Human flow analysis system using Internet of Things (IoT) information collection

Abstract

The present invention discloses a people flow analysis system using Internet of Things (IoT) information collection. The system includes a data collection module, a data storage module, a people flow statistics module, an obstruction analysis module, a combined analysis module, and a predictive analysis module. The data collection module collects all human flow data in the laboratory's machine room through an image collection device and stores the people flow data in the data storage module. The people flow statistics module analyzes crowd densities in different covered spaces and determines the average distance through hierarchical division and cluster analysis of the human head label points, and further determines the human head dimensions using a Gaussian function. When the crowd density is low and not average, the k-nearest neighbor method is used to solve the uncertainty problem. The obstruction analysis module combines the features of obstructions in the covered space with the features of adjacent crowds to obtain a new data set. The predictive analysis module performs overall people flow prediction analysis based on the combined new data set and the people flow data of crowds not adjacent to the obstructions, thereby greatly improving the accuracy of people flow analysis and the efficiency of control.

Description

[Technical Field]

[0001] This invention relates to the technology of intelligent control, and more specifically to a human flow analysis system using Internet of Things (IoT) information collection. [Background technology]

[0002] The pedestrian flow analysis system analyzes data on the dynamic changes in pedestrian flow across different time slots through pedestrian count statistics and the spatial distribution of crowds, thereby better understanding pedestrian behavioral habits and reducing pedestrian congestion by taking appropriate safety measures. In laboratory machine rooms, the accuracy of pedestrian flow statistical analysis decreases when experimental equipment obstructs pedestrians. Crowd distribution densities differ in different spaces, and the closer the crowd distribution densities are, the more accurate the model analysis results become. However, during peak pedestrian flow, the change in pedestrian flow density is rapid, reducing the accuracy of the model analysis. Crowds at different locations within the same space influence each other, making congestion highly likely. When analyzing crowds in the entire space, the accuracy of the analysis is affected by obstructions, and accurately analyzing through the distance between crowds is not suitable for all crowd density analyses. To improve traffic efficiency and meet the requirements for accuracy and real-time performance in the pedestrian flow analysis process, this invention proposes a pedestrian flow analysis system using Internet of Things (IoT) information collection. [Overview of the project] [Means for solving the problem]

[0003] In response to the above situation, and in order to overcome the shortcomings of conventional technology, the present invention aims to provide a pedestrian flow analysis system using Internet of Things (IoT) information collection. The system's pedestrian flow statistics module analyzes crowd density in different coverage spaces, determines the average distance through hierarchical partitioning and cluster analysis of labeled points for human heads, and further determines the dimensions of human heads using a Gaussian function. When the crowd density is low and not average, it solves the uncertainty problem using the k-nearest neighbor method. The obstruction analysis module features the characteristics of obstructions within the coverage space and adjacent crowds to obtain a new dataset. The predictive analysis module performs overall pedestrian flow prediction analysis based on the newly fused dataset and pedestrian flow data of crowds not adjacent to obstructions, thereby significantly improving the efficiency of remote pedestrian flow analysis and control, and enabling overall pedestrian flow analysis for data collected by different Internet of Things (IoT) collection devices.

[0004] A human flow analysis system using Internet of Things (IoT) information collection, comprising a data collection module, a data storage module, a human flow statistics module, an obstruction analysis module, an integration analysis module, and a predictive analysis module, The data acquisition module collects all human flow data within the laboratory's machine room via an image acquisition device, and then stores the human flow data in a data storage module. The image acquisition devices are installed at different locations within the laboratory's machine room. After dividing the machine room into different covered spaces, the pedestrian flow statistics module performs crowd density analysis on the pedestrian flow data within each different covered space to obtain the corresponding loss function, crowd density, and pedestrian flow rate. The loss function represents the error between the model analysis results from the pedestrian flow statistics model and the actual crowd figures. The combined analysis module combines the analysis results of the human flow statistics module to analyze changes in crowd location, obtains the degree of crowd coupling in different coverage spaces, and calculates the exchange rate of crowd numbers within adjacent coverage spaces. The aforementioned inhibition analysis module analyzes the loss function of the statistical analysis process and the effect of experimental equipment on changes in the direction of human flow to obtain inhibition factors. Then, it combines the degree of coupling obtained from the coupling analysis module to perform a positional analysis on the human flow inhibition factors in the machine room of the laboratory. The aforementioned inhibition analysis module, based on the analysis results of the inhibitory factors, features crowd characteristics in the laboratory and close to the inhibition location, and by fusing the features of the inhibition location and the crowd location, a new dataset is obtained. The predictive analysis module performs crowd density analysis based on the analysis results of the inhibition analysis module and the resulting new dataset, obtaining a crowd density map for the entire space. Based on this density map, it then performs crowd density prediction to obtain prediction results. The safety of the human flow density in the space is evaluated through these prediction results, and control measures for the experimental equipment are established.

