the fault tolerance strategy is simplistic, only pre-setting processing solutions for specific types of anomalies. This cannot cope with the complex and diverse anomalies during distributed device collaboration, resulting in insufficient consistency in the interactive experience. Summary of the Invention

[0005] To overcome the shortcomings and deficiencies of existing technologies, this invention provides an AI control method for multimedia exhibition halls based on distributed device collaboration and task scheduling.

[0006] The technical solution adopted in this invention is a multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling, comprising the following steps: S1, collecting data on pedestrian movement trajectories, dwell time, and interaction intentions through multimodal sensing terminals deployed in the designated area of ​​the exhibition hall, and constructing a basic dataset by combining the exhibition hall's spatial topology and device distribution information; S2, using an exhibition hall pedestrian situation perception and prediction algorithm to extract features and model the situation of the basic dataset, generating prediction results of pedestrian density, movement paths, and interaction needs in different areas within a preset time period; S3, based on the prediction results, activating a distributed device dynamic task game scheduling algorithm to perform task scheduling on the distributed computing devices, rendering devices, and interactive devices within the exhibition hall. S4. Task allocation and resource scheduling, establishing collaborative communication links between devices; S5. Calling the intelligent adaptation rendering model for multimedia content, dynamically adapting the resolution, frame rate, and interactive response logic of the multimedia display content based on the predicted flow of people and the device scheduling status; S6. Transmitting the adapted multimedia content to the corresponding device for display operation, and using the exhibition hall's parallel task interaction intelligent fault-tolerant processing platform to monitor and handle data transmission anomalies, device response delays, and interactive command conflicts in real time during task execution; S7. Based on the fault-tolerant processing results and real-time collected flow interaction feedback data, dynamically adjusting the parameters of the algorithm and model to form a dynamic optimization mechanism for AI interactive control of the multimedia exhibition hall.

[0007] Furthermore, the exhibition hall pedestrian flow situation perception and prediction algorithm adopts the following expression: , , in, The probability of population transfer from region i to region j These are the weighting coefficients. It is a nonlinear mapping function. The weights of the k-th type of trajectory features are... Let be the feature value of the k-th type of trajectory in region i at time t. Let be the transition coefficient of region j corresponding to the k-th type trajectory at time t. For the region correlation gain function, Let be the real-time pedestrian density in region i at time t. Let j be the spatial capacity characteristic parameter of region j. The weights of the spatial features of the m-th class are... Let m be the spatial feature value of region j. Let be the matching degree between region i and the m-th spatial feature at time t.

[0008] Furthermore, the distributed device dynamic task game scheduling algorithm adopts the following expression: , in, Let x be the payoff value for device x performing task y. For the benefit weighting coefficient, Let x be the fit between device x and task y at time t. Let be the complexity feature parameters of task y. Let be the remaining resources of device x at time t. The weights of the capability characteristics of the z-th type of equipment are: Let x be the z-th type of capability characteristic value of device x at time t. Let be the required capability value of task y for the z-th type of equipment. for The time-lapse update value of device x's fit with task y. The adaptation rate is updated to a coefficient. Let t be the average game payoff value for all device-task combinations at time t. Let t be the task execution priority of device x at time t.

[0009] Furthermore, the intelligent adaptation and rendering model for multimedia content adopts the following expression: ;in, Overall optimization values ​​for rendering adaptation of multimedia content. The weight coefficients for the adaptation dimension of the s-th class. For the content feature coefficients of the s-th class adaptation dimension, The rendering precision parameters for the s-th class adapting dimension. This represents the current energy consumption of the equipment. This represents the maximum allowable energy consumption of the equipment. Let be the interaction response coefficient for the s-th class of adaptation dimension. For the bandwidth adaptation parameters of the s-th adaptation dimension, This represents the rendering quality requirement value corresponding to the current human interaction needs. The maximum rendering quality supported by the device.

[0010] Furthermore, the multimodal sensing terminal includes a lidar sensor, a millimeter-wave radar sensor, a visual image acquisition module, an infrared thermal imaging acquisition module, and an ultrasonic ranging module. Each module aligns its data acquisition timing through a time synchronization protocol, with a sampling frequency set to 10-20Hz. The distributed devices include an edge computing gateway, heterogeneous computing nodes, a GPU rendering server, an FPGA dedicated processing module, and a 5G industrial router. Low-latency communication links are established between devices through a deterministic network protocol, with communication latency controlled within 50ms. The exhibition hall's parallel task interaction intelligent fault-tolerant processing platform has a built-in triple redundancy backup module and a dynamic load balancing unit. It adopts a state machine-based fault-tolerant decision-making mechanism, with a fault-tolerant response time of no more than 30ms. The state transition triggering conditions of the state machine are set through a multi-dimensional threshold matrix, and the threshold matrix parameters are updated in real time according to the operating status of the exhibition hall equipment.

[0011] Further, S2 includes the following sub-steps: S21, performing feature filtering on the raw data collected by the multimodal sensing terminal, selecting features such as pedestrian movement speed, trajectory curvature, density of dwell points, and frequency of interactive actions, and converting different modal data into feature vectors in a unified feature space through a feature fusion algorithm; S22, constructing a pedestrian flow situation time series based on the feature vectors, and using a sliding window mechanism to segment the time series, with the window length dynamically adjusted according to the rate of change of pedestrian flow in the exhibition hall, ranging from 5 to 15 data collection cycles; S23, inputting the segmented time series into the exhibition hall pedestrian flow situation perception and prediction algorithm, and using the algorithm to model the pedestrian flow characteristics of each time period, generating regional-level pedestrian flow density prediction values ​​and path transition probability matrices; S24, verifying the rationality of the prediction results, by comparing historical pedestrian flow data of the same period with current environmental parameters, eliminating abnormal prediction values, and ensuring the adaptability of the prediction results to the actual operating scenario of the exhibition hall.

[0012] Further, S3 includes the following sub-steps: S31, performing capability modeling on all distributed devices in the exhibition hall, constructing a device capability feature matrix, the matrix dimensions including computing power, rendering output capability, communication transmission capability, storage caching capability, and task concurrent processing capability; S32, based on the predicted flow of people and the requirements of multimedia display tasks, decomposing and generating several sub-tasks, determining the resource requirements, execution priority, and time constraints of each sub-task; S33, calling the distributed device dynamic task game scheduling algorithm to match sub-tasks with distributed devices, calculating the game payoff value of each device-sub-task combination through the algorithm, and allocating task execution permissions according to the payoff value ranking result; S34, establishing a collaborative scheduling communication link between devices, transmitting task execution instructions and device status feedback information through the link, synchronizing the task execution progress of each device in real time, and ensuring the consistency of collaborative work of distributed devices.

[0013] Further, step S4 includes the following sub-steps: S41, extracting the attribute parameters of the multimedia content, including content resolution level, frame rate range, color space type, interactive response triggering conditions, and data transmission volume parameters; S42, obtaining the device scheduling results from step S3, and determining the hardware performance parameters of each execution device, including the upper limit of display output resolution, computing processing rate, communication bandwidth capacity, and rendering processing latency; S43, calling the multimedia content intelligent adaptation rendering model, inputting the multimedia content attribute parameters and device hardware performance parameters into the model, and obtaining the optimal combination of adaptation rendering parameters through model calculation; S44, performing real-time processing on the multimedia content according to the optimal combination of adaptation rendering parameters, adjusting the content's resolution, frame rate, and interactive response logic to ensure that the content's display effect and interactive experience on the target device meet the preset standards.

