Multimedia interactive exhibition hall display system based on big data analysis

Abstract

The invention discloses a multimedia interactive exhibition hall display system based on big data analysis, and the system comprises a multi-source sensing layer which is used for collecting the multi-modal behavior data of visitors, and comprises a millimeter wave radar sensor array and a temperature and humidity sensing module; the quantization feature extractor is used for carrying out dimension reduction processing on the original data; a dynamic behavior analysis engine; the augmented reality rendering terminal is in data connection with the behavior analysis engine and generates a dynamic guide identifier in real time; and updating the nodes through federal learning, periodically aggregating the parameter updating quantity of each exhibition hall node, and performing global model updating. According to the method, multi-modal data is collected through a multi-source sensing layer, after quantization feature extraction and dimension reduction, a dynamic behavior analysis engine analyzes a behavior mode in real time, a dynamic guidance identifier is generated in combination with an augmented reality rendering terminal, differential privacy protection noise is injected through a federated learning node, accurate behavior sensing, intelligent interaction and data security are achieved, and the method is high in practicability and high in practicability. And the exhibition hall management efficiency and the user experience are improved.

Description

Technical Field

[0001] The present invention belongs to the field of electronic digital data processing, and more specifically, particularly relates to a multimedia interactive exhibition hall display system based on big data analysis. Background Art

[0002] In recent years, with the development of the Internet of Things and augmented reality technology, multimedia interactive exhibition hall systems have been widely used in museums and science and technology museums. Traditional systems mostly use a one-way display mode that links preset content playback with fixed sensors, which has the following significant defects: Existing technologies mainly rely on RGB cameras for personnel positioning, which has the problems of high light sensitivity and inability to capture micro-movement behaviors; At the same time, the existing technology mostly adopts a centralized data processing mode, resulting in a transmission delay of 200-300ms. When the number of concurrent users is too large, the response error rate rises sharply, which seriously restricts large-scale applications. In addition, the LSTM network with fixed weights is difficult to adapt to sudden changes in the flow of people in the exhibition hall. Field measurement data shows that the traditional model will have inaccurate path planning during peak hours; Based on the above, we propose a multimedia interactive exhibition hall display system based on big data analysis to effectively solve the above problems. Summary of the invention

[0003] The purpose of this invention is to solve the shortcomings of the prior art and to propose a multimedia interactive exhibition hall display system based on big data analysis. Through innovative architecture design and algorithm optimization, this system solves the technical problems of traditional multimedia interactive exhibition hall systems in perception accuracy, dynamic response, privacy protection and energy efficiency, and provides a practical solution for the intelligent upgrade of exhibition hall display systems.

[0004] To achieve the above object, the present invention provides the following technical solutions: A multimedia interactive exhibition hall display system based on big data analysis, comprising: The multi-source perception layer is used to collect multimodal behavior data of visitors, including a millimeter-wave radar sensor array and a temperature and humidity sensor module; The quantized feature extractor is used to reduce the dimension of the original data, and its window adjustment coefficient α satisfies: ; In the formula, N is the number of sampling points in the sliding window, X ti represents the i-th input feature vector at the current time step t; Dynamic behavior analysis engine, the cell state expression of the dynamic behavior analysis engine is: ; Where W co and W cc is the differentiation weight matrix, h t-1 represents the hidden state vector at the previous time step t-1, x t represents the input feature vector at the current time step t, represents the cell state vector at the previous time step t−1; The augmented reality rendering terminal is connected to the behavior analysis engine data to generate dynamic guidance signs in real time; Federated learning updates nodes and periodically aggregates parameter updates of nodes in each exhibition hall to perform global model updates.

[0005] Preferably, the workflow of the quantized feature extractor is as follows: 1) Dynamically adjust the number of wavelet decomposition layers L; Where L = +1; where is the acquisition frequency, is the effective signal bandwidth; 2) Establish an energy threshold criterion and retain the sub-band signals that meet the conditions for reconstruction. The expression is: ; In the formula, is the standard deviation of the subband energy, quantifying the intensity of energy fluctuation in the frequency domain, It is the arithmetic mean of all sub-band energy values ​​after L-layer wavelet decomposition.

[0006] Preferably, the dynamic behavior analysis engine includes: Multimodal mutual information selection module and abnormal behavior detection unit; The workflow of the multimodal mutual information selection module is as follows: 1) The millimeter-wave radar sensor array and temperature and humidity sensor module installed in the multimedia interactive exhibition hall capture the spatial coordinates of visitors in real time, and construct an initial feature pool based on the collected data; 2) Calculating a dynamic feature score for each feature in the initial feature pool, wherein the dynamic feature score calculation includes category relevance calculation, penalty term calculation, and category association strength calculation; 3) Based on the dynamic feature score calculation results, feature screening optimization is performed and the feature subset is dynamically updated.

