AI exhibition hall control method and system based on crowd thermal distribution monitoring

Abstract

The invention discloses an AI exhibition hall control method and system based on crowd thermal distribution monitoring, and the method comprises the steps: obtaining crowd density distribution data, behavior data and environment data through constructing a full-coverage sensing network, employing a thermal imaging camera, a WiFi probe and an infrared sensor, and generating a thermal distribution diagram; through converting discrete density point fusion and labeling into a data set of a continuous thermal field, crowd distribution characteristics are visually displayed, spatial-temporal correlation of crowd flow is effectively modeled by combining a convolutional neural network and a spatial-temporal diagram neural network, a prediction-adjustment-verification closed loop is realized based on a prediction result 6-8 minutes ahead of time, and the prediction efficiency is improved. The control strategy is dynamically adjusted according to the predicted crowd distribution trend and the business target, the guide route or the interaction form can be dynamically adjusted according to the crowd density, the exhibit layout can be adjusted in advance, the number of tourists going back and forth is reduced, the overall flow efficiency is improved, and therefore the display effect of the AI exhibition hall is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of AI exhibition hall control, and particularly relates to an AI exhibition hall control method and system based on crowd heat distribution monitoring. Background Art

[0002] The operation of an AI exhibition hall involves multiple links such as equipment control, security monitoring, and energy management. AI can achieve intelligent linkage of equipment, automatically generate energy consumption reports, and even reduce equipment failures through predictive maintenance, thereby reducing labor costs.

[0003] Currently, the main method for controlling an AI exhibition hall is based on AI data analysis and decision-making. Cameras and sensors are used to collect audience behavior data. AI can analyze the interests of the audience and optimize the exhibition layout or explanation content. For example, if most audiences stay in front of a certain exhibit for a short time, the system can automatically adjust the lighting or interaction form of that exhibition area.

[0004] However, when performing AI data analysis and decision-making, the AI analysis model mainly relies on training data. If the exhibition hall environment is very different from the training scenario, it may lead to misjudgment, resulting in poor real-time performance of adjusting the guided tour route or interaction form in the exhibition hall, thus affecting the display effect of the AI exhibition hall. For example, in extreme crowd flow situations, sensors may fail due to data overload, making it inconvenient to accurately obtain the distribution of the audience in the exhibition hall and timely adjust the guided tour route or interaction form. Therefore, we need to propose an AI exhibition hall control method and system based on crowd heat distribution monitoring to solve the above problems, enabling it to dynamically adjust the guided tour route or interaction form according to the crowd density and improve the display effect of the AI exhibition hall. Summary of the Invention

[0005] The purpose of the present invention is to provide an AI exhibition hall control method and system based on crowd heat distribution monitoring, which can dynamically adjust the guided tour route or interaction form according to the crowd density and improve the display effect of the AI exhibition hall, so as to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: An AI exhibition hall control method based on crowd heat distribution monitoring includes the following steps: S1. Install multi-source sensor devices on the ceiling, wall or floor of the exhibition hall to construct a full-coverage perception network. The multi-source sensor devices include thermal imaging cameras, infrared sensors, WiFi probes, RFID tag readers, and environmental sensors; S2. Real-time capture crowd density distribution data, behavior data, and environmental data through the multi-source sensor devices, and generate a heat distribution map; S3. Filter the real-time data obtained by the multi-source sensor device, then fuse the thermal distribution data, behavior data, and environmental data, and label the thermal characteristics of each area to form a dataset of a continuous thermal field; S4. Establish a real-time thermal prediction model by combining a convolutional neural network and a spatio-temporal graph neural network, and predict the crowd distribution trend within 6 - 8 minutes based on the thermal characteristics; S5. Dynamically adjust the control strategy according to the predicted crowd distribution trend and business objectives; S6. Real-time monitor the change in thermal distribution after dynamic adjustment, and optimize the parameters of the thermal prediction model according to the change in thermal distribution.