[0005] The image acquisition device collects pedestrian flow data within the machine room from different angles, and the pedestrian flow statistics module uses human head features to detect and count the crowd through the generated density map, specifically, Each person's head size is inversely proportional to the distance from the image acquisition device, and the labeled point x i Step 1 involves using a method to represent the position of human heads in an image, dividing the label points in the image into different hierarchies according to distance, assigning different weighted averages to the different hierarchies, and performing cluster analysis on the positions of the label points within each hierarchy to obtain different cluster centers, with the number of cluster centers being equal to the number of hierarchies of the label points. Convert the labels of crowd images into crowd density maps based on different weighted averages.

[0006]

number

[0007] x i The Dirichlet function δ(xx) at the position of the human head i This is represented as ), where N represents the number of label points for the human head, and further, step 2 involves performing a head simulation using radial basis functions. The human head is diffused into a single range by the radial basis function, and the value of the range depends on the size of the head,

[0008] [Mathematical Expression]

[0009] Here,

[0010] [Mathematical Expression]

[0011] where i is the subscript of a label point, β i is the weighted average of label points at different hierarchies, n is the number of cluster centers, G σi (x) is the radial basis function, d i - (where " - " represents an overline of d i denotes the average distance from label points to a cluster center within one hierarchy; step 3, step 4: after training all label points using a deep learning neural network to output a density map, performing an integration operation on the density map to obtain the number of people.

[0012] The inhibition analysis module analyzes pedestrian flow movement rules to determine inhibitory factors that affect pedestrian flow movement in the machine room, different coverage areas contain different obstacles, and the specific analysis process is establishing a biomathematical state equation based on the crowd cluster distribution in the space, and extracting activity factors of pedestrian flow diffusion, wherein the dynamic equation of crowd cluster diffusion is

[0013] [Mathematical Expression]

[0014] where step 1: Q represents the flow rate, ρ represents crowd density, and V represents average velocity, When the average speed is at its maximum,

[0015]

number

[0016]

number

[0017] Here, ρ(x,t) is the density of x locations 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 x j Step 2 depends on the position of (t) and x, Step 3 involves analyzing and extracting inhibiting factors based on changes in position vectors in the dynamic system of crowd diffusion, then performing feature extraction on the image to extract object features of the inhibiting object in the image, and determining the position of the inhibiting object in the machine room based on the inhibiting factors. The target features of the crowd adjacent to the obstruction and the features of the obstruction are fused, and calculations are performed based on the position of the pixel points to obtain the x coordinates of the heads of people in the crowd. i Let x i The position of the adjacent pixel point of the obstruction is y i Assuming that different positions correspond to different feature vectors, the feature fusion formula is as follows:

[0018]

number

[0019]

number

[0020] Here, x1 and y2 are feature vectors of the same dimension, y∈Y is the feature mapping set after fusion, and W(x1+y2)∈R nThis includes step 4, which involves fusing the inhibitors and neighboring crowds through feature fusion to obtain a new dataset.

[0021] The data acquisition module includes different image acquisition devices. After numbering the image acquisition devices, the data acquisition module installs the image acquisition devices with different numbers at different locations in the laboratory's machine room. The areas covered by the different image acquisition devices include overlapping and non-overlapping areas. The combined analysis module analyzes the crowd distribution images in the overlapping areas, and the pedestrian flow statistics module analyzes the pedestrian flow distribution rules in the non-overlapping areas.