[0014] Further, S5 includes the following sub-steps: S51, through the monitoring nodes deployed on the exhibition hall parallel task interaction intelligent fault-tolerant processing platform, real-time collection of equipment operating parameters, data transmission status, and interactive feedback information during task execution is carried out, with a monitoring node deployment density of no less than 3 per 100 square meters of exhibition hall area; S52, anomaly detection is performed on the collected monitoring data, and by setting multi-dimensional anomaly judgment thresholds, anomalies such as data transmission packet loss, equipment response timeout, and interactive command errors are identified; S53, corresponding fault-tolerant processing strategies are activated for different types of anomalies, including data retransmission mechanism, equipment switching backup mechanism, command correction compensation mechanism, and task reallocation mechanism; S54, the time, type, processing process, and processing result of the anomaly are recorded to form a fault-tolerant processing log, providing data support for subsequent algorithm and model parameter adjustments.

[0015] A multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling is proposed. This method is implemented through a multimedia exhibition hall AI control system based on distributed device collaboration and task scheduling, comprising: a multimodal sensing data acquisition and preprocessing unit, used to collect exhibition hall pedestrian flow-related data through distributed multi-type sensing devices and perform feature filtering and fusion processing; this unit is connected to each sensing terminal in the exhibition hall via an industrial bus interface for real-time data transmission and preliminary processing; a pedestrian flow situation modeling and prediction analysis unit, used to call the exhibition hall pedestrian flow situation perception and prediction algorithm to model and analyze the preprocessed data and generate pedestrian flow situation prediction results; this unit is connected to the multimodal sensing data acquisition and preprocessing unit via a high-speed data interface to receive feature data and output prediction results; and a distributed device task game scheduling unit, used to optimize the matching of devices and tasks and schedule resources based on prediction results and task requirements using a distributed device dynamic task game scheduling algorithm. The system establishes communication connections with the crowd flow situation modeling and prediction analysis unit and distributed equipment within the exhibition hall to transmit scheduling instructions and equipment status information. The multimedia content intelligent adaptation and rendering unit calls upon the multimedia content intelligent adaptation and rendering model to adapt multimedia content based on equipment scheduling status and crowd flow demands. This unit connects with the distributed equipment task game scheduling unit via a data transmission link, receiving equipment parameters and outputting the adapted multimedia content. The parallel task interaction fault-tolerant processing unit monitors and handles anomalies in the task execution process in real time. This unit establishes a bidirectional communication link with distributed equipment and the multimedia content intelligent adaptation and rendering unit, synchronizing equipment operating status and task execution information and executing fault-tolerant strategies. The dynamic optimization and adjustment unit adjusts algorithm and model parameters based on fault-tolerant processing results and interactive feedback data. This unit connects with all functional units, collects operating data from each unit, outputs parameter adjustment instructions, and dynamically optimizes the entire AI interactive control process.

[0016] Beneficial Effects: This invention proposes an AI control method for multimedia exhibition halls based on distributed device collaboration and task scheduling. It collects multi-dimensional pedestrian flow data through multimodal sensing terminals, and uses a pedestrian flow situational awareness and prediction algorithm to accurately predict pedestrian density, paths, and interaction needs. Combined with a distributed device dynamic task game scheduling algorithm, it globally optimizes the allocation of distributed devices such as edge computing gateways and heterogeneous computing nodes within the exhibition hall, solving the disconnect between traditional pedestrian flow perception and task scheduling. This ensures precise matching of device resources with pedestrian flow needs, avoiding excessive load on local devices or idle resources, and significantly improving interaction response efficiency. Furthermore, through an intelligent multimedia content adaptation rendering model, it dynamically adjusts the rendering based on device performance and pedestrian flow needs. The system dynamically adjusts the adaptation parameters of the displayed content, abandoning fixed rendering configurations. Simultaneously, through the triple redundancy backup and dynamic load balancing design of the exhibition hall's parallel task interaction intelligent fault-tolerant processing platform, it constructs a multi-dimensional fault-tolerant strategy to comprehensively address various abnormal situations in distributed collaboration, overcoming the shortcomings of traditional single fault-tolerant strategies. By dynamically optimizing the linkage between functional units, it optimizes the communication synchronization between devices, achieving an overall improvement in distributed device collaboration efficiency and enhanced consistency of interactive experience. This addresses the shortcomings of existing technologies, such as low device collaboration efficiency, inaccurate interactive response, lack of flexibility in content adaptation, and insufficient fault tolerance, providing a smart and efficient AI interactive control solution for multimedia exhibition halls. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of method step S2 of the present invention; Figure 3 This is a flowchart of method step S3 of the present invention; Figure 4 This is a flowchart of method step S4 of the present invention; Figure 5 This is a flowchart of step S5 of the method of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1As shown, the AI ​​control method for multimedia exhibition halls based on distributed device collaboration and task scheduling is characterized by the following steps: S1, collecting data on pedestrian movement trajectories, dwell time, and interaction intentions through multimodal sensing terminals deployed in the designated area of ​​the exhibition hall, and constructing a basic dataset by combining the exhibition hall's spatial topology and device distribution information; S2, using an exhibition hall pedestrian situation perception and prediction algorithm to extract features and model the situation of the basic dataset, generating prediction results of pedestrian density, movement paths, and interaction needs in different areas within a preset time period; S3, based on the prediction results, activating a distributed device dynamic task game scheduling algorithm to allocate tasks among the distributed computing devices, rendering devices, and interactive devices in the exhibition hall. S4. Allocate resources and establish collaborative communication links between devices; S5. Call the intelligent adaptation and rendering model for multimedia content, and dynamically adapt the resolution, frame rate, and interactive response logic of the multimedia display content based on the predicted flow of people and the device scheduling status; S6. Transmit the adapted multimedia content to the corresponding devices for display operations, and use the exhibition hall's parallel task interaction intelligent fault-tolerant processing platform to monitor and handle data transmission anomalies, device response delays, and interactive command conflicts in real time during task execution; S7. Based on the fault-tolerant processing results and real-time collected feedback data from the flow of people, dynamically adjust the parameters of the algorithm and model to form a dynamic optimization mechanism for AI interactive control in the multimedia exhibition hall.