[0007] Preferably, the abnormal behavior detection unit has the following working process: 1) Initialize basic parameters and extract real-time trajectory parameters, including trajectory smoothing, instantaneous speed calculation, and path deviation calculation; 2) The anomaly judgment threshold is obtained through sliding window statistics and robust standard deviation calculation, and anomaly judgment is made based on the anomaly judgment threshold.

[0008] Preferably, the federated learning update node implements a differential privacy protection mechanism and injects noise according to the following calculation formula: ; In the formula, represents the update amount of the security parameters after noise processing, represents the model update direction of node feature distribution, N represents the non-traceable random noise source, It represents the reference parameter for controlling the noise level, and its value is ∈[0.01,0.1], is a nonlinear scaling term that characterizes the node credibility, and its value is ∈[0,1], In order to balance the key indicator of privacy protection strength and model accuracy, the value is ∈[1,10]; Among them, sensitivity Dynamically adjusted according to the node trust, the expression is: = ; In the formula, Indicates the quality of node historical data, Represents the data quality after adjusting the sensitivity.

[0009] Preferably, the operating parameters of the millimeter wave radar sensor array satisfy: Carrier frequency 76-81GHz; Angular resolution ≤ 0.5°; Spatial coordinate refresh rate ≥ 50Hz; Multi-target tracking capability supports ≥30 concurrent targets.

[0010] Preferably, the instantaneous speed sensitivity factor needs to be considered in the process of determining the abnormality determination threshold, and the dynamic abnormality determination threshold calculation formula is: ; Where, T alert is the dynamic abnormality judgment threshold, μ traj is the weighted moving average of trajectory velocity, σ traj is the velocity fluctuation of the trajectory after removing discrete points, inst is the instantaneous velocity based on the coordinate change, avg is the average moving speed of the visitors to the venue; The sliding window weight allocation calculation formula is: ; is the relative importance attenuation coefficient of historical data in the sliding window, W is the context coverage, and i represents the index value of the instantaneous speed, which is used to identify the instantaneous speed of the current frame.

[0011] Preferably, the anomaly adjudication adopts a multi-level anomaly adjudication, which is specifically divided into a first-level key indicator rapid screening, a second-level model risk scoring, and a third-level space-time continuity check. After the first-level key indicator rapid screening meets the conditions, it enters the second-level model risk scoring. When the second-level model risk scoring meets the conditions, it enters the third-level space-time continuity check. After the third-level space-time continuity check, the anomaly adjudication data is output.

[0012] Preferably, the first-level key indicator rapid screening includes speed overlimit determination and residence timeout determination, and the triggering conditions are: Speed ​​Exceeding Limit Judgment: like inst >3 base ,in, base Indicates the reference speed, the value is 1.2m / s; Immediately trigger a red alarm and activate the AR dynamic warning beam; Resident timeout determination: If Δt>15min, where Δt is the residence time; Trigger the yellow reminder and give a voice prompt; The risk score expression of the secondary model risk score is: ; In the formula, D path is the path deviation, represents the speed fluctuation coefficient; Among them, when Score risk >0.85, enter the third level of space-time continuity check; The judgment conditions for the three-level space-time continuity check are: 1) The instantaneous speed of three consecutive frames exceeds the dynamic threshold; 2) ≥2 abnormal points are detected within a 2m³ space; If the above two groups of conditions are met at the same time, it is determined to be an abnormal event.