[0022] The aforementioned pedestrian flow statistics module extracts crowd features within the coverage space of the image acquisition device, then performs data training based on pedestrian flow data and density maps, and obtains the training loss error between the predicted crowd and the actual crowd. The formula for the loss error is as follows:

[0023]

number

[0024] Here, N is the number of images, and F(x;θ) and F represent points in the density map.

[0025] In the laboratory, obstacles that hinder human movement exert the same pressure level on adjacent crowds. The obstacle analysis module performs a fusion analysis on the datasets of adjacent crowds of the obstacles to analyze the obstacles and human flow as a whole. The predictive analysis module considers the impact of different parts of the obstacles on human flow when analyzing the entire crowd within the entire covered space.

[0026] The predictive analysis module performs an overall pedestrian flow density analysis based on data collected by all image acquisition devices in the space, and the obstruction analysis module and the pedestrian flow statistics module obtain a complete dataset by analyzing crowd density under different coverage spaces. The predictive analysis module then further predicts crowd density at different times on the dataset and analyzes pedestrian flow paths within the entire space based on the prediction results.

[0027] The combined analysis module analyzes the transformation of crowd density within different coverage spaces, obtains changes in the number of crowds at different times through the pedestrian flow statistics module, and then determines the change in the position of the pedestrian flow based on the start and end coverage spaces of the pedestrian flow. [Effects of the Invention]

[0028] By employing the above technical means, the present invention has the following advantages compared to the prior art: Firstly, the human flow statistics module of the present invention assigns hierarchies to label points for human heads, performs cluster analysis on label points of different hierarchies, calculates the average distance between the label point of a human head at the cluster center and the label points of adjacent human heads, avoids incompatibility on other datasets when distance coefficients are selected, solves the uncertainty problem of the k-nearest neighbors method when the crowd density is low and not average, and improves the accuracy of head dimension estimation. Secondly, the inhibition analysis module of the present invention determines inhibiting factors that obstruct human flow movement based on a dynamic analysis process of human flow movement in different cover spaces, determines the location of obstructions that obstruct human flow movement within the cover space based on the inhibiting factors, and obtains a new crowd dataset by feature fusing the crowd feature vectors adjacent to the obstructions with the feature vectors of the obstructions. The inhibition analysis module then analyzes the inhibiting pressure that the obstructions exert on human flow movement through feature fusing, thereby enabling the predictive analysis module to have higher accuracy and real-time performance in the overall human flow analysis process. Thirdly, the predictive analysis module of the present invention combines pedestrian flow data collected at different locations to perform overall predictive analysis. Crowd density distributions differ in crowd images collected at different locations, and within the same space, all pedestrian movement trajectories influence each other. The predictive analysis module combines the analysis results of the inhibition analysis module to significantly improve the efficiency of pedestrian flow prediction, allowing for timely management measures against safety hazards that arise when pedestrian flow is dense. [Brief explanation of the drawing]

[0029] [Figure 1] This is an overall analysis flowchart according to the present invention. [Figure 2] This is an overall module diagram according to the present invention. [Figure 3] This is an analysis flowchart for the pedestrian flow statistics module according to the present invention. [Figure 4] This is an analysis flowchart for the inhibition analysis module according to the present invention. [Modes for carrying out the invention]

[0030] The aforementioned and other technical details, features, and effects of the present invention can be clearly shown in the detailed description of the examples in the following formulation reference figures 1 to 4. The embodiments and features of the embodiments of this application may be combined with each other, and the terms used in this specification have the meanings generally understood by those skilled in the art of the present invention.

[0031] To meet the requirements for real-time performance and enhance the need for real-time control of human flow, the human flow analysis system must process large amounts of data in a short time and provide analysis results in a timely manner. By using statistical and visual survey methods based on data analysis, it predicts the flow of human flow within the machine room space, enabling more real-time statistical analysis and management of human flow. Furthermore, the distribution of human flow within a space is of great significance to the spatial distribution of human flow within the machine room. Experimental equipment in machine rooms within school laboratories provides shielding for human flow, and when student human flow leaves the machine room, analyzing the rules of human flow movement optimizes the movement paths of students, improving traffic efficiency within the machine room and avoiding congestion. At the same time, analyzing different human flow movement statistics within the machine room helps machine room managers improve the efficiency of setting up equipment within the machine room.

