People flow analysis system based on internet of things information acquisition

Abstract

Disclosed in the present invention is a people flow analysis system based on Internet of Things information acquisition. The system comprises a data acquisition module, a data storage module, a people flow statistics module, an obstruction analysis module, a coupling analysis module and a predictive analysis module. The data acquisition module acquires all people flow data of a laboratory server room by means of an image acquisition apparatus; then, the people flow statistics module separately performs crowd density analysis, wherein an average distance is determined by means of hierarchical and clustering analysis of head labeling points, and further a Gaussian function is used to determine the size of heads, thus solving the problem of uncertainty caused by a nearest neighbor method when crowd densities are low and uneven; the obstruction analysis module performs feature fusion on features of an obstacle in a coverage space and neighboring crowds to obtain a new data set; the predictive analysis module performs overall people flow predictive analysis on the basis of the new data set and people flow data of crowds not adjacent to the obstacle. Therefore, the accuracy of people flow analysis and control efficiency are greatly improved.

Description

A crowd flow analysis system for collecting information from the Internet of Things Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a crowd flow analysis system for collecting information from the Internet of Things. Background Art

[0002] The crowd flow analysis system analyzes the dynamic changes of crowd flow in different time periods by counting the number of pedestrians and the spatial distribution of the crowd, so as to better understand the behavioral habits of pedestrians and take corresponding safety measures to reduce walking congestion. In the laboratory room, the experimental equipment will block the pedestrians, which will reduce the accuracy of the crowd flow statistical analysis. In addition, the crowd distribution density in different spaces is different. Although the closer the crowd distribution density is, the more accurate the model analysis result is, but during the peak period of crowd flow, the crowd density changes rapidly and the accuracy of the model analysis decreases. In the same space, people at different positions affect each other and congestion is very likely to occur. When analyzing the crowd in the entire space, the accuracy of the analysis is affected by obstacles. The accuracy of the analysis based on the distance between people is not suitable for all crowd density analyses. In order to improve traffic efficiency and meet the accuracy and real-time requirements of the crowd flow analysis process, the present invention proposes a crowd flow analysis system for Internet of Things information collection.

[0003] Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the purpose of the present invention is to provide a crowd flow analysis system for Internet of Things information collection. The crowd flow statistics module of the system analyzes the crowd density in different coverage spaces, determines the average distance through stratification and clustering analysis of the head marking points, and then uses the Gaussian function to determine the size of the head, which solves the uncertainty problem caused by the nearest distance method when the crowd density is low and uneven. The obstacle analysis module fuses the features of the obstacles in the coverage space with the features of the adjacent crowds to obtain a new data set. The prediction analysis module then performs an overall crowd flow prediction analysis based on the fused data set and the crowd flow data of the crowd without adjacent obstacles, which greatly improves the efficiency of remote crowd flow analysis and control, and enables the data collected by different Internet of Things collection devices to be used for overall crowd flow analysis.

[0005] A crowd flow analysis system for collecting information from the Internet of Things, comprising a data collection module, a data storage module, a crowd flow statistics module, an obstruction analysis module, a coupling analysis module, and a prediction analysis module;

[0006] The data acquisition module collects all the human traffic data in the laboratory computer room through the image acquisition device and stores the human traffic data in the data storage module;

[0007] The image acquisition devices are installed at different locations in the laboratory computer room, dividing the computer room into different coverage spaces. The crowd flow statistics module then performs crowd density analysis on the crowd flow data in different coverage spaces to obtain corresponding loss functions, crowd density, and crowd flow. The loss function represents the error between the model analysis results of the crowd flow statistics model and the actual crowd value.