[0013] Technical effects and advantages of the present invention: The present invention provides a multimedia interactive exhibition hall display system based on big data analysis. Compared with the prior art, the present invention has the following advantages: Through the coordinated work of the millimeter-wave radar sensor array and the temperature and humidity sensor module, comprehensive collection of multimodal behavior data is achieved, solving the problem of insufficient perception accuracy of traditional single-modal perception systems in complex environments. The multi-source perception layer can not only accurately capture the spatial coordinates and movement trajectories of visitors, but also monitor environmental parameters in real time, providing a richer and more reliable data basis for subsequent behavior analysis; The differential privacy protection mechanism is adopted to dynamically adjust the amount of noise injection, achieving a balance between privacy protection and model accuracy. Compared with the static noise injection scheme, the present invention can adaptively adjust the sensitivity according to the node trust, effectively reducing the risk of data leakage while ensuring the update efficiency of the model; The multi-level abnormality adjudication mechanism significantly improves the detection efficiency and accuracy of abnormal events. From the first-level rapid screening of key indicators to the third-level spatiotemporal continuity check, layer-by-layer verification effectively reduces the false alarm rate and missed alarm rate, and enhances the security and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a structural diagram of a multimedia interactive exhibition hall display system based on big data analysis of the present invention; Figure 2 The figure is a flow chart of calculating the instantaneous speed sensitivity factor of the abnormality determination threshold in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] The present invention provides a multimedia interactive exhibition hall display system based on big data analysis. The system solves the technical problems of traditional multimedia interactive exhibition hall systems in terms of perception accuracy, dynamic response, privacy protection and energy efficiency ratio through innovative architecture design and algorithm optimization, and provides a practical solution for the intelligent upgrade of exhibition hall display systems. like Figure 1 As shown, the system includes: The multi-source perception layer is used to collect multi-modal behavior data of visitors, including a millimeter-wave radar sensor array and a temperature and humidity sensor module; the working parameters of the millimeter-wave radar sensor array meet the following requirements: carrier frequency 76-81GHz; angular resolution ≤ 0.5°; spatial coordinate refresh rate ≥ 50Hz; multi-target tracking capability supports ≥ 30 concurrent targets; edge preprocessing module is used to perform preliminary processing on the raw data collected by the multi-source perception layer. This module improves the quality and consistency of the data through noise filtering, data alignment and format conversion operations, providing a reliable basis for subsequent feature extraction and analysis; The quantized feature extractor is used to reduce the dimension of the original data, and its window adjustment coefficient α satisfies: ; In the formula, N is the number of sampling points in the sliding window, X ti represents the i-th input feature vector at the current time step t; It should be noted that the specific working process of the quantized feature extractor is as follows: 1) Dynamically adjust the number of wavelet decomposition layers L; Where L = +1; where is the acquisition frequency, is the effective signal bandwidth; 2) Establish an energy threshold criterion and retain the sub-band signals that meet the conditions for reconstruction. The expression is: ; In the formula, is the standard deviation of the subband energy, quantifying the intensity of energy fluctuation in the frequency domain, It is the arithmetic mean of all sub-band energy values ​​after L-layer wavelet decomposition.

[0017] Dynamic behavior analysis engine, the cell state expression of the dynamic behavior analysis engine is: ; Where Wco and Wcc are differential weight matrices, ht-1 represents the hidden state vector of the previous time step t-1, xt represents the input feature vector of the current time step t, and represents the cell state vector of the previous time step t−1; The dynamic behavior analysis engine includes: Multimodal mutual information selection module and abnormal behavior detection unit; The workflow of the multimodal mutual information selection module is as follows: The millimeter-wave radar sensor array and temperature and humidity sensor modules installed in the multimedia interactive exhibition hall capture the spatial coordinates of visitors in real time, and build an initial feature pool based on the collected data; Calculating a dynamic feature score for each feature in the initial feature pool, wherein the dynamic feature score calculation includes category relevance calculation, penalty term calculation, and category association strength calculation; Based on the dynamic feature score calculation results, feature screening optimization is performed and the feature subset is dynamically updated.

[0018] In addition, the abnormal behavior detection unit has the following working process: Initialize basic parameters and extract real-time trajectory parameters, including trajectory smoothing, instantaneous speed calculation, and path deviation calculation; The anomaly determination threshold is obtained through sliding window statistics and robust standard deviation calculation, and anomaly judgment is made based on the anomaly determination threshold; As needs to be explained, the instantaneous speed sensitivity factor needs to be considered in the process of determining the abnormality determination threshold, and the dynamic abnormality determination threshold calculation formula is: ; Where, T alert is the dynamic abnormality judgment threshold, μ traj is the weighted moving average of trajectory velocity, σ traj is the velocity fluctuation of the trajectory after removing discrete points, inst is the instantaneous velocity based on the coordinate change, avg is the average moving speed of the visitors to the venue; The sliding window weight allocation calculation formula is: ; is the relative importance attenuation coefficient of historical data in the sliding window, W is the context coverage, and i represents the index value of the instantaneous speed, which is used to identify the instantaneous speed of the current frame.