[0020] Step S1 involves constructing a multi-dimensional foundational dataset to provide precise data support for subsequent algorithm calculations and task scheduling. During implementation, multimodal sensing terminals need to be deployed in key areas such as the exhibition hall entrance, main passageways, boundaries between exhibition areas, and around key exhibits. The distance between terminals is strictly controlled between 5 and 8 meters to ensure no blind spots throughout the exhibition hall. The deployed sensing terminals integrate LiDAR sensors, millimeter-wave radar sensors, visual image acquisition modules, infrared thermal imaging acquisition modules, and ultrasonic ranging modules. Each module uses a time synchronization protocol to achieve data acquisition timing alignment, and the sampling frequency is uniformly set to 15Hz to ensure the synchronization and continuity of data acquisition. The collected data includes pedestrian movement trajectory coordinates, dwell time in each area, human body postures, and interactive intent-related behavioral characteristics, among other pedestrian-related data. Simultaneously, exhibition hall spatial topology data is collected, including area boundaries, aisle widths, coordinates of display equipment locations, and maximum space capacity. After collection, a categorized database is established based on pedestrian, spatial, and equipment data. Each data entry is accompanied by a collection timestamp and terminal number to ensure traceability and verification, ultimately forming a complete basic dataset encompassing pedestrian flow, space, and equipment dimensions. This step eliminates the limitations of single-sensor methods through multi-source data fusion, providing comprehensive and accurate raw data input for subsequent pedestrian flow pattern modeling, task scheduling, and content adaptation, ensuring the accuracy and reliability of the entire control process.

[0021] Step S2 involves pedestrian flow situation perception and prediction. Through algorithmic modeling, it achieves accurate prediction of future pedestrian flow dynamics. Implementation requires the foundational dataset built in S1. First, core features are extracted from the collected multimodal data, filtering out key features such as pedestrian movement speed, trajectory curvature, density of stopping points, frequency of interaction actions, and frequency of entry and exit from areas. Data cleaning removes invalid data caused by changes in ambient light or temporary equipment malfunctions. Then, combining the exhibition hall's spatial topology and equipment distribution information, a multi-dimensional pedestrian flow situation analysis model is constructed. The model's input parameters include the filtered core feature data, spatial capacity parameters, equipment service radius, and regional connectivity coefficients. Algorithms perform in-depth data mining and correlation analysis to generate standardized situation feature vectors. The prediction period is set to the next 10 to 30 minutes, and a sliding window mechanism is used to dynamically update the prediction results. The window length is adaptively adjusted according to the pedestrian flow change rate, ranging from 5 to 15 data collection cycles, with each cycle corresponding to 1 second of data at a 15Hz sampling frequency. During the prediction process, the algorithm incorporates historical pedestrian flow data for trend calibration. The calibration weights are dynamically allocated based on the real-time changes in pedestrian flow. When the change exceeds 30%, the weight of real-time data is increased to 70%; when the change is less than 15%, the weight of historical data is increased to 50%. The final output includes predicted pedestrian density for each area, probability distribution of movement paths between areas, and predicted intensity of interaction demand for each area. This step proactively understands pedestrian flow trends, providing a forward-looking decision-making basis for subsequent equipment scheduling and content adaptation, avoiding resource waste and poor user experience caused by reactive responses.

[0022] Step S3 implements dynamic task scheduling for distributed devices, optimizing resource allocation efficiency based on the crowd flow prediction results from S2. During implementation, a comprehensive capability model is first performed on all distributed devices within the exhibition hall, including edge computing gateways, heterogeneous computing nodes, GPU rendering servers, FPGA dedicated processing modules, and 5G industrial routers. Modeling dimensions include computing processing speed, maximum rendering output resolution, communication bandwidth capacity, storage cache space, number of concurrent tasks, and device energy consumption thresholds. Parameters for each dimension are collected in real-time using device self-testing tools at 2-second intervals to ensure real-time performance. Subsequently, based on the intensity of regional interaction demand in the crowd flow prediction results, the overall multimedia display task is broken down into several independent sub-tasks. The resource requirement threshold, execution priority level (1-5, with level 1 being the highest priority), and time constraint range for each sub-task are determined, with the longest execution latency strictly controlled within 50ms. When invoking the distributed device dynamic task game scheduling algorithm, the algorithm calculates the suitability and game payoff value of each device and subtask. The suitability calculation comprehensively considers the device's real-time remaining resources (devices with ≥30% remaining computing resources can participate in task allocation), task complexity coefficient, physical distance between the device and the task execution area (devices ≤20 meters are given priority), and the device's historical execution success rate. After ranking the payoff values, task execution permissions are allocated according to the Top 10 matching results. Simultaneously, devices establish low-latency communication links through deterministic network protocols, with 20% bandwidth redundancy reserved and communication latency controlled within 50ms. A real-time device status feedback mechanism is established at a frequency of 10Hz to ensure accurate transmission of scheduling instructions and synchronization of execution status. This step breaks the traditional fixed task allocation model, achieving dynamic and accurate matching of device resources and user demand, improving resource utilization and task execution efficiency, and laying the foundation for a high-quality interactive experience.

[0023] Step S4 performs intelligent adaptation and rendering of multimedia content. Based on the device scheduling results and user flow demands in S3, content optimization is achieved. During implementation, the core attribute parameters of the multimedia content are first extracted, including the original resolution level (1080P, 4K, 8K), frame rate range (24fps-60fps), color space type, interactive response trigger conditions, and data transmission volume. Parameter extraction is completed using a dedicated content parsing tool, with parsing time controlled within 10ms. Subsequently, the hardware performance parameters of each executing device are acquired, including the upper limit of display output resolution, computing processing speed, communication bandwidth capacity, rendering processing latency, and supported color space formats. Parameter acquisition is performed in real-time through the device communication interface, with the acquisition frequency consistent with the device status feedback frequency at 10Hz. When calling the intelligent adaptation rendering model for multimedia content, the core attribute parameters of the content and the hardware performance parameters of the device are input into the model. The model will adjust the adaptation weights based on the intensity of interaction demand in the predicted pedestrian flow. When the intensity of interaction demand is high (pedestrian density ≥ 3 people / square meter), priority is given to ensuring interaction response speed, and the rendering latency weight is increased to 40%. When the intensity of interaction demand is low (pedestrian density < 1 person / square meter), priority is given to ensuring display effect, and the resolution weight is increased to 45%. The adapted parameters include target resolution (rounded down from the device's upper limit, not lower than 720P), target frame rate (dynamically adjusted from 24fps to 60fps, automatically reduced by 5fps when the device load is ≥ 80%), and interaction response trigger threshold (adjusted according to the frequency of pedestrian actions, reduced by 30% when the frequency is ≥ 2 times / second). The adaptation processing time is controlled within 20ms. This step achieves accurate matching between multimedia content and device performance and pedestrian flow demand, avoiding display stuttering or resource waste caused by incompatible content parameters, and improving the smoothness and adaptability of the interactive experience.