[0008] The coupling analysis module combines the analysis results of the crowd flow statistics module to analyze the changes in crowd positions to obtain the coupling degree of crowds in different coverage spaces, and calculates the exchange rate of the number of people in adjacent coverage spaces;

[0009] The obstruction analysis module analyzes the loss function of the statistical analysis process and the impact of experimental equipment on the change of the direction of human flow to obtain an obstruction factor, and then combines the coupling degree analyzed by the coupling analysis module to perform a position analysis of the human flow obstruction factor in the laboratory computer room;

[0010] The obstacle analysis module fuses the features of the crowd near the obstacle location in the laboratory according to the analysis results of the obstacle factor, and fuses the features of the obstacle location and the crowd location to obtain a new data set;

[0011] The prediction analysis module performs crowd density analysis based on the analysis results of the damage analysis module and the fused new data set to obtain a density map of the crowd in the entire space, and then predicts the crowd density based on the density map to obtain a prediction result. The prediction result is used to evaluate the safety of the crowd density in the space and formulate management measures for experimental equipment.

[0012] The image acquisition device collects the human flow data in the computer room from different angles. The human flow counting module detects the human head features and counts the crowd through the generated density map. The specific process is as follows:

[0013] Step 1: The head size of each person is inversely proportional to the distance from the image acquisition device. i Indicates the position of the head in the image. The annotation points in the image are divided into different levels according to the distance. Different levels correspond to different weight values. Cluster analysis is performed on the positions of the annotation points in different levels to obtain different cluster centers. The number of cluster centers is the same as the number of levels of the annotation points.

[0014] Step 2: Convert the crowd image annotation into a crowd density map according to different weighted values.

[0015] x i The head at the position is represented by the Diktral function δ(xx i ) is used to represent it, N represents the number of labeled points on the human head, and then the Gaussian kernel function is used to simulate the head;

[0016] Step 3: The Gaussian kernel function diffuses the human head into a range. The range value is related to the head size.

[0017] in, i represents the subscript of the marked point, β i is the weighted value of the annotation points at different levels, n is the number of cluster centers, is the Gaussian kernel function, Represents the average distance from the labeled points to the cluster center within a layer;

[0018] Step 4: Use the deep learning neural network to train all the marked points to output a density map, and then perform an integral operation on the density map to obtain the number of people.

[0019] The obstruction analysis module analyzes the movement patterns of people to determine the obstruction factors that affect the movement of people in the computer room. Different coverage areas include different obstacles. The specific analysis process is as follows:

[0020] Step 1: Establish a biomathematical state analysis equation based on the distribution of crowd groups in space, extract the activity factors of crowd diffusion, and the dynamic equation of crowd group diffusion: Q(x,t)=ρ(x,t)V(x,t),

[0021] Where Q is the flow rate, ρ is the crowd density, V is the average speed,

[0022] Step 2: When the average speed is the maximum, V(ρ max )=0

[0023] Among them, ρ(x,t) represents the density at x at time t, m is the total number of people in the measurement area, and c(x j (t)-x) is a Gaussian function, x j (t) represents the coordinate vector of pedestrian j at time t, and the measurement area is based on x j (t) is related to the position of x;

[0024] Step 3: Analyze and extract the obstruction factor based on the change of the position vector in the crowd diffusion dynamic system, perform feature extraction on the image, extract the object features of the obstruction in the image, and determine the location of the obstruction in the computer room based on the obstruction factor;

[0025] Step 4: Fuse the target features of the crowd near the obstacle with the features of the obstacle. Perform feature fusion and calculate based on the position of the pixel points. The coordinate of the head of the person in the crowd is x i , and x i The pixel position of the adjacent obstacle is recorded as y i, different positions correspond to different feature vectors, and the feature fusion formula is as follows: Y=W(x1+y2),

[0026] Where x1 and y2 represent feature vectors of the same dimension, y∈Y is the fused feature map set, and W(x1+y2)∈R n ,Through feature fusion, the obstacles and the adjacent crowd are fused to obtain a new dataset.