[0032] When crowd density distributions differ within different covered spaces, and crowd densities are similar, the model analysis process is more accurate. However, when the flow of people within the entire space changes and sudden events occur, the degree of change in crowd density changes, significantly reducing the accuracy of the model analysis process.

[0033] Within the same space, coupling occurs between pedestrian flow data from different coverage areas. Computation yields the degree of coupling under different analysis models. When predicting and analyzing pedestrian flow within the entire space, obstacles exert pressure in the direction of pedestrian flow, altering its trajectory. To analyze the impact of obstacles, feature fusion of obstacle features and adjacent crowd features is performed to obtain a new dataset. This weakens the analysis of obstacles during overall pedestrian flow prediction. However, considering the impact of obstacles, model training is performed on different training datasets, resulting in different accuracy of analysis results. The predictive analysis module predicts based on the fused dataset, obtaining analysis results for pedestrian flow trajectories that fit the boundaries of obstacles. Based on the results of pedestrian flow analysis, administrators can optimize allocation, gain a more accurate understanding of the needs and behavioral characteristics of different people, thereby improving resource utilization efficiency. By analyzing pedestrian movement patterns and behavioral habits within the laboratory's machine room, future pedestrian flow and traffic conditions can be predicted, providing decision-making support to laboratory administrators and enabling the prior formulation of appropriate management measures.

[0034] Embodiments of the present invention disclose a human flow analysis system using Internet of Things (IoT) information collection, comprising a data collection module, a data storage module, a human flow statistics module, an obstruction analysis module, a coupling analysis module, and a predictive analysis module. The data acquisition module collects all human flow data within the laboratory's machine room through an image acquisition device, and then stores the human flow data in a data storage module. The image acquisition device includes a camera, a thermal sensor, and an infrared device. Different image acquisition devices are located at different distances to the crowd, resulting in different image clarity, and the degree of shading generated by experimental equipment within the laboratory's machine room in different covered spaces also varies. The image acquisition devices are installed at different locations within the laboratory's machine room. After dividing the machine room into different coverage spaces, the pedestrian flow statistics module performs crowd density analysis on the pedestrian flow data within each different coverage space, obtaining corresponding loss functions, crowd density, and pedestrian flow rates. The loss function represents the error between the model analysis results from the pedestrian flow statistics model and the actual crowd figures. Since the crowd density distribution differs within different coverage spaces, the accuracy and real-time performance of the pedestrian flow analysis are significantly improved by having the pedestrian flow statistics module analyze images with different crowd densities. The combined analysis module combines the analysis results of the human flow statistics module to analyze changes in crowd position, obtains the degree of crowd coupling in different coverage spaces, calculates the exchange rate of crowd numbers within adjacent coverage spaces, the machine room of the implementation room is a fixed space, students conduct experimental learning using experimental equipment, changes in the laws of human flow motion during non-peak times become gradual, crowd densities in different coverage spaces mutually transform, and the degree of change in human flow within different coverage spaces is described using the degree of coupling. The aforementioned inhibition analysis module analyzes the loss function of the statistical analysis process and the effect of experimental equipment on changes in the direction of human flow to obtain inhibition factors. Then, it combines the degree of coupling obtained from the coupling analysis module to perform a positional analysis on the human flow inhibition factors in the machine room of the laboratory. The aforementioned obstruction analysis module, based on the analysis results of the obstruction factors, features crowd characteristics within the laboratory and near the obstruction location, and merges the obstruction location and crowd location characteristics to obtain a new dataset. When the prediction analysis module performs an overall analysis based on all flow data, there is no image acquisition device that can collect distribution images of all unobstructed crowds. However, within the entire space of the laboratory, the flow direction of people in the entire space during the peak period follows the prediction analysis results, and the obstruction analysis module processes the pressure exerted on the flow motion by obstructions during the overall analysis process. The predictive analysis module performs crowd density analysis based on the analysis results of the inhibition analysis module and the resulting new dataset, obtaining a crowd density map for the entire space. Based on this density map, it then performs crowd density prediction to obtain prediction results. The safety of the human flow density in the space is evaluated through these prediction results, and control measures for the experimental equipment are established.