[0024] Step S5 ensures task execution stability through the exhibition hall's parallel task interaction intelligent fault-tolerant processing platform. Based on the multimedia content transmission and display process adapted in S4, comprehensive monitoring and fault tolerance are implemented. During implementation, three monitoring nodes are deployed per 100 square meters within the exhibition hall. These nodes establish bidirectional communication with distributed devices and content rendering units, collecting real-time device operating parameters (CPU load, memory usage, network bandwidth utilization, energy consumption), data transmission status (transmission rate, packet loss rate, latency), and interactive feedback information (command response success rate, action recognition accuracy, content display completeness) during task execution. The collection frequency is 15Hz to ensure no monitoring omissions. The collected monitoring data is analyzed in real-time, and multi-dimensional anomaly judgment thresholds are set. Device CPU load ≥90%, memory usage ≥85%, data transmission packet loss rate ≥3%, command response latency ≥50ms, and action recognition accuracy <95% are all considered abnormal states. The false judgment rate of the anomaly detection algorithm is controlled within 1%. Corresponding fault-tolerance strategies are implemented for different anomaly types: A retransmission mechanism is activated when data transmission packets are lost, with a maximum of 3 retransmissions and a 5ms interval between each retransmission; a device switchover backup mechanism is activated when a device response times out, with a switchover time of ≤20ms; a command correction and compensation mechanism is activated for interaction command conflicts, adjusting command priority based on historical interaction logic; and a task redistribution mechanism is activated when device load is too high, migrating some tasks to neighboring devices with ≥40% remaining resources. The platform has a built-in triple redundancy backup module, and data storage adopts a "master-slave-backup" three-node synchronization mode with a synchronization latency of ≤10ms. The dynamic load balancing unit adjusts device load distribution every 5 seconds to ensure that the load on a single device does not exceed 80%. This comprehensive approach addresses various anomalies during distributed device collaboration, preventing overall interaction interruptions due to local anomalies, ensuring system stability and reliability, and improving user experience consistency.

[0025] Step S6 involves dynamic optimization and adjustment to construct a closed-loop control mechanism. Based on the fault-tolerant processing results of S5 and real-time pedestrian interaction feedback data, system parameters are continuously optimized. During implementation, fault-tolerant processing log data from S5 is first collected, including the time, type, processing method, processing result, and scope of impact of the anomaly. Simultaneously, real-time pedestrian interaction feedback data is collected through multimodal sensing terminals, including interaction success rate, changes in dwell time (compared to historical averages), action feedback delay, and content viewing completion rate. The data collection cycle is 10 seconds, and the storage cycle is 72 hours to ensure sufficient sample size for analysis. Subsequently, the collected data is statistically analyzed, and parameter fit scores for each algorithm and model are calculated. The scoring dimensions include pedestrian prediction accuracy (target value ≥ 85%), equipment scheduling matching rate (target value ≥ 90%), content fit satisfaction (indirectly determined by changes in dwell time; an extension ≥ 10% is considered satisfactory), fault-tolerant processing success rate (target value ≥ 98%), and interaction response smoothness (feedback delay ≤ 30ms is considered satisfactory). The parameters are dynamically adjusted based on the scoring results: When the pedestrian flow prediction accuracy is <85%, the feature weights of the pedestrian flow situation perception prediction algorithm are adjusted, increasing the weights of real-time trajectory features and regional entry / exit frequency features by 5%-10%; when the equipment scheduling matching rate is <90%, the benefit calculation coefficient of the distributed equipment dynamic task game scheduling algorithm is adjusted, increasing the weights of equipment remaining resources and communication latency; when the content adaptation satisfaction is low, the resolution and frame rate adaptation weights of the multimedia content intelligent adaptation rendering model are adjusted; when the fault tolerance processing success rate is <98%, the anomaly judgment threshold and fault tolerance strategy triggering conditions are optimized to shorten the anomaly response time. Parameter adjustments adopt an incremental adjustment mode, with each adjustment not exceeding 10% to avoid excessive system fluctuations. After adjustment, the effect is verified through a small-scale pilot (selecting 2 display areas) for 5 minutes. Once the effect meets the standards, it is fully rolled out. This step constructs a closed-loop dynamic optimization mechanism to continuously improve the adaptability of the algorithm and model, enabling the system to adapt to the dynamic changes in pedestrian flow and equipment status in the exhibition hall, maintain a high-efficiency and stable operating state in the long term, and continuously optimize the AI ​​interactive control effect.

[0026] Preferably, the exhibition hall pedestrian flow situation perception and prediction algorithm adopts the following expression: , , in, The probability of population transfer from region i to region j These are the weighting coefficients. It is a nonlinear mapping function. The weights of the k-th type of trajectory features are... Let be the feature value of the k-th type of trajectory in region i at time t. Let be the transition coefficient of region j corresponding to the k-th type trajectory at time t. For the region correlation gain function, Let be the real-time pedestrian density in region i at time t. Let j be the spatial capacity characteristic parameter of region j. The weights of the spatial features of the m-th class are... Let m be the spatial feature value of region j. Let be the matching degree between region i and the m-th spatial feature at time t.

[0027] Specifically, the exhibition hall pedestrian flow situation perception and prediction algorithm is based on the correlation between pedestrian flow transfer and the influence mechanism of regional characteristics. First, by analyzing the intrinsic correlation between the trajectory characteristics of pedestrian movement and regional spatial attributes, it determines that the probability of pedestrian flow transfer needs to comprehensively consider two core factors: real-time trajectory contribution and regional correlation gain. The basic transfer probability calculation part is constructed first, selecting trajectory feature weights, real-time trajectory feature values, and regional transfer coefficients as key parameters. Multiple types of trajectory information are integrated through linear superposition and nonlinear mapping. The trajectory feature weights range from 0.1 to 0.3, dynamically allocated according to the degree of influence of different trajectory features on transfer behavior. Real-time trajectory feature values ​​are obtained after standardization processing of multimodal perception data, and the regional transfer coefficient is determined based on statistical analysis of historical transfer data, ranging from 0.2 to 0.8. Subsequently, the regional correlation gain function is derived, considering the rate of change of real-time pedestrian density and the matching degree of regional spatial characteristics. The spatial feature weights range from 0.15 to 0.4, and the spatial feature values ​​are obtained by quantifying attributes such as regional capacity and channel width. The matching degree parameter is calculated based on inter-regional connectivity, ranging from 0.3 to 0.9. Calibration was performed using extensive on-site data from the exhibition hall to avoid affecting the accuracy of the results. During implementation, multimodal data was first collected to extract trajectory and regional feature parameters, which were then substituted into formulas to calculate the initial transfer probability. This was then corrected using a gain function, ultimately outputting accurate pedestrian flow transfer prediction results. Through phased derivation and parameter optimization, the accuracy and reliability of pedestrian flow trend prediction were achieved, providing a scientific basis for subsequent scheduling.

[0028] Preferably, the distributed device dynamic task game scheduling algorithm adopts the following expression: , in, Let x be the payoff value for device x performing task y. For the benefit weighting coefficient, Let x be the fit between device x and task y at time t. Let be the complexity feature parameters of task y. Let be the remaining resources of device x at time t. The weights of the capability characteristics of the z-th type of equipment are: Let x be the z-th type of capability characteristic value of device x at time t. Let be the required capability value of task y for the z-th type of equipment. for The time-lapse update value of device x's fit with task y. The adaptation rate is updated to a coefficient. Let t be the average game payoff value for all device-task combinations at time t. Let t be the task execution priority of device x at time t.