[0027] The data acquisition module includes different image acquisition devices. The data acquisition module numbers the image acquisition devices and then installs the image acquisition devices with different numbers at different locations in the laboratory computer room. The range covered by the different image acquisition devices includes overlapping parts and non-overlapping parts. The coupling analysis module analyzes the image of the crowd distribution in the overlapping part, and the crowd flow statistics module analyzes the crowd flow distribution pattern in the non-overlapping part.

[0028] The crowd counting module extracts the characteristics of the crowd in the space covered by the image acquisition device, and performs data training based on the crowd flow data and density map to obtain the training loss error between the predicted crowd and the actual crowd. The formula for the loss error is as follows:

[0029] Where N represents the number of images, F(x;θ) and F are points in the density map.

[0030] Obstacles that hinder people's walking in the laboratory have the same degree of pressure on the adjacent groups of people. The obstacle analysis module analyzes the obstacles and the human flow as a whole by fusing and analyzing the data sets of the people adjacent to the obstacles. When the prediction analysis module analyzes the people in all covered spaces as a whole, the impact of different parts of the obstacles on the human flow is taken into account.

[0031] The prediction and analysis module performs an overall crowd density analysis based on the data collected by all image acquisition devices in the space. The obstacle analysis module and the crowd statistics module obtain a complete data set by analyzing the crowd density in different coverage spaces. The prediction and analysis module then predicts the crowd density at different times on the data set, and analyzes the movement path of the crowd in the entire space based on the prediction results.

[0032] The coupling analysis module analyzes the transformation of crowd density in different coverage spaces, obtains the change of crowd number at different times through the crowd flow statistics module, and then determines the position change of crowd flow according to the starting coverage space and ending coverage space of the crowd flow.

[0033] Due to the adoption of the above technical solution, the present invention has the following advantages compared with the prior art:

[0034] 1. The crowd counting module of the present invention divides the head annotation points into layers, then performs cluster analysis on the annotation points at different layers, and calculates the average distance between the head annotation at the cluster center and the annotation points of adjacent heads. This avoids the inapplicability of the distance coefficient when selecting other data sets, solves the uncertainty problem caused by the nearest distance method when the crowd density is low and uneven, and improves the accuracy of head size estimation.

[0035] 2. The obstruction analysis module of the present invention determines the obstruction factors that hinder human flow based on the dynamic analysis process of human flow in different coverage areas, and determines the location of the obstructions that hinder human flow within the coverage area based on the obstruction factors. The feature vectors of the people near the obstructions are fused with the feature vectors of the obstructions to obtain a new population data set. The obstruction analysis module analyzes the obstruction pressure brought by the obstacles to human flow through feature fusion, making the prediction analysis module more accurate and real-time in the overall human flow analysis process.

[0036] 3. The prediction and analysis module of the present invention combines the crowd flow data collected at different locations to perform overall prediction and analysis. The crowd density distribution in the crowd images collected at different locations is different. In the same space, all the movement trajectories of the crowds affect each other. The prediction and analysis module combined with the analysis results of the obstacle analysis module greatly improves the efficiency of crowd flow prediction, and timely management measures are taken for the safety hazards caused by dense crowds. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] FIG1 is an overall analysis flow chart of the present invention;

[0038] FIG2 is an overall module diagram of the present invention;

[0039] FIG3 is an analysis flow chart of a crowd counting module of the present invention;

[0040] FIG4 is an analysis flow chart of the obstacle analysis module of the present invention. DETAILED DESCRIPTION

[0041] The foregoing and other technical contents, features, and effects of the present invention will be clearly presented in the following detailed description of the embodiments with reference to Figures 1 to 4. The embodiments and features of the embodiments of the present application may be combined with each other. The terms used in the description have the meanings commonly understood by those skilled in the art in the art of the present invention.