[0007] Preferably, the resolution of the thermal imaging camera is [X] pixels, and the deployment spacing ≤ 5 meters; the WiFi probe supports the Bluetooth 5.0 + WiFi6 protocol, and the positioning error ≤ 1.5 meters; the infrared sensor coverage angle ≥ 120°, and the detection distance ≤ 8 meters.

[0008] Preferably, the process for generating the thermal distribution map is as follows: A1. Divide the field of view of the thermal camera into multiple pixel blocks, calculate the number of human targets in each pixel block, and calculate the density value in combination with the actual area corresponding to each pixel block; A2. Smooth the discrete density points to generate a continuous density field; A3. Map the density value into a thermal level of 0 - 100; A4. Use a cold-warm gradient method for color coding for high-density and low-density distributions, and output the thermal distribution map after GIS rendering.

[0009] Preferably, the process for establishing the dataset of the continuous thermal field is as follows: B1. Identify and remove data that deviates from the normal range through an outlier detection algorithm; B2. Fill in the missing data using linear interpolation and filter the data using a low-pass filter; B3. Fuse the thermal distribution data, behavior data, and environmental data according to the timestamp to obtain a visual record containing multi-dimensional information; B4. Calculate the crowd density based on the number of people in the area and the area of the area, and use it as a basic indicator of the thermal characteristics to add labels to each area; B5. Normalize the numerical data, and then organize the normalized data into a structured table or database table structure; B6. Add metadata of the dataset to the structured table or database table structure to form a standardized dataset.

[0010] Preferably, in step B3, the fusion process is as follows: B31. Synchronize the heat distribution data, behavior data, and environmental data according to the time stamp to make different types of data consistent in the time dimension; B32. Map the heat distribution data, behavior data, and environmental data to a unified spatial coordinate system to make the data comparable in space; B33. Map the features of different types of data to the color and intensity values in the heat distribution map, unify the data scale through normalization processing, and display the data distribution density with the depth of color on a two-dimensional plane to form a visual record containing multi-dimensional information.

[0011] Preferably, in step S4, the process of establishing the real-time heat prediction model is as follows: C1. Divide the standardized data according to time windows and spatial grids to form an input tensor; C2. Process the input tensor through a multi-layer convolutional neural network; C3. Calculate the spatial attention weight according to the processing result of the input tensor; C4. Combine the convolution result with the time series data and use the gated recurrent algorithm to learn the dependence of crowd flow in the time dimension; C5. Calculate the time attention weight according to the dependence of crowd flow in the time dimension; C6. Establish a prediction model by combining the time attention weight, spatial attention weight, spatial features, and time features; C7. Train the prediction model with the data of the standardized data set and optimize the parameters of the prediction model using the loss function. The optimized prediction model is the real-time heat prediction model.

[0012] Preferably, the process of predicting the crowd distribution trend is as follows: D1. Real-time collect multi-source data in the 6 - 8 minutes before the current moment and normalize the collected data; D2. Predict the normalized data through the heat prediction model to obtain the prediction result; D3. Restore the prediction result to the actual heat level; D4. Convert the predicted heat level to a color heat map and use different colors to represent the crowd density level.

[0013] Preferably, the control content of the control strategy includes control based on exhibit attention, control based on visiting fluency, and control of interactive experience; among them, the control based on exhibit attention includes optimizing the flow of popular exhibits and activating cold areas; The control method based on visiting fluency is: when it is predicted that the density value in the passage area ≥ 60 people / m 2When the flow velocity < 0.5 m / s, congestion warning and diversion control are carried out; The control method based on interactive experience is as follows: If it is predicted that the number of people queuing in the interactive device area ≥ 5 and the waiting time > 10 minutes, dynamic scheduling operation of interactive devices is carried out.

[0014] Preferably, the process of parameter optimization of the thermal prediction model is as follows: E1. Collect actual thermal data through a thermal imaging camera every two minutes, compare it with the predicted thermal map, and calculate the error value of each grid; E2. Add the latest actual data to the standardized data set, and update the parameters of the convolutional neural network and the spatio-temporal graph neural network by using incremental learning; E3. If the spatial attention weight of a certain area is continuously low but the actual crowd is dense, increase the weight of the features of this area in the spatial attention calculation; If the prediction error in a certain period of the time series is large, enhance the time attention weight of this period; E4. Adopt the sliding window method, use the data of the past 1 hour to verify the optimized model. If the prediction error is reduced to less than 5%, save the new model; otherwise, repeat E1 - E5.