[0035] Furthermore, the image acquisition device collects pedestrian flow data within the machine room from different angles, and the pedestrian flow statistics module uses human head features to detect and count the crowd through the generated density map, specifically, If the k-nearest neighbor method is used to calculate the average distance between all label points and adjacent human head label points within the coverage space, and then a distance coefficient is selected based on the average, the fit on other datasets will decrease, and if the crowd density is low and not average, it will introduce uncertainty problems with the k-nearest neighbor method, and further affect the accuracy of head dimension estimation. Therefore, it is preferable to first perform hierarchical partitioning and cluster analysis on the label points within the coverage space, and then calculate the average distance. The head dimensions of each person are inversely proportional to the distance of the image acquisition device, and the label point x iStep 1 involves using a method to represent the position of human heads in an image, dividing the label points in the image into different hierarchies according to distance, assigning different weighted averages to the different hierarchies, and performing cluster analysis on the positions of the label points within each hierarchy to obtain different cluster centers, with the number of cluster centers being equal to the number of hierarchies of the label points. Convert the labels of crowd images into crowd density maps based on different weighted averages.

[0036]

number

[0037] x i The Dirichlet function δ(xx) is used to describe the position of the human head. i This is represented as ), where N represents the number of label points for the human head, and further, step 2 involves performing a head simulation using radial basis functions. The radial basis function diffuses the human head into a single range, and the range value depends on the head dimensions.

[0038]

number

[0039] Here,

[0040]

number

[0041] i is a subscript of the label dot, β i is the weighted average of label points at different hierarchies, n is the number of cluster centers, G σi (x) is a radial basis function, d i - (Here, " - " is, d i The overline (shown) represents the average distance from the label point to the cluster center within a single hierarchy, σ i Step 3 determines the accuracy of the Gaussian convolution, This includes step 4, which involves using a deep learning neural network to train all labeled points and output a density map, and then performing an integral operation on the density map to obtain the number of people.

[0042] During peak pedestrian flow periods, remote control of the overall pedestrian flow through pedestrian flow rules cannot be achieved without analyzing the pedestrian flow within the entire space corresponding to the laboratory's machine room. However, the data collected by the image acquisition device has the problem of high occlusion rates. Furthermore, the obstruction analysis module obtains a new dataset by fusing obstructions and crowd features. The obstruction analysis module analyzes pedestrian flow rules to determine the obstructing factors that affect pedestrian flow within the machine room, and different obstructions are included in different coverage areas. The specific analysis process is as follows: Based on the distribution of crowds within a space, a biomathematical equation of state is established, and activity factors of human flow diffusion are extracted. The dynamic equation for crowd diffusion is:

[0043]

number

[0044] Here, in Step 1, Q represents the flow rate, ρ represents the density of the crowd, and V represents the average velocity. When the average speed is at its maximum,

[0045]

number

[0046]

number

[0047] Here, ρ(x,t) is the density of x locations 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 x j (Depending on the position of t and x,

[0048]

number

[0049]

number

[0050] Here, the range of the R0 value depends on step 2, which depends on the dimensions of the human head at the label point. Step 3 involves analyzing and extracting inhibiting factors based on changes in position vectors in the dynamic system of crowd diffusion. Since the degree of inhibition differs at different locations of the inhibiting factors, and experimental equipment in the laboratory inhibits students' walking, Step 3 involves extracting the object features of the inhibiting factors in the machine room based on the inhibiting factors, and then performing feature extraction on the image to extract the object features of the inhibiting factors in the image. The target features of the crowd adjacent to the obstruction and the features of the obstruction are fused, and calculations are performed based on the position of the pixel points to obtain the x coordinates of the heads of people in the crowd. i Let x i The position of the adjacent pixel point of the obstruction is y i Assuming that different positions correspond to different feature vectors, the feature fusion formula is as follows:

[0051]

number

[0052]

number

[0053] Here, x1 and y2 are feature vectors of the same dimension, y∈Y is the feature mapping set after fusion, and W(x1+y2)∈R n By fusing the inhibitors and neighboring crowds through feature fusion, a new dataset was obtained.