[0042] The crowd flow analysis system needs to process a large amount of data in a short period of time and give analysis results in time to meet the real-time requirements and enhance the real-time control requirements of crowd flow. It uses the statistical and visualization methods of data analysis to predict the flow direction of crowd flow in the computer room space, so that the crowd flow can be analyzed and managed more in real time. At the same time, the distribution of crowd flow in the space is of great significance to the spatial distribution of crowd flow in the computer room. The experimental instruments in the computer room of the school laboratory will block the flow of pedestrians. When students leave the computer room, the movement path of students is optimized through the analysis of the movement rules of crowd flow, the traffic efficiency in the computer room is improved, and congestion is avoided. At the same time, the analysis of the different crowd flow movement statistics in the computer room helps the computer room manager improve the efficiency of the placement of facilities in the computer room.

[0043] The crowd density distribution in different coverage spaces is different. When the crowd density is close, the model analysis process is more accurate. When the flow of people in the entire space changes or there is an emergency, the degree of change in the crowd density will change, and the model analysis process will greatly reduce the accuracy of the analysis.

[0044] In the same space, the crowd flow data of different coverage areas are coupled. The coupling degree of data under different analysis models is obtained by calculation. When predicting and analyzing the crowd flow in the entire space, obstacles exert pressure on the direction of the crowd flow and change the movement trajectory of the crowd flow. In order to analyze the impact of obstacles, the characteristics of the obstacles are fused with the characteristics of the nearby crowds to obtain a new data set. When predicting the overall crowd flow, the analysis of obstacles is weakened, but the impact of obstacles is taken into account. The accuracy of the analysis results varies when training models for different training data sets. The prediction analysis module predicts the trajectory analysis results of the crowd flow fitting the obstacle boundary based on the fused data set.

[0045] Managers can optimize resource allocation based on the results of crowd flow analysis, understand the needs and behavioral characteristics of different groups of people more accurately, and thus improve resource utilization efficiency. By analyzing the movement patterns and behavioral habits of pedestrians in the laboratory room, they can predict future pedestrian flow and traffic conditions, provide decision support for laboratory managers, and formulate corresponding management measures in advance.

[0046] The embodiment of the present invention discloses a crowd flow analysis system for collecting information from the Internet of Things, including a data collection module, a data storage module, a crowd flow statistics module, an obstruction analysis module, a coupling analysis module, and a prediction analysis module;

[0047] The data acquisition module collects all the crowd data in the laboratory computer room through the image acquisition device and stores the crowd data in the data storage module. The image acquisition device includes a camera, a thermal sensor and an infrared instrument. The image acquisition devices at different positions have different clarity of the collected images due to the different distances from the crowd. In addition, the degree of obstruction caused by the experimental equipment in the laboratory computer room in different coverage spaces is also different.

[0048] The image acquisition devices are installed at different locations in the laboratory computer room, dividing the computer room into different coverage spaces. The crowd statistics module then performs crowd density analysis on the crowd flow data in different coverage spaces to obtain corresponding loss functions, crowd density, and crowd flow. The loss function represents the error between the model analysis results of the crowd statistics model and the actual crowd value. The crowd density distribution in different coverage spaces is different. The analysis of images with different crowd densities by the crowd statistics module greatly improves the accuracy and real-time performance of the crowd flow analysis.

[0049] The coupling analysis module combines the analysis results of the crowd flow statistics module to analyze the changes in crowd position and obtain the coupling degree of crowds in different coverage spaces, and calculates the exchange rate of the number of people in adjacent coverage spaces. The implementation room is a fixed space, and students use experimental equipment to conduct experimental learning. During non-peak hours, the flow of people changes slowly, and the crowd density in different coverage spaces transforms into each other. The coupling degree is used to describe the degree of change of the crowd flow in different coverage spaces.