[0019] The abnormality adjudication adopts multi-level abnormality adjudication, which is specifically divided into primary key indicator rapid screening, secondary model risk scoring, and third-level spatiotemporal continuity inspection. After the primary key indicator rapid screening meets the conditions, it enters the secondary model risk scoring. When the secondary model risk scoring meets the conditions, it enters the third-level spatiotemporal continuity inspection. After the third-level spatiotemporal continuity inspection, the abnormality adjudication data is output, which is specifically: The first-level key indicator rapid screening includes speed overlimit determination and dwell timeout determination, and the triggering conditions are: Speed ​​over limit judgment: If inst >3 base ,in, base Indicates the reference speed, the value is 1.2m / s; Immediately trigger a red alarm and activate the AR dynamic warning beam; Dwell timeout judgment: if Δt>15min, where Δt is the dwell time; Trigger the yellow reminder and give a voice prompt; The risk score expression of the secondary model risk score is: ; In the formula, D path is the path deviation, Represents the speed fluctuation coefficient; when Score risk >0.85, enter the third level of space-time continuity check; The judgment conditions for the three-level spatiotemporal continuity check are: 1) the instantaneous speed of 3 consecutive frames exceeds the dynamic threshold; 2) ≥ 2 abnormal points are detected within a 2m³ space range; if the above two sets of conditions are met at the same time, it is judged as an abnormal event.

[0020] The augmented reality rendering terminal is connected to the behavior analysis engine data to generate dynamic guidance signs in real time; specifically: Wearable AR components: micro-OLED-based holographic lens glasses, supporting 120° field of view and gesture tracking; environmental projection module: DLP laser projection array deployed on the top of the exhibition hall to realize dynamic light spot projection on the ground; edge computing unit: FPGA chip equipped with parallel rendering engine, directly connected to the behavior analysis engine through the PCIe bus; multimodal interaction interface: integrated bone conduction audio module and infrared gesture capture sensor.

[0021] Real-time data stream mapping: Receive the structured data stream output by the behavior analysis engine, including: real-time interest vectors of visitors (based on time series encoding generated by duration of stay and focus of sight); dynamic path planning suggestions (including weight matrices for avoiding congested areas); abnormal event level labels (light effect parameters used to trigger warning signs); build a three-dimensional space data pool, and map logical data to physical exhibit locations through coordinate system transformation.

[0022] Scene adaptive matching engine: Establishing the knowledge graph of exhibits, the node attributes include: theme classification weight (science / art / history); interaction complexity classification; content update timeliness mark; cosine similarity matching based on the current visitor's interest vector and the exhibit feature vector, and outputting the priority queue.

[0023] Exemplarily, the above terminal implements intelligent guidance through the following submodules working together: The above terminal interacts with the federated learning node: Periodically upload anonymized interaction data (click hot spots, content dwell time); Receive updated personalized recommended model parameters (automatically switch rendering styles in night mode); Adaptive environment tuning: Light compensation algorithm: Combine temperature and humidity sensor data to dynamically optimize virtual light intensity; Multi-terminal collaborative rendering: unify the AR perspectives of multiple visitors through a distributed consensus protocol.

[0024] Federated learning update nodes periodically aggregate parameter updates of nodes in each exhibition hall to update the global model; Furthermore, the federated learning update node implements a differential privacy protection mechanism and injects noise according to the following calculation formula: ; In the formula, represents the update amount of the security parameters after noise processing, represents the model update direction of node feature distribution, N represents the non-traceable random noise source, It represents the reference parameter for controlling the noise level, and its value is ∈[0.01,0.1], is a nonlinear scaling term that characterizes the node credibility, and its value is ∈[0,1], In order to balance the key indicator of privacy protection strength and model accuracy, the value is ∈[1,10]; Among them, Figure 2 , sensitivity Dynamically adjusted according to the node trust, the expression is: = ; In the formula, Indicates the quality of node historical data, It is expressed as the data quality after adjusting the sensitivity; As mentioned above, the privacy-utility balance is improved: through Adaptive adjustment of , under the same ϵ constraint: When the node credibility ↑, ↑ →Noise amount↓→Model accuracy↑; When the node credibility is ↓, ↓ →Noise level↑→Privacy protection↑.

[0025] As an option, the exhibit autonomous evolution module is deployed in the federated learning update node to dynamically optimize the content and presentation of the exhibits. This module generates improvement suggestions for the exhibits by analyzing the visitor's interaction data and feedback information, combined with the content update timeliness mark, and pushes them to the augmented reality rendering terminal to achieve autonomous optimization and update of the exhibits.