[0054]

number

[0055] w i Step 4 includes the fusion weight, where i is a subscript.

[0056] The data acquisition module includes different image acquisition devices. After numbering the image acquisition devices, the data acquisition module installs the image acquisition devices with different numbers at different locations in the laboratory's machine room. The areas covered by the different image acquisition devices include overlapping and non-overlapping areas. The combined analysis module analyzes the crowd distribution images in the overlapping areas, and the pedestrian flow statistics module analyzes the pedestrian flow distribution rules in the non-overlapping areas.

[0057] The aforementioned pedestrian flow statistics module extracts crowd features within the coverage space of the image acquisition device, then performs data training based on pedestrian flow data and density maps, and obtains the training loss error between the predicted crowd and the actual crowd. The formula for the loss error is as follows:

[0058]

number

[0059] Here, N is the number of images, and F(x;θ) and F represent points in the density map. The density distribution of the crowd differs depending on the coverage space, and the loss error that occurs during the corresponding data analysis also differs.

[0060] In the laboratory, obstacles that hinder human movement exert the same pressure level on adjacent crowds. As the characteristics of the crowd distribution change, the obstructive pressure acting on the crowd at the obstruction location does not change. Therefore, when predicting and analyzing the overall human flow, the degree of influence of the obstacles does not change. The obstruction analysis module performs a fusion analysis on the dataset of adjacent crowds of the obstacles to analyze the obstacles and human flow as a whole. When the prediction analysis module analyzes the entire crowd within the entire covered space, it considers the influence of different parts of the obstacles on human flow, and the obstruction pressure is calculated by the following formula.

[0061]

number

[0062]

number

[0063] Here, <·> represents the average value at the observation center point x, and Var(V) represents the change in the velocity of the crowd.

[0064] The predictive analysis module performs an overall pedestrian flow density analysis based on data collected by all image acquisition devices in the space, and the obstruction analysis module and the pedestrian flow statistics module obtain a complete dataset by analyzing crowd density under different coverage spaces. The predictive analysis module then further predicts crowd density at different times on the dataset and analyzes pedestrian flow paths within the entire space based on the prediction results.

[0065] The coupling analysis module analyzes the transformation of crowd density within different coverage spaces, obtains changes in the number of crowds at different times through the pedestrian flow statistics module, determines the positional changes of the pedestrian flow based on the start and end coverage spaces of the pedestrian flow, and recognizes that coupling relationships exist between obstructions detected in different coverage spaces. The coupling analysis module then analyzes the overlapping and connecting parts between these obstructions.

[0066] Specifically, when used, the present invention includes a data acquisition module, a data storage module, a pedestrian flow statistics module, an obstruction analysis module, a fusion analysis module, and a predictive analysis module. The data acquisition module collects all pedestrian flow data within the laboratory machine room through an image acquisition device and stores the pedestrian flow data in the data storage module. The pedestrian flow statistics module analyzes crowd density in different coverage spaces, determines the average distance through hierarchical partitioning and cluster analysis of labeled points for human heads, and further determines the dimensions of human heads using a Gaussian function. If the crowd density is low and not average, it solves the uncertainty problem using the k-nearest neighbor method. The obstruction analysis module features the characteristics of obstructions within the coverage space and adjacent crowds to obtain a new dataset. The predictive analysis module performs overall pedestrian flow predictive analysis based on the newly fused dataset and pedestrian flow data of crowds not adjacent to obstructions, thereby significantly improving the efficiency of remote pedestrian flow analysis and control, and ensuring that data is not obstructed when performing overall pedestrian flow analysis on data collected by different Internet of Things (IoT) collection devices.