[0050] The obstruction analysis module analyzes the loss function of the statistical analysis process and the impact of experimental equipment on the change of the direction of human flow to obtain an obstruction factor, and then combines the coupling degree analyzed by the coupling analysis module to perform a position analysis of the human flow obstruction factor in the laboratory computer room;

[0051] The obstruction analysis module fuses the features of people near obstructions within the laboratory based on the analysis results of the obstruction factors, fusing the features of the obstruction location with the features of the crowd location to generate a new data set. When the prediction analysis module performs a holistic analysis based on all the crowd data, no single image acquisition device can capture all unobstructed crowd distribution images. However, within the entire laboratory computer room, the direction of human traffic during peak hours follows the predicted analysis results. The obstruction analysis module addresses the pressure of obstacles on human traffic movement during the overall analysis process.

[0052] The prediction analysis module performs crowd density analysis based on the analysis results of the damage analysis module and the fused new data set to obtain a density map of the crowd in the entire space, and then predicts the crowd density based on the density map to obtain a prediction result. The prediction result is used to evaluate the safety of the crowd density in the space and formulate management measures for experimental equipment.

[0053] Furthermore, the image acquisition device collects the human flow data in the computer room from different angles. The human flow counting module detects the head features of the human body and counts the crowd through the generated density map. The specific process is as follows:

[0054] Step 1: If the nearest distance method is used to calculate the average distance between all the marked points and the marked points of the adjacent heads in the coverage space, and then the distance coefficient is selected based on the average value, the applicability to other data sets will be reduced. When the crowd density is low and uneven, the nearest distance method will bring uncertainty problems, thereby affecting the accuracy of head size estimation. Preferably, the marked points in the coverage space are pre-divided and clustered, and then the average distance is calculated. The head size of each person is inversely proportional to the distance from the image acquisition device. The distance coefficient is calculated based on the distance between the marked points x and the image acquisition device. i Indicates the position of the head in the image. The annotation points in the image are divided into different levels according to the distance. Different levels correspond to different weight values. Cluster analysis is performed on the positions of the annotation points in different levels to obtain different cluster centers. The number of cluster centers is the same as the number of levels of the annotation points.

[0055] Step 2: Convert the crowd image annotation into a crowd density map according to different weighted values.

[0056] x i The head at the position is represented by the Diktral function δ(xx i ) is used to represent it, N represents the number of labeled points on the human head, and then the Gaussian kernel function is used to simulate the head;

[0057] Step 3: The Gaussian kernel function diffuses the human head into a range. The range value is related to the head size.

[0058] in, i represents the subscript of the marked point, β i is the weighted value of the annotation points at different levels, n is the number of cluster centers, is the Gaussian kernel function, Represents the average distance from the labeled points to the cluster center within a level, σ i Determines the accuracy of Gaussian convolution;

[0059] Step 4: Use the deep learning neural network to train all the marked points to output a density map, and then perform an integral operation on the density map to obtain the number of people.

[0060] During peak hours of human movement, it is necessary to analyze the human flow in the entire space corresponding to the laboratory computer room in order to remotely control the overall human flow based on the human flow analysis rules. However, the data in the space collected by the image acquisition device has a high occlusion rate. Furthermore, the obstacle analysis module integrates the characteristics of obstacles and the crowd to obtain a new data set. The obstacle analysis module analyzes the human flow movement rules to determine the obstacle factors that affect human flow in the computer room. Different coverage areas include different obstacles. The specific analysis process is as follows:

[0061] Step 1: Establish a biomathematical state analysis equation based on the distribution of crowd groups in space, extract the activity factors of crowd diffusion, and the dynamic equation of crowd group diffusion: Q(x,t)=ρ(x,t)V(x,t),

[0062] Where Q is the flow rate, ρ is the crowd density, V is the average speed,

[0063] Step 2: When the average speed is the maximum, V(ρ max )=0

[0064] Among them, ρ(x,t) represents the density at x at time t, m is the total number of people in the measurement area, and c(x j (t)-x) is a Gaussian function, x j (t) represents the coordinate vector of pedestrian j at time t, and the measurement area is based on x j (t) is related to the position of x;

[0065] The value range of R0 is related to the size of the head at the marked point.