[0029] Specifically, the distributed device dynamic task game scheduling algorithm is based on the game equilibrium principle of device capabilities and task requirements. It first determines that the payoff for a device executing a task requires a comprehensive consideration of adaptability, task complexity, and device capability matching, constructing a game payoff calculation model. The adaptability parameter is determined based on the ratio of the device's real-time remaining resources to the task complexity, with a remaining resource threshold set at no less than 30%. The task complexity coefficient is quantified by breaking down the task's computation and communication requirements, ranging from 1 to 5. The device capability feature weight ranges from 0.2 to 0.5, allocated according to the degree of influence of different device capabilities on task execution. Subsequently, the adaptability update formula is derived. Based on the difference between the current game payoff and the average payoff, and combined with the device's execution priority, the adaptability is dynamically adjusted. The update rate coefficient ranges from 0.1 to 0.3, ensuring smooth adaptability adjustment and rapid response to resource changes. In implementation, device capabilities and task requirements are first modeled, various parameters are extracted and substituted into the payoff formula for calculation, tasks are allocated according to payoff ranking, and the matching relationship between devices and tasks is adjusted in real time through the adaptability update formula. Simultaneously, a device status feedback mechanism is established, updating resource parameters every 2 seconds. By using game theory algorithms to achieve optimal allocation of device resources, task execution efficiency and resource utilization are improved, ensuring the stable collaborative operation of distributed devices.

[0030] Preferably, the intelligent adaptation rendering model for multimedia content adopts the following expression: ;in, Overall optimization values ​​for rendering adaptation of multimedia content. The weight coefficients for the adaptation dimension of the s-th class. For the content feature coefficients of the s-th class adaptation dimension, The rendering precision parameters for the s-th class adapting dimension. This represents the current energy consumption of the equipment. This represents the maximum allowable energy consumption of the equipment. Let be the interaction response coefficient for the s-th class of adaptation dimension. For the bandwidth adaptation parameters of the s-th adaptation dimension, This represents the rendering quality requirement value corresponding to the current human interaction needs. The maximum rendering quality supported by the device.

[0031] Specifically, the intelligent adaptation rendering model for multimedia content is based on a dynamic adaptation logic of content features, device performance, and user flow needs. The derivation process revolves around the goal of optimizing rendering quality, integrating three key factors: content rendering accuracy, device energy consumption constraints, and interactive response efficiency. The weights of the adaptation dimensions are determined to range from 0.1 to 0.4, dynamically adjusted according to the intensity of user flow needs; the response coefficient weight is increased when interaction needs are high, and the rendering accuracy weight is increased when needs are low. The content feature coefficient is derived by quantifying attributes such as content resolution and frame rate, ranging from 0.3 to 0.9. The rendering accuracy parameter is set according to the device's supported upper limit, and the ratio of energy consumption to the maximum allowed energy consumption is used to constrain rendering resource allocation and avoid device overload. The interactive response coefficient is calculated by combining device communication bandwidth and processing latency, ranging from 0.2 to 0.7. The bandwidth adaptation parameter is dynamically adjusted based on real-time bandwidth utilization, and the ratio of the required rendering quality value to the maximum rendering quality is used to match user flow interaction expectations. During implementation, content attributes and device performance parameters are first extracted, and adaptation weights are determined based on the needs of pedestrian flow. The optimal combination of rendering parameters is then calculated by substituting them into the model. Parameters such as content resolution and frame rate are adjusted in real time, and the adaptation processing time is controlled within 20ms. This achieves accurate adaptation of multimedia content with devices and pedestrian flow needs, avoids rendering stutters or resource waste, and improves the smoothness and adaptability of the interactive experience.

[0032] Preferably, the multimodal sensing terminal includes a lidar sensor, a millimeter-wave radar sensor, a visual image acquisition module, an infrared thermal imaging acquisition module, and an ultrasonic ranging module. Each module aligns its data acquisition timing through a time synchronization protocol, and the sampling frequency is set to 10-20Hz. The distributed devices include an edge computing gateway, heterogeneous computing nodes, a GPU rendering server, an FPGA dedicated processing module, and a 5G industrial router. The devices establish low-latency communication links through a deterministic network protocol, and the communication latency is controlled within 50ms. The exhibition hall's parallel task interaction intelligent fault-tolerant processing platform has a built-in triple redundancy backup module and a dynamic load balancing unit. It adopts a state machine-based fault-tolerant decision-making mechanism, and the fault-tolerant response time does not exceed 30ms. The state transition triggering conditions of the state machine are set through a multi-dimensional threshold matrix, and the threshold matrix parameters are updated in real time according to the operating status of the exhibition hall equipment.

[0033] Preferred, such as Figure 2As shown, S2 includes the following sub-steps: S21, performing feature filtering on the raw data collected by the multimodal sensing terminal, selecting features such as pedestrian movement speed, trajectory curvature, density of dwell points, and frequency of interactive actions, and converting different modal data into feature vectors in a unified feature space through a feature fusion algorithm; S22, constructing a pedestrian flow situation time series based on the feature vectors, and using a sliding window mechanism to segment the time series, with the window length dynamically adjusted according to the rate of change of pedestrian flow in the exhibition hall, ranging from 5 to 15 data collection cycles; S23, inputting the segmented time series into the exhibition hall pedestrian flow situation perception prediction algorithm, and using the algorithm to model the pedestrian flow characteristics of each time period, generating regional-level pedestrian density prediction values ​​and path transition probability matrices; S24, verifying the rationality of the prediction results, by comparing historical pedestrian flow data with current environmental parameters, eliminating abnormal prediction values, and ensuring the adaptability of the prediction results to the actual operating scenario of the exhibition hall.

[0034] Specifically, step S2 achieves pedestrian flow situation perception and prediction through four sub-steps. S21 first performs feature filtering on the raw data collected by the multimodal sensing terminal, selecting core features such as pedestrian movement speed, trajectory curvature, density of stopping points, and frequency of interaction actions. A feature fusion algorithm converts different modal data into feature vectors in a unified feature space. Feature filtering employs a threshold filtering mechanism to remove outlier data deviating from the mean by more than three standard deviations, ensuring feature effectiveness. S22 constructs a pedestrian flow situation time-series sequence based on the feature vectors. A sliding window mechanism is used to segment the time-series sequence, with the window length dynamically adjusted according to the rate of change of pedestrian flow in the exhibition hall, ranging from 5 to 15 data collection cycles. Each cycle... S23 uses a 1-second data volume at a 15Hz sampling frequency to ensure the continuity and relevance of the time-series data. The segmented time-series sequence is input into the exhibition hall pedestrian flow situation perception and prediction algorithm. The algorithm models the pedestrian flow characteristics of each time period, generating regional-level pedestrian density prediction values ​​and path transition probability matrices. Multi-dimensional feature weighting calculations are used in the modeling process, with weights allocated according to the influence of features on the prediction results, ranging from 0.1 to 0.4. S24 verifies the reasonableness of the prediction results by comparing historical pedestrian flow data with current environmental parameters, eliminating abnormal prediction values. The verification threshold is set at 40% deviation between the predicted value and the historical average, ensuring the adaptability of the prediction results to the actual operating scenario of the exhibition hall. Through precise step-by-step control, the accuracy of pedestrian flow situation prediction is improved, providing a reliable basis for subsequent task scheduling. The macro-level requirements of step S2 are transformed into concrete, implementable operations, ensuring the standardization and effectiveness of each technical implementation step.