[0026] In summary, the technical solution of the present invention has the following significant advantages over the prior art: Advantages of the multi-source perception layer: Through the coordinated work of the millimeter-wave radar sensor array and the temperature and humidity sensor module, comprehensive multi-modal behavioral data collection is achieved, solving the problem of insufficient perception accuracy of traditional single-modal perception systems in complex environments. The multi-source perception layer can not only accurately capture the spatial coordinates and movement trajectories of visitors, but also monitor environmental parameters in real time, providing a richer and more reliable data foundation for subsequent behavioral analysis.

[0027] Advantages of the quantized feature extractor: It uses the improved Haar wavelet packet transform to reduce the dimensionality of the original data, and by dynamically adjusting the window coefficient, it effectively reduces data redundancy and improves the accuracy and efficiency of feature extraction. Compared with the traditional fixed window dimensionality reduction method, the present invention can better adapt to the data characteristics in different scenarios and ensure the efficient retention of key information.

[0028] Advantages of the dynamic behavior analysis engine: An improved long-tail LSTM network was designed, and a composite weight matrix was introduced to solve the problem of inaccurate prediction of the traditional LSTM network in sudden changes in the flow of people. The engine can analyze visitor behavior in real time and dynamically adjust model parameters, significantly improving the accuracy of behavior prediction and the adaptability of the system.

[0029] The advantages of augmented reality rendering terminal: seamless connection with dynamic behavior analysis engine, real-time generation of dynamic guide signs, solving the problem of low matching between traditional system guide prompts and actual visitor needs. Through AR technology, a more intuitive and personalized tour experience is achieved, which improves the interaction and satisfaction of visitors.

[0030] Advantages of updating nodes in federated learning: It adopts differential privacy protection mechanism and dynamically adjusts the amount of noise injection to achieve a balance between privacy protection and model accuracy. Compared with the static noise injection scheme, this invention can adaptively adjust the sensitivity according to the node trust, effectively reducing the risk of data leakage while ensuring the update efficiency of the model.

[0031] Advantages of the abnormal behavior detection unit: Through the multi-level abnormal adjudication mechanism, the detection efficiency and accuracy of abnormal events are significantly improved. From the first-level key indicator rapid screening to the third-level spatiotemporal continuity check, layer-by-layer verification effectively reduces the false alarm rate and missed alarm rate, and enhances the security and reliability of the system.

[0032] Advantages of dynamic feature scoring calculation: Through the comprehensive calculation of category relevance, redundancy penalty terms and category association strength, intelligent feature screening is achieved, solving the problem that traditional static feature selection methods cannot adapt to dynamic scenarios. The dynamic update mechanism of feature subsets further ensures the efficient operation of the system in different scenarios.

[0033] Advantages of comprehensive performance: The overall solution achieves full-process optimization of perception, analysis, rendering and updating through the collaborative work of multiple modules. The system can efficiently process large-scale concurrent data, significantly improving the intelligence level and operational efficiency of the exhibition hall display system, and providing visitors with a better service experience.

[0034] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multimedia interactive exhibition hall display system based on big data analysis, characterized in that: include: The multi-source perception layer is used to collect multimodal behavior data of visitors, including a millimeter-wave radar sensor array and a temperature and humidity sensor module; The quantized feature extractor is used to reduce the dimension of the original data, and its window adjustment coefficient α satisfies: ； Where N is the number of sampling points in the sliding window; X ti represents the i-th input feature vector at the current time step t; Dynamic behavior analysis engine, the cell state expression of the dynamic behavior analysis engine is: ； Where W co and W cc is the differentiation weight matrix, h t-1 represents the hidden state vector at the previous time step t-1, x t represents the input feature vector at the current time step t, represents the cell state vector at the previous time step t−1; The augmented reality rendering terminal is connected to the behavior analysis engine data to generate dynamic guidance signs in real time; Federated learning updates nodes and periodically aggregates parameter updates of nodes in each exhibition hall to perform global model updates.

2. A multimedia interactive exhibition hall display system based on big data analysis according to claim 1, characterized in that: The specific workflow of the quantized feature extractor is as follows: 1) Dynamically adjust the number of wavelet decomposition layers L; Where L = +1; where is the acquisition frequency, is the effective signal bandwidth; 2) Establish an energy threshold criterion and retain the sub-band signals that meet the conditions for reconstruction. The expression is: ; In the formula, is the standard deviation of the subband energy, quantifying the intensity of energy fluctuation in the frequency domain, It is the arithmetic mean of all sub-band energy values ​​after L-layer wavelet decomposition.