[0066] Step 3: During the dynamic change of crowd diffusion, the change in the pressure of the obstacle corresponding to the moment of the dynamic factor change is recorded as the obstruction factor. The degree of obstruction varies depending on the location of the obstacle. The experimental equipment in the laboratory computer room hinders the movement of students. The obstruction factor is extracted based on the change of the position vector in the dynamic system of crowd diffusion. Feature extraction is performed on the image to extract the object features of the obstacle in the image. The location of the obstacle in the computer room is determined based on the obstruction factor, and the distribution characteristics of the nearby crowd are determined based on the location of the obstacle.

[0067] Step 4: Fuse the target features of the crowd near the obstacle with the features of the obstacle. Perform feature fusion and calculate based on the position of the pixel points. The coordinate of the head of the person in the crowd is x i , and x i The pixel position of the adjacent obstacle is recorded as y i, different positions correspond to different feature vectors, and the feature fusion formula is as follows: Y=W(x1+y2),

[0068] Where x1 and y2 represent feature vectors of the same dimension, y∈Y is the fused feature map set, and W(x1+y2)∈R n ,Through feature fusion, the obstacles and the adjacent crowds are fused to obtain a new dataset, w i is the fusion weight, and i is the subscript.

[0069] The data acquisition module includes different image acquisition devices. The data acquisition module numbers the image acquisition devices and then installs the image acquisition devices with different numbers at different locations in the laboratory computer room. The range covered by the different image acquisition devices includes overlapping parts and non-overlapping parts. The coupling analysis module analyzes the image of the crowd distribution in the overlapping part, and the crowd flow statistics module analyzes the crowd flow distribution pattern in the non-overlapping part.

[0070] The crowd counting module extracts the characteristics of the crowd in the space covered by the image acquisition device, and performs data training based on the crowd flow data and density map to obtain the training loss error between the predicted crowd and the actual crowd. The formula for the loss error is as follows:

[0071] Where N represents the number of images, F(x;θ) and F are points in the density map. The population density distribution in different coverage areas is different, and the loss error generated during the corresponding data analysis is also different.

[0072] Obstacles that hinder people's walking in the laboratory exert the same pressure on the adjacent crowds. When the characteristics of the crowd distribution change, the effect of the obstruction position and the obstruction pressure of the crowd do not change. Therefore, when the overall crowd flow is predicted and analyzed, the impact of the obstruction remains unchanged. The obstruction analysis module analyzes the obstruction and the crowd flow as a whole by fusing the data sets of the crowds adjacent to the obstruction. When the prediction analysis module analyzes the crowds in all covered spaces as a whole, the impact of different parts of the obstruction on the crowd flow is taken into account. The calculation of the obstacle pressure is: P(t) = ρ(t)Var(V), Var t (V)=<[V(x,t)-<V(x,t)> ] 2 >,

[0073] Where <·> represents the average value of the observation center point x, and Var(V) represents the change in the speed of the crowd.

[0074] The prediction and analysis module performs an overall crowd density analysis based on the data collected by all image acquisition devices in the space. The obstacle analysis module and the crowd statistics module obtain a complete data set by analyzing the crowd density in different coverage spaces. The prediction and analysis module then predicts the crowd density at different times on the data set, and analyzes the movement path of the crowd in the entire space based on the prediction results.

[0075] The coupling analysis module analyzes the transformation of crowd density in different coverage spaces, obtains the changes in the number of people at different times through the crowd flow statistics module, and then determines the position changes of the crowd flow based on the starting coverage space and ending coverage space of the crowd flow. There is also a coupling relationship between the obstacles detected in different coverage areas. The coupling analysis module analyzes the overlapping and connected parts between the obstacles.