[0067] Although the present invention has been described in more detail above based on specific embodiments, it will be readily apparent to those skilled in the art that the scope of protection of the present invention is clearly not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent modifications or substitutions to the relevant technical features, and all such modified or substituted technical proposals fall within the scope of protection of the present invention.

Claims

1. It includes a data collection module, a data storage module, a human flow statistics module, an inhibition analysis module, a coupling analysis module, and a predictive analysis module. The data acquisition module collects all human flow data within the laboratory's machine room via an image acquisition device, and then stores the human flow data in a data storage module. The image acquisition devices are installed at different locations within the laboratory's machine room. After dividing the machine room into different coverage spaces, the pedestrian flow statistics module performs crowd density analysis on the pedestrian flow data within each coverage space, using the labeled points of people's heads as the core analysis target. This analysis combines hierarchical clustering results and a Gaussian function to obtain the corresponding loss function, crowd density, and pedestrian flow rate. The loss function represents the error between the model analysis results from the pedestrian flow statistics model and the actual crowd figures. The pedestrian flow statistics model is a deep learning neural network model that outputs crowd density and pedestrian flow rate. The model analysis results are the density map and population statistics results that the model outputs for the input pedestrian flow data. The combined analysis module combines the analysis results of the human flow statistics module to analyze changes in crowd location, obtains the degree of crowd coupling in different coverage spaces, and calculates the exchange rate of crowd numbers within adjacent coverage spaces. The aforementioned inhibition analysis module analyzes the loss function of the statistical analysis process and the effect of experimental equipment on changes in the direction of human flow to obtain inhibition factors. Then, it combines the degree of coupling obtained from the coupling analysis module to perform a positional analysis on the human flow inhibition factors in the machine room of the laboratory. The aforementioned inhibition analysis module, based on the analysis results of the inhibitory factors, features crowd characteristics in the laboratory and close to the inhibition location, and by fusing the features of the inhibition location and the crowd location, a new dataset is obtained. The aforementioned predictive analysis module performs crowd density analysis based on a new dataset formed by fusing the analysis results of the inhibition analysis module, obtains a crowd density map for the entire space, then performs crowd density prediction based on the density map to obtain prediction results, evaluates the safety of the human flow density in the space through the prediction results, and formulates management measures for experimental equipment. This is a human flow analysis system using Internet of Things (IoT) information collection.

2. The image acquisition device collects pedestrian flow data within the machine room from different angles, and the pedestrian flow statistics module uses human head features to detect and count the crowd through the generated density map, specifically, Each person's head size is inversely proportional to the distance from the image acquisition device, and the labeled point x i Step 1 involves using a method to represent the position of human heads in the image, arranging the distance between the human head corresponding to the label point and the image acquisition device from closest to furthest, dividing the label points in the image into multiple layers, setting a higher weighted average for layers with closer distances, assigning different weighted averages to different layers, and performing cluster analysis on the positions of label points within different layers to obtain different cluster centers, with the number of cluster centers being equal to the number of layers of the label points. Convert the labels of crowd images into crowd density maps based on different weighted averages. [Number 24] x i The Dirichlet function δ(xx) is used to describe the position of the human head. i ) is represented as, N represents the number of label points for the human head, and step 2 involves performing a head simulation using radial basis functions. The radial basis function diffuses the human head into a single range, and the range value depends on the head dimensions. [Number 25] Here, [Number 26] i is a subscript character of label points, β i is the weighted average of label points at different hierarchies, n is the number of cluster centers, G σi (x) is a radial basis function, d i - (wherein, " - " represents the overline of d i ), step 3 of representing the average value of distances from label points to cluster centers within one hierarchy, The human flow analysis system for collecting information from the Internet of Things (IoT) according to claim 1, characterized by comprising step 4, which involves training all label points using a deep learning neural network to output a density map, and then performing an integral operation on the density map to obtain the number of people.