[0015] Based on the AI exhibition hall control method based on crowd thermal distribution monitoring described above, the present invention also provides an AI exhibition hall control system based on crowd thermal distribution monitoring, including: a data acquisition module, which obtains the original data stream in the exhibition hall in real time through multi-source sensor devices, and the original data stream includes crowd thermal coordinates, stay duration, movement speed, and environmental parameters; A thermal map generation module, which converts the perception data into a visual thermal distribution for real-time monitoring and analysis; A data processing and integration module, which cleans, complements, and integrates multi-source data to form a standardized data set of a continuous thermal field; A real-time prediction module, which establishes a real-time thermal prediction module by combining a convolutional neural network and a spatio-temporal graph neural network to predict the crowd distribution trend 6 - 8 minutes in advance; An execution control module, which dynamically adjusts the exhibition hall resources according to the prediction results and business goals. The exhibition hall resources include optimizing the attention of exhibits, optimizing the fluency, and optimizing the interactive experience; A feedback optimization module, which is used to construct a closed loop of monitoring and optimization, real-time monitor the change of the thermal distribution after dynamic adjustment, and optimize the parameters of the thermal prediction model according to the change of the thermal distribution.

[0016] The AI exhibition hall control method and system based on crowd thermal distribution monitoring proposed by the present invention have the following advantages compared with the prior art: 1. The present invention obtains crowd density distribution data, behavior data, and environmental data through a thermal imaging camera, a WiFi probe, and an infrared sensor, and generates a heat map; by fusing and annotating discrete density points into a dataset of a continuous heat field, it intuitively displays the crowd distribution characteristics, and effectively models the spatio-temporal correlation of crowd flow by combining a convolutional neural network and a spatio-temporal graph neural network. Based on the prediction results 6 - 8 minutes in advance, a closed-loop of prediction - adjustment - verification is achieved. By dynamically adjusting the control strategy according to the predicted crowd distribution trend and business objectives, it can dynamically adjust the guided tour route or interaction form according to the crowd flow density, and can also adjust the exhibit layout in advance, reduce the number of round trips of tourists, improve the overall flow efficiency, and thus enhance the display effect of the AI exhibition hall.