3. A multimedia interactive exhibition hall display system based on big data analysis according to claim 1, characterized in that: The dynamic behavior analysis engine includes: Multimodal mutual information selection module and abnormal behavior detection unit; The workflow of the multimodal mutual information selection module is as follows: 1) The millimeter-wave radar sensor array and temperature and humidity sensor module installed in the multimedia interactive exhibition hall capture the spatial coordinates of visitors in real time, and construct an initial feature pool based on the collected data; 2) Calculating a dynamic feature score for each feature in the initial feature pool, wherein the dynamic feature score calculation includes category relevance calculation, penalty term calculation, and category association strength calculation; 3) Based on the dynamic feature score calculation results, feature screening optimization is performed and the feature subset is dynamically updated.

4. A multimedia interactive exhibition hall display system based on big data analysis according to claim 3, characterized in that: The abnormal behavior detection unit has the following working process: 1) Initialize basic parameters and extract real-time trajectory parameters, including trajectory smoothing, instantaneous speed calculation, and path deviation calculation; 2) The anomaly judgment threshold is obtained through sliding window statistics and robust standard deviation calculation, and anomaly judgment is made based on the anomaly judgment threshold.

5. The multimedia interactive exhibition hall display system based on big data analysis according to claim 1 is characterized in that: The federated learning update node implements a differential privacy protection mechanism and injects noise according to the following calculation formula: ； In the formula, represents the update amount of the security parameters after noise processing, represents the model update direction of node feature distribution, N represents the non-traceable random noise source, It represents the reference parameter for controlling the noise level, and its value is ∈[0.01,0.1], is a nonlinear scaling term that characterizes the node credibility, and its value is ∈[0,1], In order to balance the key indicator of privacy protection strength and model accuracy, the value is ∈[1,10]; Among them, sensitivity Dynamically adjusted according to the node trust, the expression is: = ； In the formula, Indicates the quality of node historical data, Represents the data quality after adjusting the sensitivity.

6. The multimedia interactive exhibition hall display system based on big data analysis according to claim 1 is characterized in that: The operating parameters of the millimeter wave radar sensor array meet the following requirements: Carrier frequency 76-81GHz; Angular resolution ≤ 0.5°; Spatial coordinate refresh rate ≥ 50Hz; Multi-target tracking capability supports ≥30 concurrent targets.

7. The multimedia interactive exhibition hall display system based on big data analysis according to claim 4 is characterized in that: The instantaneous speed sensitivity factor needs to be considered in the process of determining the abnormality determination threshold. The dynamic abnormality determination threshold calculation formula is: ； Where, T alert is the dynamic abnormality judgment threshold, μ traj is the weighted moving average of trajectory velocity, σ traj is the velocity fluctuation of the trajectory after removing discrete points, inst is the instantaneous velocity based on the coordinate change, avg is the average moving speed of the visitors to the venue; The sliding window weight allocation calculation formula is: ； is the relative importance attenuation coefficient of historical data in the sliding window, W is the context coverage, and i represents the index value of the instantaneous speed, which is used to identify the instantaneous speed of the current frame.

8. The multimedia interactive exhibition hall display system based on big data analysis according to claim 7 is characterized in that: The anomaly adjudication adopts a multi-level anomaly adjudication, which is specifically divided into a first-level key indicator rapid screening, a second-level model risk scoring, and a third-level space-time continuity check. After the first-level key indicator rapid screening meets the conditions, it enters the second-level model risk scoring. When the second-level model risk scoring meets the conditions, it enters the third-level space-time continuity check. After the third-level space-time continuity check, the anomaly adjudication data is output.

9. The multimedia interactive exhibition hall display system based on big data analysis according to claim 8 is characterized in that: The first-level key indicator rapid screening includes speed overlimit determination and dwell timeout determination, and the triggering conditions are: Speed ​​Exceeding Limit Judgment: like inst >3 base ,in, base Indicates the reference speed, the value is 1.2m / s; Immediately trigger a red alarm and activate the AR dynamic warning beam; Resident timeout determination: If Δt>15min, where Δt is the residence time; Trigger the yellow reminder and give a voice prompt; The risk score expression of the secondary model risk score is: ； In the formula, D path is the path deviation, represents the speed fluctuation coefficient; Among them, when Score risk >0.85, enter the third level of space-time continuity check; The judgment conditions for the three-level space-time continuity check are: 1) The instantaneous speed of three consecutive frames exceeds the dynamic threshold; 2) ≥2 abnormal points are detected within a 2m³ space; If the above two groups of conditions are met at the same time, it is determined to be an abnormal event.

Images