3. The aforementioned inhibition analysis module analyzes the rules of human flow movement to determine the inhibiting factors that affect human flow movement within the machine room, and different inhibiting factors are included in different coverage areas. The specific analysis process is as follows: Based on the distribution of crowds within a space, a biomathematical equation of state is established, and activity factors of human flow diffusion are extracted. This equation of state is a dynamic equation of crowd diffusion that describes the state of crowd diffusion. [Number 27] Here, Q is the flow rate, ρ is the crowd density, V is the average velocity, and the activity factors are time-varying dynamic parameters corresponding to the flow rate Q, density ρ, and average velocity V in the formula. Step 1 involves collecting crowd distribution data at different time points t, substituting them into the formula, and performing calculations. Based on the dynamic equation of crowd dispersal obtained in Step 1, when the average velocity is maximum, [Number 28] At this point, the crowd density reaches its peak value, and the formula for calculating the corresponding crowd density is as follows: [Number 29] Here, ρ(x,t) is the density of x locations 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 x j Step 2 depends on the position of (t) and x, Step 3 involves analyzing and extracting inhibiting factors based on changes in position vectors in the dynamic system of crowd diffusion, then performing feature extraction on the image to extract object features of the inhibiting object in the image, and determining the position of the inhibiting object in the machine room based on the inhibiting factors. The target features of the crowd adjacent to the obstruction and the features of the obstruction are fused, and calculations are performed based on the position of the pixel points to obtain the x coordinates of the heads of people in the crowd. i Let x i The position of the adjacent pixel point of the obstruction is y i Assuming that different locations correspond to different feature vectors, the feature fusion formula is as follows: [Number 30] [Number 31] Here, x １ and y ２ is a feature vector of the same dimension, y∈Y is the feature mapping set after fusion, W(x 1 +y 2 )∈R n The human flow analysis system using Internet of Things (IoT) information collection according to claim 1, characterized by comprising step 4, which is a new dataset obtained by fusing obstructors and neighboring crowds through feature fusion.

4. 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 of different numbers at different locations in the laboratory's machine room, the covered areas of the different image acquisition devices include overlapping and non-overlapping areas, the combined analysis module analyzes the crowd distribution image of the overlapping area, and the pedestrian flow statistics module analyzes the pedestrian flow distribution rules of the non-overlapping area, characterized in that the pedestrian flow analysis system using Internet of Things (IoT) information acquisition is as described in claim 1.

5. The aforementioned pedestrian flow statistics module extracts crowd features within the coverage space of the image acquisition device, then performs data training based on pedestrian flow data and density maps, obtaining a training loss error between the predicted crowd and the actual crowd. This training loss error is identical to the loss function described in claim 1 and is used to measure the output accuracy of the pedestrian flow statistics model. The formula for the loss error is as follows: [Number 32] The human flow analysis system using Internet of Things (IoT) information collection according to claim 1, characterized in that, here, N is the number of images, and F(x;θ) and F represent points in the density map.

6. The human flow analysis system using Internet of Things (IoT) information collection according to claim 3, characterized in that, in a laboratory setting, obstacles that hinder human movement exert the same pressure level on adjacent crowds, that is, the numerical value of the obstruction pressure caused by the same obstacle on all adjacent crowds is the same and does not change with changes in the position of the crowd, the obstruction analysis module analyzes the obstacle and human flow as a whole by performing fusion analysis on the dataset of adjacent crowds of the obstacle, and the predictive analysis module considers the effect of different parts of the obstacle on human flow when analyzing the entire crowd within the entire covered space.

7. The predictive analysis module performs an overall pedestrian flow density analysis based on data collected by all image acquisition devices in the space, the obstruction analysis module and the pedestrian flow statistics module obtain a complete dataset by analyzing crowd density under different coverage spaces, the predictive analysis module further performs crowd density predictions for different times on the dataset, and analyzes pedestrian flow paths within the entire space based on the prediction results, characterized in that the pedestrian flow analysis system using Internet of Things (IoT) information collection according to claim 1.

8. The combined analysis module analyzes the transformation of crowd density within different coverage spaces, obtains changes in the number of crowds at different times through the pedestrian flow statistics module, and then determines the change in the position of the pedestrian flow based on the start and end coverage spaces of the pedestrian flow, as described in claim 1, which is a pedestrian flow analysis system using Internet of Things (IoT) information collection.

Images