[0017] 2. The present invention realizes the full-chain intelligence from data collection to business implementation through the collaboration of a data collection module, a heat map generation module, a data processing and integration module, a real-time prediction module, an execution control module, and a feedback optimization module. The modules are linked by a data stream, combining deep learning and dynamic control, which not only meets the real-time monitoring requirements of the exhibition hall, but also dynamically adjusts the guided tour route or interaction form according to the crowd flow density, enhancing the display effect of the AI exhibition hall. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Shows a flowchart of the AI exhibition hall control method according to an embodiment of the present invention; Figure 2 Shows a flowchart of the heat map generation process according to an embodiment of the present invention; Figure 3 Shows a flowchart of the establishment of a dataset of a continuous heat field according to an embodiment of the present invention; Figure 4 Shows a flowchart of the establishment of a real-time heat prediction model according to an embodiment of the present invention; Figure 5 Shows a system block diagram of the AI exhibition hall control system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0020] The present invention provides an AI exhibition hall control method based on crowd heat distribution monitoring as Figures 1-4 shown, including the following steps: S1. Install multi-source sensor devices on the ceiling, wall or floor of the exhibition hall to build a full-coverage perception network. The multi-source sensor devices include thermal imaging cameras, infrared sensors, WiFi probes, RFID tag readers and environmental sensors; Among them, the resolution of the thermal imaging camera pixels, the deployment spacing ≤ 5 meters, ensuring the accuracy of the heat map grid ; The WiFi probe supports the Bluetooth 5.0 + WiFi6 protocol, and the positioning error ≤ 1.5 meters; The infrared sensor has a coverage angle ≥ 120°, and the detection distance ≤ 8 meters; The environmental sensor monitors the air quality in real time and triggers a ventilation warning when the threshold exceeds 800 ppm; S2. Real-time capture crowd density distribution data, behavior data and environmental data through multi-source sensor devices, and generate a heat map according to the crowd density distribution data; The process of generating the heat map is as follows: A1. Divide the field of view of the thermal camera into multiple pixel blocks, calculate the number of human targets in each pixel block, and calculate the density value in combination with the actual area corresponding to each pixel block. The density value calculation formula is: , where is the density value of each pixel block, is the number of human targets in each pixel block, is the row index, is the column index, is the actual area corresponding to a single pixel block; A2. Smooth the discrete density points to generate a continuous density field. The smoothing formula is: , where is the density estimate value at the coordinate , is the total number of sample points, is the bandwidth parameter used to control the smoothing degree, is the th sample point coordinate, is the Gaussian kernel function; A3. Map the density value to the heat level of 0-100. The mapping formula is: , where is the heat level value at the coordinate , is the crowd density value at the coordinate , and are the global maximum crowd density value and the global minimum crowd density value respectively; A4. Use a cold-warm gradient method for color coding for high-density and low-density distributions, and output a heat map after GIS rendering; GIS data is managed by layers (such as base maps, hotspot layers, dynamic layers), and each layer corresponds to a specific type of geographical feature. When rendering, it is drawn layer by layer by calling a display list (such as the display list of OpenGL), and only the changed parts are updated to improve efficiency.

[0021] S3. Filter the real-time data obtained by multi-source sensor devices, then fuse the heat distribution data, behavior data, and environmental data, and label the heat characteristics of each area to form a dataset of a continuous heat field; The process of establishing the dataset of the continuous heat field is as follows: B1. Identify and remove data that deviates from the normal range through an outlier detection algorithm; B2. Fill in the missing data using linear interpolation to ensure the integrity and continuity of the data, and use a low-pass filter to filter the data to remove high-frequency noise in the data; B3. Fuse the heat distribution data, behavior data, and environmental data according to the time stamp to obtain a visual record containing multi-dimensional information; The fusion process is as follows: B31. Synchronize the heat distribution data, behavior data, and environmental data according to the time stamp to make different types of data consistent in the time dimension; B32. Map the heat distribution data, behavior data, and environmental data to a unified spatial coordinate system to make the data comparable in space; B33. Map the characteristics of different types of data to the color and intensity values in the heat map, unify the data scale through normalization processing, and display the data distribution density in terms of color depth on a two-dimensional plane to form a visual record containing multi-dimensional information; B4. Calculate the population density based on the number of people in the area and the area of the area as a basic indicator of the heat characteristics, add labels to each area, and the added labels include but are not limited to heat levels, area types, and time stamp information; B5. Perform normalization processing on the numerical data, and then organize the normalized data into a structured table or database table structure; B6. Add meta-information of the dataset to the structured table or database table structure to form a standardized dataset. The meta-information includes data sources, collection times, sensor types, etc., to enhance the understandability and traceability of the data; S4. Establish a real-time heat prediction model by combining a convolutional neural network and a spatio-temporal graph neural network, and predict the population distribution trend within 6-8 minutes based on the heat characteristics; The process of establishing the real-time heat prediction model is as follows: C1. Divide the standardized data by time window and spatial grid to form the input tensor X. The formula expression of the input tensor is: , where X is the value of the input tensor, is the size of the spatial grid, T is the time step of the time window, C is the number of feature channels, is the set of real numbers; C2. Process the input tensor X through a multi-layer convolutional neural network. The formula expression of the multi-layer convolutional neural network is: , where, is the -th layer of spatial feature map, is the convolutional kernel weight of the -th layer, is the activation function, is the convolutional operation, is the -th layer of bias term, is the -th layer of spatial feature map; C3. Calculate the spatial attention weight according to the processing result of the input tensor X. The formula for calculating the spatial attention weight is: , where, is the feature at the -th spatial position, is the feature at the -th position among all spatial positions, is the spatial attention weight, is the exponential function used to perform the exponential operation on the calculation result, is the -th spatial position feature calculated attention score, which measures the degree of attention of the model to the feature at this position. The higher the score, the more important the model considers the feature at this position for focusing on the key areas of the exhibition hall (such as the popular exhibits area); C4. Combine the convolutional result with the time series data and use the gated recurrent algorithm to learn the dependence of crowd flow in the time dimension. The formula expression of the gated recurrent algorithm is: , where, is the hidden state at time t, is the input data at time t, is the graph convolutional operation that can perform feature extraction and transformation on graph-structured data, is the adjacency matrix in the graph structure, which describes the connection relationship and weights between nodes in the graph, The gated recurrent algorithm that controls information flow through a gating mechanism is the hidden state at time t-1; C5. Calculate the temporal attention weights based on the temporal dependencies of crowd flow. The formula for the temporal attention weights is: , where is the feature at the t-th time step, is the feature at the -th time step among all time steps, is the temporal attention weight, is the exponential function used to perform an exponential operation on the calculation result of , is the -th time step feature The calculated attention score; C6. Combine the temporal attention weights, spatial attention weights, spatial features, and temporal features to establish a prediction model. The formula expression of the prediction model is: , where is the predicted heatmap, is the spatial attention weight, is element-wise multiplication to integrate the weighted spatial and temporal features, is the spatial feature matrix, is the temporal attention weight, is the temporal feature matrix; C7. Train the prediction model with the standardized dataset data and optimize the parameters of the prediction model using a loss function. The optimized prediction model is a real-time heat prediction model.