[0076] When the present invention is specifically used, the system includes a data acquisition module, a data storage module, a crowd statistics module, an obstacle analysis module, a coupling analysis module, and a prediction analysis module. The data acquisition module collects all crowd data in the laboratory computer room through an image acquisition device and stores the crowd data in the data storage module. The crowd statistics module of the system analyzes the crowd density of different coverage spaces, determines the average distance through stratification and clustering analysis of head marking points, and then uses a Gaussian function to determine the size of the head, which solves the uncertainty problem caused by the nearest distance method when the crowd density is low and uneven. The obstacle analysis module fuses the features of the obstacles in the coverage space with the features of the adjacent crowds to obtain a new data set. The prediction analysis module then performs an overall crowd prediction analysis based on the fused data set and the crowd data of the crowd without adjacent obstacles, which greatly improves the efficiency of remote crowd analysis and control, and ensures that the data collected by different Internet of Things collection devices are not obstructed when performing overall crowd analysis.

[0077] The above is a further detailed description of the present invention in conjunction with specific embodiments. However, it is readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A crowd flow analysis system for Internet of Things information collection, characterized in that, it includes a data collection module, a data storage module, a crowd flow statistics module, an obstruction analysis module, a coupling analysis module, and a prediction analysis module; The data collection module collects all crowd flow data in the laboratory computer room through an image collection device and stores the crowd flow data in the data storage module; The image collection device is installed at different positions in the laboratory computer room, dividing the computer room into different covered spaces. Then, the crowd flow statistics module respectively analyzes the crowd density of the crowd flow data in different covered spaces to obtain the corresponding loss function, crowd density, and crowd flow rate. The loss function represents the error between the model analysis result of the crowd flow statistics model and the actual crowd value; The coupling analysis module analyzes the change of the crowd position combined with the analysis result of the crowd flow statistics module to obtain the coupling degree of the crowd in different covered spaces, and calculates the exchange rate of the number of people in adjacent covered spaces; The obstruction analysis module analyzes the loss function in the statistical analysis process and the influence of experimental equipment on the change of the crowd flow direction to obtain an obstruction factor, and then combines the coupling degree analyzed by the coupling analysis module to analyze the position of the crowd flow obstruction factor in the laboratory computer room; The obstruction analysis module fuses the characteristics of the crowd near the obstruction position in the laboratory according to the analysis result of the obstruction factor, and fuses the characteristics of the obstruction position and the crowd position to obtain a new data set; The prediction analysis module performs crowd density analysis based on the analysis result of the obstruction analysis module and the fused new data set to obtain a density map of the crowd in the entire space, and then performs crowd density prediction based on the density map to obtain a prediction result, and evaluates the safety of the crowd flow density in the space through the prediction result and formulates management measures for experimental equipment.

2. The crowd flow analysis system for Internet of Things information collection according to claim 1, characterized in that, The image collection device collects the crowd flow data in the computer room from different angles. The crowd flow statistics module uses the head characteristics of the human body for detection and counts the crowd through the generated density map. The specific process is as follows: Step 1: The head size of each person is inversely proportional to the distance from the image acquisition device. Using the marked point x i to represent the position of the human head in the image, the marked points in the image are divided into different levels according to the distance. Different levels correspond to different weighting values. Cluster analysis is performed on the positions of the marked points within different levels to obtain different cluster centers. The number of cluster centers is the same as the number of levels of the marked points; Step 2: Convert the crowd image annotation into a crowd density map according to different weighting values. Take x i The human head at the position is represented by the Dirac function δ(x - x i ), N represents the number of labeled points of the human head, and then the Gaussian kernel function is used for head simulation; Step 3: The Gaussian kernel function spreads the human head into a range, and the value of the range is related to the head size. Among them, The subscript of the marked point represented by i, β i is the weighted value of marked points at different levels, and n is the number of clustering centers, is the Gaussian kernel function, represents the average distance from the labeled points in a layer to the cluster center; Step 4: Use the neural network of deep learning to train all the labeled points to output a density map, and then perform integral operation on the density map to obtain the number of people.