[0035] Preferred, such as Figure 3As shown, step S3 includes the following sub-steps: S31, performing capability modeling on all distributed devices in the exhibition hall, constructing a device capability feature matrix, the matrix dimensions including computing power, rendering output capability, communication transmission capability, storage caching capability, and task concurrency processing capability; S32, based on the predicted flow of people and the requirements of multimedia display tasks, decomposing and generating several sub-tasks, determining the resource requirements, execution priority, and time constraints of each sub-task; S33, calling the distributed device dynamic task game scheduling algorithm to match sub-tasks with distributed devices, calculating the game payoff value of each device-sub-task combination through the algorithm, and allocating task execution permissions according to the payoff value ranking result; S34, establishing a collaborative scheduling communication link between devices, transmitting task execution instructions and device status feedback information through the link, synchronizing the task execution progress of each device in real time, and ensuring the consistency of collaborative work of distributed devices.

[0036] Specifically, step S3 implements task scheduling for distributed devices through four sub-steps. S31 involves modeling the capabilities of all distributed devices in the exhibition hall, constructing a device capability feature matrix. The matrix dimensions include computing power, rendering output capability, communication transmission capability, storage caching capability, and task concurrency processing capability. Parameters for each dimension are collected in real-time using a device self-testing tool at 2-second intervals. After quantization, the parameters range from 0 to 100, with higher values ​​indicating stronger capabilities in that dimension. S32, based on the predicted pedestrian flow and multimedia display task requirements, decomposes and generates several sub-tasks, determining the resource requirements, execution priority, and time constraints for each sub-task. Execution priorities are divided into five levels, with level 1 being the highest. The time constraint is set to a maximum execution delay of 50 milliseconds. Resource requirements are quantified separately for computing, communication, and storage. S33 invokes a distributed device dynamic task game scheduling algorithm to match subtasks with distributed devices. The algorithm calculates the game payoff value of each device-subtask combination and allocates task execution permissions based on the payoff value ranking. In the payoff value calculation process, the weight of the device's remaining resources ratio is set to 0.3, the weight of task adaptability is set to 0.4, and the weight of communication latency is set to 0.3. S34 establishes a collaborative scheduling communication link between devices. The link transmits task execution instructions and device status feedback information, and synchronizes the task execution progress of each device in real time. The communication link adopts a deterministic network protocol with a bandwidth reserve of 20% redundancy and a status feedback frequency of 10 Hz. By breaking down the task scheduling process step by step, the precise matching of device capabilities and task requirements is achieved, improving the collaborative efficiency of distributed devices and ensuring the timeliness and stability of task execution.

[0037] Preferred, such as Figure 4As shown, step S4 includes the following sub-steps: S41, extracting the attribute parameters of the multimedia content, including content resolution level, frame rate range, color space type, interactive response triggering conditions, and data transmission volume parameters; S42, obtaining the device scheduling results from step S3, and determining the hardware performance parameters of each execution device, including the upper limit of display output resolution, computing processing rate, communication bandwidth capacity, and rendering processing latency; S43, calling the multimedia content intelligent adaptation rendering model, inputting the multimedia content attribute parameters and device hardware performance parameters into the model, and obtaining the optimal adaptation rendering parameter combination through model calculation; S44, processing the multimedia content in real time according to the optimal adaptation rendering parameter combination, adjusting the content resolution, frame rate, and interactive response logic to ensure that the display effect and interactive experience of the content on the target device meet the preset standards.

[0038] Specifically, step S4 implements intelligent adaptation rendering of multimedia content. Step S31 extracts core attribute parameters of the multimedia content, including key parameters such as content resolution level, frame rate range, color space type, interactive response trigger conditions, and data transmission volume. Parameter extraction is completed using a dedicated content parsing tool, with parsing time controlled within 10 milliseconds. Resolution levels are quantified into three levels, and the frame rate range is set to 24 to 60 frames per second. Step S42 obtains the device scheduling results from step S3, determining the hardware performance parameters of each executing device, including the upper limit of display output resolution, computing processing speed, communication bandwidth capacity, and rendering processing latency. Parameter acquisition is performed in real-time through the device communication interface at a frequency of 10 Hz. The upper limit of display output resolution is divided into three levels according to device hardware specifications, and the computing processing speed is quantified by the amount of data processed per second. Step S43 calls the intelligent adaptation rendering model for multimedia content, rendering the multimedia... The core attribute parameters of the content and the hardware performance parameters of the device are input into the model. The optimal combination of rendering parameters is obtained through model calculation. During the model calculation process, the weight of human interaction needs is dynamically adjusted according to the density of human traffic in the area. When the density is ≥3 people per square meter, the interaction response weight is set to 0.4, and when the density is <1 person per square meter, the rendering quality weight is set to 0.45. S44 processes the multimedia content in real time according to the optimal combination of rendering parameters, adjusting the content resolution, frame rate and interaction response logic. The resolution adjustment is rounded down from the upper limit of the device, with a minimum of 720P. The frame rate is dynamically adjusted according to the device load. When the load is ≥80%, it automatically reduces by 5 frames per second. The adaptation processing time is controlled within 20 milliseconds. By refining the content adaptation process step by step, the precise matching of multimedia content with device performance and human traffic needs is achieved, avoiding display stuttering or resource waste caused by parameter mismatch, and improving the smoothness of the interactive experience.

[0039] Preferred, such as Figure 5As shown, S5 includes the following steps: S51, through the monitoring nodes deployed on the exhibition hall parallel task interaction intelligent fault-tolerant processing platform, real-time collection of equipment operating parameters, data transmission status, and interactive feedback information during task execution is carried out, with a monitoring node deployment density of no less than 3 per 100 square meters of exhibition hall area; S52, anomaly detection is performed on the collected monitoring data, and by setting multi-dimensional anomaly judgment thresholds, anomalies such as data transmission packet loss, equipment response timeout, and interactive command errors are identified; S53, corresponding fault-tolerant processing strategies are activated for different types of anomalies, including data retransmission mechanism, equipment switching backup mechanism, command correction compensation mechanism, and task reallocation mechanism; S54, the time, type, processing process, and processing result of the anomaly are recorded to form a fault-tolerant processing log, providing data support for subsequent algorithm and model parameter adjustments.

[0040] Specifically, step S5 implements fault-tolerant processing for parallel task interaction through four sub-steps. S51 utilizes monitoring nodes deployed on the exhibition hall's intelligent fault-tolerant processing platform for parallel task interaction to collect real-time device operating parameters, data transmission status, and interactive feedback information during task execution. The monitoring node deployment density is no less than 3 per 100 square meters of exhibition hall area. Collected parameters include CPU load, memory usage, transmission rate, packet loss rate, and command response success rate, with a collection frequency of 15 Hz to ensure comprehensive and real-time data collection. S52 performs anomaly detection on the collected monitoring data. By setting multi-dimensional anomaly judgment thresholds, it identifies abnormal situations such as data transmission packet loss, device response timeouts, and interactive command errors. The anomaly judgment thresholds are set as follows: CPU load ≥ 90%, memory usage ≥ 85%, packet loss rate ≥ 3%, and response latency ≥ 50 milliseconds. The algorithm's false positive rate is controlled within 1%. S53 initiates corresponding fault-tolerant handling strategies for different types of anomalies, including data retransmission mechanisms, device switching backup mechanisms, instruction correction and compensation mechanisms, and task reallocation mechanisms. The maximum number of data retransmissions is 3, with a 5-millisecond interval between each retransmission. Device switching time is ≤20 milliseconds. When reallocating tasks, neighboring devices with ≥40% remaining resources are prioritized. S54 records the time, type, processing procedure, and result of the anomaly, forming a fault-tolerant handling log. The log storage period is 72 hours, and each log entry includes 12 key pieces of information, providing data support for subsequent algorithm and model parameter adjustments. By constructing a full-process fault-tolerant system step by step, it comprehensively addresses various anomalies in the distributed device collaboration process, ensuring system stability and consistent user experience, and preventing local anomalies from causing overall interaction interruptions.