[0022] The process of crowd distribution trend prediction is as follows: D1. Real-time collect multi-source data from 6 - 8 minutes before the current moment and normalize the collected data; D2. Pass the normalized data through the heat prediction model for prediction to obtain the prediction result; D3. Restore the prediction result to the actual heat level. The formula for the heat level restoration algorithm is: , where is the actual heat level, is the normalized heat level, is the global maximum crowd density value, is the global minimum crowd density value; D4. Convert the predicted heat level to a color heatmap, using different colors to represent the crowd density levels; S5. Dynamically adjust the control strategy according to the predicted population distribution trend and business objectives, where the business objectives are exhibit attention, visit fluency, and interactive experience; The control content of the control strategy includes control based on exhibit attention, control based on visit fluency, and control of interactive experience. Among them, the control based on exhibit attention includes optimizing the flow of popular exhibits and activating cold areas; the control method for exhibit attention is: if it is predicted that the heat level of a certain area ≥ 80% and the average stay time > 15 minutes, then adjust the optimization of the flow of popular exhibits, that is, trigger operations to enhance the display effect, intelligent explanation scheduling, and space expansion. Enhancing the display effect is to adjust the lighting brightness of the area (increase by 30%) through the IoT system, increase the frequency of holographic projection or AR interaction ( + 2 times per hour), and attract more attention; Intelligent explanation scheduling is to automatically dispatch mobile tour guide robots to this area (or trigger voice broadcasts), supplement the pressure of manual explanations, and at the same time push detailed information of exhibits to tourists' mobile phones through WiFi probes; Space expansion is to temporarily remove non - critical exhibits around, widen the passage width (such as adjusting the exhibition area layout through electric partitions), and relieve congestion; If it is predicted that the heat level of a certain area ≤ 20% and the average stay time < 5 minutes, then adjust the activation of cold areas, that is, trigger operations of content update and path guidance. Content update is to automatically switch the display form of exhibits (such as rotating exhibition stands, dynamic screen content), or release interactive tasks (such as scanning codes to answer questions to win souvenirs); Path guidance is through the LED screen in the exhibition hall, floor projection arrows, or APP push to guide tourists to this area; The control method based on visit fluency is: when it is predicted that the density value of the passage area ≥ 60 people / m 2 ², and the flow speed < 0.5 m / s, then carry out congestion warning and diversion control, that is, trigger physical diversion, digital guidance, and flow limiting control. Physical diversion is to open the emergency passage (such as automatic unlocking of electric doors), monitor the passage flow through infrared sensors, and dynamically adjust the opening and closing frequency of the gates; Digital guidance is to broadcast in the exhibition hall, mark the red congested area on the real - time heat map of the APP, and recommend a detour route (tracking the location of tourists based on RFID tags and planning the shortest path); Flow limiting control is to count the number of people in the museum in real - time at the entrance through RFID / face recognition. If it exceeds 80% of the maximum carrying capacity, start a time - sharing reservation mechanism (such as restricting the entry speed of new tourists); The control method based on the interactive experience is as follows: If it is predicted that the number of people queuing in the interactive device area (such as the VR experience area) is ≥ 5, and the waiting time > 10 minutes, then perform dynamic scheduling operations on the interactive devices, that is, trigger device expansion, time management, and queuing visualization operations; Device expansion is to remotely start standby interactive terminals (such as VR devices in the sleep state), and push a notification of "New device added, you can experience immediately" through the APP; Time management is to automatically shorten the duration of each experience (such as adjusting from 8 minutes to 5 minutes), and at the same time increase the number of sessions (such as +3 sessions per hour); Queuing visualization is to set up an electronic screen in this area to display the real-time queuing number and the estimated waiting time, reducing the anxiety of tourists; S6. Monitor the change of the heat distribution after the dynamic adjustment in real time, and optimize the parameters of the heat prediction model according to the change of the heat distribution.