3. The crowd flow analysis system for Internet of Things information collection according to claim 1, characterized in that, The obstruction analysis module analyzes the crowd flow movement law to determine the obstruction factors that affect the crowd flow movement in the computer room. Different covered ranges include different obstructions. The specific analysis process is as follows: Step 1: Establish a state analysis equation of biomathematics according to the distribution of the crowd communities in the space, and extract the activity factors of the crowd diffusion. The kinetic equation of the crowd community diffusion: Q(x,t) = ρ(x,t)V(x,t), where Q is the flow rate, ρ is the crowd density, and V is the average velocity. Step 2. When the average speed is maximum, V(ρ max ) = 0 Among them, ρ(x,t) represents the density at position x at time t, m is the total number of people in the measurement area, and c(x j (t)-x) is a Gaussian function, where x j (t) represents the coordinate vector of pedestrian j at time t, and the measurement area is related to the positions of x j (t) and x; Step 3: Analyze and extract the obstruction factors based on the change of the position vector in the dynamic system of crowd diffusion, perform feature extraction on the image, extract the object features of the obstacles in the image, and determine the positions of the obstacles in the computer room according to the obstruction factors. Step 4: Perform feature fusion by fusing the target features of the crowd near the obstacle with the features of the obstacle, calculate according to the position of the pixel points, and the coordinate of the human head in the crowd is x i , and the pixel point position of the obstacle adjacent to x i is denoted as y i , different positions correspond to different feature vectors, and the feature fusion formula is as follows: Y = W(x 1 + y 2 ), where x 1 and y 2 represent feature vectors of the same dimension, y ∈ Y is the fused feature mapping set, W(x 1 + y 2 ) ∈ R n , and the obstacle and the adjacent crowd are fused through feature fusion to obtain a new data set.

4. The crowd flow analysis system for Internet of Things information collection according to claim 1, characterized in that the data acquisition module includes different image acquisition devices. The data acquisition module numbers the image acquisition devices, and then installs the image acquisition devices with different numbers at different positions in the laboratory computer room. The coverage ranges of different image acquisition devices include overlapping parts and non-overlapping parts. The coupling analysis module analyzes the images of the crowd distribution in the overlapping parts, and the crowd flow statistics module analyzes the crowd flow distribution rules in the non-overlapping parts.

5. The crowd flow analysis system for Internet of Things information collection according to claim 1, characterized in that The above-mentioned crowd flow statistics module extracts the characteristics of the crowd in the space covered by the image acquisition device, and performs data training based on the crowd flow data and the density map to obtain the training loss error between the predicted crowd and the real crowd. The formula for the loss error is as follows: where N represents the number of images, and F(x; θ) and F are points in the density map.

6. The crowd flow analysis system for Internet of Things information collection according to claim 3, characterized in that the pressure degrees of the obstacles that hinder people's walking in the laboratory on the adjacent crowds are the same. The obstruction analysis module analyzes the obstacles and the crowd flow as a whole through the fusion analysis of the data sets of the crowds adjacent to the obstacles. When the prediction analysis module performs an overall analysis of the crowds in all covered spaces, the influences of different parts of the obstacles on the crowd flow are taken into account.

7. The crowd flow analysis system for Internet of Things information collection according to claim 1, characterized in that the prediction analysis module performs an overall crowd flow density analysis based on the data collected by all the image acquisition devices in the space. The obstruction analysis module and the crowd flow statistics module obtain a complete data set through the crowd density analysis under different covered spaces, and the prediction analysis module then predicts the crowd density at different times for the data set, and analyzes the movement paths of the crowd flow in the entire space according to the prediction results.

8. The crowd flow analysis system for Internet of Things information collection according to claim 1, characterized in that the coupling analysis module analyzes the conversion of the crowd density in different covered spaces, obtains the change of the number of people at different times through the crowd flow statistics module, and then determines the position change of the crowd flow according to the starting covered space and the ending covered space of the crowd flow.

Images