[0041] A multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling is proposed. This method is implemented through a multimedia exhibition hall AI control system based on distributed device collaboration and task scheduling, comprising the following functional units: a multimodal sensing data acquisition and preprocessing unit, used to collect exhibition hall pedestrian flow-related data through distributed multi-type sensing devices and perform feature filtering and fusion processing. This unit is connected to each sensing terminal in the exhibition hall via an industrial bus interface for real-time data transmission and preliminary processing; a pedestrian flow situation modeling and prediction analysis unit, used to call the exhibition hall pedestrian flow situation perception and prediction algorithm to model and analyze the preprocessed data, generating pedestrian flow situation prediction results. This unit is connected to the multimodal sensing data acquisition and preprocessing unit via a high-speed data interface to receive feature data and output prediction results; and a distributed device task game scheduling unit, used to optimize the matching of devices and tasks and schedule resources based on prediction results and task requirements through a distributed device dynamic task game scheduling algorithm. The unit establishes communication connections with the pedestrian flow situation modeling and prediction analysis unit and the distributed equipment in the exhibition hall, transmitting scheduling instructions and equipment status information; the multimedia content intelligent adaptation and rendering unit is used to call the multimedia content intelligent adaptation and rendering model, and adapt the multimedia content according to the equipment scheduling status and pedestrian flow demand. This unit is connected to the distributed equipment task game scheduling unit through a data transmission link, receives equipment parameters and outputs the adapted multimedia content; the parallel task interaction fault-tolerant processing unit is used to monitor and handle anomalies in real time during task execution. This unit establishes a two-way communication link with the distributed equipment and the multimedia content intelligent adaptation and rendering unit, synchronizes equipment operating status and task execution information and executes fault-tolerant strategies; the dynamic optimization and adjustment unit is used to adjust the algorithm and model parameters based on the fault-tolerant processing results and interactive feedback data. This unit is connected to all functional units, collects the operating data of each unit and outputs parameter adjustment instructions to dynamically optimize the entire AI interactive control process.

[0042] This AI control method for multimedia exhibition halls, based on distributed device collaboration and task scheduling, constructs a deep collaborative mechanism between pedestrian flow perception and task scheduling, completely overcoming the shortcomings of the separation between the two in existing technologies. Multi-modal perception terminals collect multi-dimensional pedestrian flow data in the exhibition hall, and a specialized pedestrian flow situation perception and prediction algorithm accurately predicts pedestrian distribution, movement paths, and interaction needs. This is then combined with a distributed device dynamic task game scheduling algorithm to globally optimize the allocation of various distributed devices, ensuring precise matching between device resource supply and pedestrian demand. This avoids response delays caused by excessive device load in certain areas and eliminates waste caused by idle device resources, significantly improving the overall system's interactive response efficiency and resource utilization. Simultaneously, a closed-loop control is formed through a dynamic optimization adjustment unit, continuously adjusting the scheduling strategy based on pedestrian flow feedback, further enhancing adaptability.

[0043] This invention utilizes an intelligent multimedia content adaptation rendering model to dynamically adjust various parameters of the displayed content based on the hardware performance of different devices and the real-time demand of visitor flow. It abandons the traditional fixed rendering configuration, ensuring that the content display effect is highly compatible with device capabilities and visitor flow needs. The exhibition hall's parallel task interaction intelligent fault-tolerant processing platform features multiple redundant backups and dynamic load balancing design, constructing a multi-dimensional fault-tolerant strategy covering the entire process of data transmission, device operation, and task execution. This allows for flexible handling of various anomalies that occur during distributed collaboration, replacing traditional single fault-tolerant solutions. It significantly improves system stability and the consistency of the interactive experience, providing more reliable and intelligent AI interactive control support for multimedia exhibition halls.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multimedia exhibition hall AI control method based on distributed device cooperation and task scheduling, characterized in that, Includes the following steps: S1. Collect data on pedestrian movement trajectories, dwell time, and interaction intentions using multimodal sensing terminals deployed in the designated area of ​​the exhibition hall. Combine this data with the exhibition hall's spatial topology and equipment distribution information to construct a basic dataset. S2. Utilize an exhibition hall pedestrian flow situational awareness and prediction algorithm to extract features and model the situation from the basic dataset, generating predictions of pedestrian density, movement paths, and interaction needs in different areas over a preset time period. S3. Based on the prediction results, initiate a distributed device dynamic task game scheduling algorithm to allocate tasks and schedule resources for distributed computing devices, rendering devices, and interactive devices within the exhibition hall, establishing collaborative communication links between devices. S4. The system calls upon a multimedia content intelligent adaptation rendering model to dynamically adapt the resolution, frame rate, and interactive response logic of the multimedia display content based on the predicted flow of people and the device scheduling status. In step S5, the adapted multimedia content is transmitted to the corresponding device for display operations. The exhibition hall's parallel task interaction intelligent fault-tolerant processing platform monitors and handles data transmission anomalies, device response delays, and interactive command conflicts in real time during task execution. Finally, based on the fault-tolerant processing results and real-time collected pedestrian interaction feedback data, the parameters of the algorithm and model are dynamically adjusted, forming a dynamic optimization mechanism for the multimedia exhibition hall's AI interactive control.

2. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, The exhibition hall pedestrian flow situation perception and prediction algorithm uses the following expression: ， ， in, The probability of population transfer from region i to region j These are the weighting coefficients. It is a nonlinear mapping function. The weights of the k-th type of trajectory features are... Let be the feature value of the k-th type of trajectory in region i at time t. Let be the transition coefficient of region j corresponding to the k-th type trajectory at time t. For the region correlation gain function, Let be the real-time pedestrian density in region i at time t. Let j be the spatial capacity characteristic parameter of region j. The weights of the spatial features of the m-th class are... Let m be the spatial feature value of region j. Let be the matching degree between region i and the m-th spatial feature at time t.

3. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, The distributed device dynamic task game scheduling algorithm uses the following expression: ， in, Let x be the payoff value for device x performing task y. For the benefit weighting coefficient, Let x be the fit between device x and task y at time t. Let be the complexity feature parameters of task y. Let be the remaining resources of device x at time t. The weights of the capability characteristics of the z-th type of equipment. Let x be the z-th type of capability characteristic value of device x at time t. Let z be the required capability value of task y for the z-th type of equipment. for The time-lapse update value of device x's fit with task y. The adaptation rate is updated to a coefficient. Let t be the average game payoff value for all device-task combinations at time t. Let t be the task execution priority of device x at time t.

4. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, The intelligent adaptation and rendering model for multimedia content adopts the following expression: ;in, Overall optimization values ​​for rendering adaptation of multimedia content. The weight coefficients for the adaptation dimension of the s-th class. For the content feature coefficients of the s-th class adaptation dimension, The rendering precision parameters for the s-th class adapting dimension. This represents the current energy consumption of the equipment. This represents the maximum allowable energy consumption of the equipment. Let be the interaction response coefficient for the s-th class of adaptation dimension. For the bandwidth adaptation parameters of the s-th adaptation dimension, This represents the rendering quality requirement value corresponding to the current human interaction needs. This is the maximum rendering quality supported by the device.

5. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, The multimodal sensing terminal includes a lidar sensor, a millimeter-wave radar sensor, a visual image acquisition module, an infrared thermal imaging acquisition module, and an ultrasonic ranging module. Each module aligns its data acquisition timing through a time synchronization protocol, with a sampling frequency set to 10-20Hz. The distributed devices include an edge computing gateway, heterogeneous computing nodes, a GPU rendering server, an FPGA dedicated processing module, and a 5G industrial router. The devices establish low-latency communication links through a deterministic network protocol, with communication latency controlled within 50ms. The exhibition hall's parallel task interaction intelligent fault-tolerant processing platform has a built-in triple redundancy backup module and a dynamic load balancing unit. It adopts a state machine-based fault-tolerant decision-making mechanism, with a fault-tolerant response time of no more than 30ms. The state transition triggering conditions of the state machine are set through a multi-dimensional threshold matrix, and the threshold matrix parameters are updated in real time according to the operating status of the exhibition hall equipment.

6. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, S2 includes the following steps: S21, feature filtering is performed on the raw data collected by the multimodal sensing terminal, selecting features such as pedestrian movement speed, trajectory curvature, density of dwell points, and frequency of interactive actions. The feature fusion algorithm is used to convert different modal data into feature vectors in a unified feature space; S22, a pedestrian flow situation time series is constructed based on the feature vectors. A sliding window mechanism is used to segment the time series. The window length is dynamically adjusted according to the rate of change of pedestrian flow in the exhibition hall, with an adjustment range of 5-15 data collection cycles; S23, the segmented time series is input into the exhibition hall pedestrian flow situation perception and prediction algorithm. The algorithm models the pedestrian flow characteristics of each time period and generates regional-level pedestrian density prediction values ​​and path transition probability matrices; S24, the prediction results are validated for reasonableness. By comparing historical pedestrian flow data with current environmental parameters, abnormal prediction values ​​are eliminated to ensure the adaptability of the prediction results to the actual operating scenario of the exhibition hall.

7. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, S3 includes the following steps: S31, performing capability modeling on all distributed devices in the exhibition hall, constructing a device capability feature matrix, the matrix dimensions including computing power, rendering output capability, communication transmission capability, storage caching capability, and task concurrency processing capability; S32, based on the predicted flow of people and the requirements of multimedia display tasks, decomposing and generating several sub-tasks, determining the resource requirements, execution priority, and time constraints of each sub-task; S33, calling the distributed device dynamic task game scheduling algorithm to match sub-tasks with distributed devices, calculating the game payoff value of each device-sub-task combination through the algorithm, and allocating task execution permissions according to the payoff value ranking result; S34, establishing a collaborative scheduling communication link between devices, transmitting task execution instructions and device status feedback information through the link, synchronizing the task execution progress of each device in real time, and ensuring the consistency of collaborative work of distributed devices.

8. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, The S4 includes the following sub-steps: S41, extracting the attribute parameters of the multimedia content, including the content resolution level, frame rate range, color space type, interactive response triggering conditions, and data transmission volume parameters; S42, obtaining the device scheduling results in step S3, and determining the hardware performance parameters of each execution device, including the upper limit of display output resolution, computing processing rate, communication bandwidth capacity, and rendering processing latency. S43 calls the intelligent adaptation rendering model for multimedia content, inputs the multimedia content attribute parameters and device hardware performance parameters into the model, and obtains the optimal combination of adaptation rendering parameters through model calculation. S44 processes multimedia content in real time based on the optimal combination of rendering parameters, adjusting the content's resolution, frame rate, and interactive response logic to ensure that the content's display effect and interactive experience on the target device meet the preset standards.

9. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, S5 includes the following steps: S51, through the monitoring nodes deployed on the exhibition hall parallel task interaction intelligent fault-tolerant processing platform, real-time collection of equipment operating parameters, data transmission status, and interactive feedback information during task execution is carried out, with a deployment density of no less than 3 monitoring nodes per 100 square meters of exhibition hall area; S52, anomaly detection is performed on the collected monitoring data, and by setting multi-dimensional anomaly judgment thresholds, anomalies such as data transmission packet loss, equipment response timeout, and interactive command errors are identified; S53, corresponding fault-tolerant processing strategies are activated for different types of anomalies, including data retransmission mechanism, equipment switching backup mechanism, command correction compensation mechanism, and task reallocation mechanism; S54, the time, type, processing process, and processing result of the anomaly are recorded to form a fault-tolerant processing log, providing data support for subsequent algorithm and model parameter adjustments.

10. The multimedia exhibition hall AI control method based on distributed device collaboration and task scheduling according to claim 1, characterized in that, This method is implemented through a multimedia exhibition hall AI control system based on distributed device collaboration and task scheduling. It includes: a multimodal perception data acquisition and preprocessing unit, used to collect exhibition hall pedestrian flow-related data through distributed multi-type perception devices and perform feature filtering and fusion processing. This unit is connected to each perception terminal in the exhibition hall via an industrial bus interface for real-time data transmission and preliminary processing; a pedestrian flow situation modeling and prediction analysis unit, used to call the exhibition hall pedestrian flow situation perception and prediction algorithm to model and analyze the preprocessed data and generate pedestrian flow situation prediction results. This unit is connected to the multimodal perception data acquisition and preprocessing unit via a high-speed data interface to receive feature data and output prediction results; and a distributed device task game scheduling unit, used to optimize the matching of devices and tasks and schedule resources based on prediction results and task requirements using a distributed device dynamic task game scheduling algorithm. This unit is connected to the pedestrian flow situation modeling and prediction analysis unit. The system establishes communication connections between the units and distributed devices within the exhibition hall, transmitting scheduling instructions and device status information. The multimedia content intelligent adaptation and rendering unit calls upon the multimedia content intelligent adaptation and rendering model to adapt multimedia content based on device scheduling status and visitor flow demands. This unit connects with the distributed device task game scheduling unit via a data transmission link, receiving device parameters and outputting the adapted multimedia content. The parallel task interaction fault-tolerant processing unit monitors and handles anomalies in the task execution process in real time. This unit establishes a bidirectional communication link with the distributed devices and the multimedia content intelligent adaptation and rendering unit, synchronizing device operating status and task execution information and executing fault-tolerant strategies. The dynamic optimization and adjustment unit adjusts algorithm and model parameters based on fault-tolerant processing results and interactive feedback data. This unit connects to all functional units, collects operating data from each unit, outputs parameter adjustment instructions, and dynamically optimizes the entire AI interactive control process.

Images