[0023] The process of optimizing the parameters of the heat prediction model is as follows: E1. Collect actual heat data through a thermal imaging camera every two minutes, compare it with the predicted heat map, and calculate the error value of each grid:

[0024] If the average error of the area > 15%, trigger the model optimization process; E2. Add the latest actual data (including the adjusted control strategy label) to the standardized data set, and update the parameters of the convolutional neural network and the spatio-temporal graph neural network in an incremental learning manner; E3. If the spatial attention weight of a certain area is continuously low but the actual flow of people is dense, increase the weight of the features of this area in the spatial attention calculation, such as adjusting the bandwidth parameter of the Gaussian kernel function ; If the prediction error in a certain period of the time series is large, then enhance the time attention weight of this period, such as adjusting the forgetting gate parameter of the gated recurrent unit; E4. Adopt the sliding window method to verify the optimized model using the data of the past 1 hour. If the prediction error is reduced to within 5%, save the new model; otherwise, repeat E1 - E5; Obtain crowd density distribution data, behavior data, and environmental data through thermal imaging cameras, WiFi probes, and infrared sensors, and generate a heat map; by fusing and annotating discrete density points into a dataset of continuous heat fields, visually display the crowd distribution characteristics, and use the combination of convolutional neural networks and spatio-temporal graph neural networks to effectively model the spatio-temporal correlation of crowd flow. Based on the prediction results 6-8 minutes in advance, a closed-loop of prediction-adjustment-verification is achieved. By dynamically adjusting the control strategy according to the predicted crowd distribution trend and business objectives, it is possible to dynamically adjust the guided tour route or interaction form according to the crowd flow density, and can also adjust the exhibit layout in advance, reduce the number of round trips of visitors, improve the overall flow efficiency, and thus enhance the display effect of the AI exhibition hall.

[0025] Based on the AI exhibition hall control method based on crowd heat distribution monitoring described above, the present invention also provides an AI exhibition hall control system based on crowd heat distribution monitoring, as Figure 5 shown, including a data acquisition module, a heat map generation module, a data processing and integration module, a real-time prediction module, an execution control module, and a feedback optimization module. The data acquisition module obtains the original data stream in the exhibition hall in real time through multi-source sensor devices. The original data stream includes crowd heat coordinates, stay duration, moving speed, and environmental parameters; the multi-source sensor devices generate the original data stream by deploying thermal imaging cameras, infrared sensors, WiFi probes, RFID tag readers, and environmental sensors. The heat map generation module converts the sensed data into a visual heat distribution for real-time monitoring and analysis; The data processing and integration module cleans, complements, and integrates the multi-source data to form a standardized dataset of continuous heat fields; The real-time prediction module uses the combination of convolutional neural networks and spatio-temporal graph neural networks to establish a real-time heat prediction module to predict the crowd distribution trend 6-8 minutes in advance; The execution control module dynamically adjusts the exhibition hall resources according to the prediction results and business objectives. The exhibition hall resources include optimization of exhibit attention, optimization of fluency, and optimization of interactive experience; The feedback optimization module is used to construct a closed-loop of monitoring and optimization, real-time monitor the change of heat distribution after dynamic adjustment, and optimize the parameters of the heat prediction model according to the change of heat distribution; Through the cooperation of the data acquisition module, the heat map generation module, the data processing and integration module, the real-time prediction module, the execution control module, and the feedback optimization module, the full-chain intelligence from data acquisition to business implementation is realized. The modules are linked by data streams, combined with deep learning and dynamic control, which not only meets the real-time monitoring requirements of the exhibition hall, but also dynamically adjusts the guided tour route or interaction form according to the crowd flow density, and enhances the display effect of the AI exhibition hall.

[0026] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An AI exhibition hall control method based on crowd heat distribution monitoring, characterized in that: It includes the following steps: S1. Install multi-source sensor devices on the ceiling, wall or floor of the exhibition hall to build a full-coverage perception network. The multi-source sensor devices include thermal imaging cameras, infrared sensors, WiFi probes, RFID tag readers and environmental sensors; S2. Real-time capture crowd density distribution data, behavior data and environmental data through the multi-source sensor devices, and generate a heat map; S3. Filter the real-time data obtained by the multi-source sensor devices, then fuse the heat distribution data, behavior data and environmental data, and label the heat characteristics of each area to form a dataset of a continuous heat field; S4. Establish a real-time heat prediction model by combining a convolutional neural network and a spatio-temporal graph neural network, and predict the crowd distribution trend within 6-8 minutes according to the heat characteristics; S5. Dynamically adjust the control strategy according to the predicted crowd distribution trend and business objectives; S6. Real-time monitor the change of the heat distribution after dynamic adjustment, and optimize the parameters of the heat prediction model according to the change of the heat distribution.

2. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 1, characterized in that: The resolution of the thermal imaging camera Pixels, deployment spacing ≤ 5 m; The WiFi probe supports Bluetooth 5.0 + WiFi6 protocol, positioning error ≤ 1.5 m; The infrared sensor coverage angle ≥ 120°, detection distance ≤ 8 m.

3. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 2, wherein: The process of generating the heat map is as follows: A1. Divide the field of view of the thermal camera into multiple pixel blocks, calculate the number of human targets in each pixel block, and calculate the density value in combination with the actual area corresponding to each pixel block; A2. Smooth the discrete density points to generate a continuous density field; A3. Map the density value to a heat level of 0-100; A4. Adopt a cold and warm gradient method for color coding for high-density and low-density distributions, and output the heat map after rendering through GIS.

4. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 3, wherein: The process of establishing the dataset of the continuous heat field is as follows: B1. Identify and eliminate data deviating from the normal range through an outlier detection algorithm; B2. Fill in the missing data using linear interpolation, and filter the data using a low-pass filter; B3. Fuse the heat distribution data, behavior data and environmental data according to the time stamp to obtain a visual record containing multi-dimensional information; B4. Calculate the crowd density according to the number of people in the area and the area of the area, which is used as the basic index of the heat characteristics, and add labels to each area; B5. Normalize the numerical data, and then organize the normalized data into a structured table or database table structure; B6. Add meta-information of the dataset to the structured table or database table structure to form a standardized dataset.

5. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 4, wherein: In step B3, the fusion process is as follows: B31. Synchronize the heat distribution data, behavior data and environmental data according to the time stamp to make different types of data consistent in the time dimension; B32. Map the heat distribution data, behavior data and environmental data to a unified spatial coordinate system to make the data comparable in space; B33. Map the characteristics of different types of data to the color and intensity values in the heat map, unify the data scale through normalization, and display the data distribution density with the depth of color on a two-dimensional plane to form a visual record containing multi-dimensional information.

6. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 5, characterized in that: In step S4, the process of establishing the real-time heat prediction model is as follows: C1. Divide the standardized data according to time windows and spatial grids to form an input tensor; C2. Process the input tensor through a multi-layer convolutional neural network; C3. Calculate the spatial attention weights based on the processing result of the input tensor; C4. Combine the convolution result with the time series data and use the gated recurrent algorithm to learn the dependencies of crowd flow in the time dimension; C5. Calculate the temporal attention weights based on the dependencies of crowd flow in the time dimension; C6. Establish a prediction model by combining the temporal attention weights, spatial attention weights, spatial features, and temporal features; C7. Train the prediction model with the standardized dataset data and optimize the parameters of the prediction model using a loss function. The optimized prediction model is a real-time heat prediction model.

7. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 6, wherein: The process of predicting the crowd distribution trend is as follows: D1. Real-time collect multi-source data for the 6 - 8 minutes before the current moment and normalize the collected data; D2. Pass the normalized data through the heat prediction model for prediction to obtain the prediction result; D3. Restore the prediction result to the actual heat level; D4. Convert the predicted heat level into a color heat map, using different colors to represent the crowd density levels.

8. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 7, characterized in that: The control content of the control strategy includes control based on exhibit attention, control based on visit fluency, and control of interactive experience; among them, the control based on exhibit attention includes optimizing the flow of popular exhibits and activating unpopular areas; The control method based on the visit fluency is as follows: when it is predicted that the density value of the passage area ≥ 60 people / m 2 , and the flow velocity < 0.5 m / s, congestion warning and guidance control are carried out; The control method for interactive experience is: if it is predicted that the number of people queuing in the interactive device area ≥ 5 and the waiting time > 10 minutes, then perform dynamic scheduling operations on the interactive devices.

9. The AI exhibition hall control method based on crowd heat distribution monitoring according to claim 8, characterized in that: The process of optimizing the parameters of the heat prediction model is as follows: E1. Collect actual heat data through a thermal imaging camera every two minutes, compare it with the predicted heat map, and calculate the error value for each grid; E2. Add the latest actual data to the standardized dataset and update the parameters of the convolutional neural network and spatio-temporal graph neural network in an incremental learning manner; E3. If the spatial attention weight of a certain area is continuously low but the actual crowd flow is dense, increase the weight of the features of this area in the spatial attention calculation; If the prediction error is large in a certain period of the time series, then enhance the temporal attention weight of this period; E4. Use the sliding window method to verify the optimized model with the data of the last 1 hour. If the prediction error is reduced to within 5%, then save the new model; otherwise, repeat steps E1 - E5.

10. An AI exhibition hall control system based on crowd heat distribution monitoring, based on the AI exhibition hall control method based on crowd heat distribution monitoring according to any one of claims 1-9, characterized in that: Including: A data collection module that real-time obtains the original data stream in the exhibition hall through multi-source sensor devices. The original data stream includes crowd heat coordinates, stay duration, movement speed, and environmental parameters; A heat map generation module that converts the perceived data into a visual heat distribution for real-time monitoring and analysis; A data processing and integration module that cleans, completes, and integrates the multi-source data to form a standardized dataset of a continuous heat field; A real-time prediction module that establishes a real-time heat prediction module by combining a convolutional neural network and a spatio-temporal graph neural network to predict the crowd distribution trend 6 - 8 minutes in advance; An execution control module that dynamically adjusts the exhibition hall resources according to the prediction result and business objectives. The exhibition hall resources include optimizing exhibit attention, optimizing fluency, and optimizing interactive experience; A feedback optimization module is used to construct a closed-loop of monitoring and optimization, monitor the change of the thermal distribution after dynamic adjustment in real time, and optimize the parameters of the thermal prediction model according to the change of the thermal distribution.

Images