Visitor flow dynamic visualization interactive display interface generation method and system

Abstract

The invention discloses a human traffic dynamic visualization interactive display interface generation method and system, relates to the technical field of data visualization, and is used for solving the problem that macroscopic situations and microscopic details cannot be considered in large-scale human traffic monitoring. Firstly, trajectory vector features are analyzed, an interaction entropy field reflecting the disorder degree of microcosmic motion is constructed, and global continuous data are generated; then, extracting features by using Gaussian difference convolution, and constructing a visual saliency weight matrix; probability sampling and weighted Thiessen polygon subdivision are executed based on the weight, and a space geometric constraint type self-adaptive viewport topology grid is generated; and finally, the grids are used as projection containers to execute differential rendering, microscopic trajectory dynamic reconstruction at a small viewport and pressure color gradation filling generation at a large viewport, so that fusion interactive display of multi-granularity data is realized, and the decision efficiency of public safety monitoring is improved.

Description

Technical Field

[0001] This invention relates to the field of data visualization technology, specifically to a method and system for generating a dynamic visualization interactive display interface for human traffic. Background Technology

[0002] With the acceleration of urbanization, the density of people flowing through large public places such as transportation hubs, commercial complexes, and sports venues has increased significantly. In these complex scenarios, real-time monitoring and analysis of crowd flow is crucial for ensuring public safety, optimizing spatial layout, and improving emergency response efficiency. Managers need to quickly perceive the situation on-site through visual interfaces, extracting potential congestion risks and abnormal behavior patterns from massive amounts of trajectory data, thereby achieving a shift from passive data monitoring to proactive decision-making.

[0003] However, existing methods for visualizing pedestrian flow typically employ fixed-resolution grid heatmaps or globally uniform vector flow field rendering. This uniform rendering approach has significant drawbacks when dealing with non-uniformly distributed pedestrian flow data: firstly, in high-density areas, fixed-resolution grids cause pixel aliasing and visual occlusion of microscopic individual movement details, making it difficult for managers to distinguish whether crowds are queuing in an orderly fashion or pushing and shoving; secondly, in low-density, open areas, globally uniform high-precision rendering consumes a large amount of graphics processor computing resources, leading to a decrease in system frame rate and rendering latency. Furthermore, existing interface generation logic is often statically preset and cannot dynamically adjust screen space allocation strategies based on the real-time data's complexity, causing critical risk information to be easily buried in a large amount of redundant background data, failing to meet the precise interaction needs of complex dynamic scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for generating a dynamic, visual, interactive display interface for pedestrian flow, thus solving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for generating a dynamic and visually interactive display interface for pedestrian flow, comprising the following steps: S1. Vectorizing and parsing the collected time-series data of pedestrian trajectories, extracting trajectory segments through a sliding time window, calculating the angular dispersion and velocity magnitude attenuation rate between adjacent trajectory vectors, constructing a local interaction entropy value reflecting the degree of disorder in the micro-motion of the group, and performing inverse distance weighted interpolation to map the discrete local interaction entropy value to physical space grid cells, generating a global pedestrian flow interaction entropy field data; S2. Performing multi-scale Gaussian difference convolution operation on the global pedestrian flow interaction entropy field data, extracting the spatial extreme points and edge gradient features of the local entropy value, and constructing a visual saliency weight distribution matrix corresponding to the screen pixel coordinate system through normalization mapping; S3. Using the visual saliency weight distribution matrix as a probability density function, performing... Monte Carlo importance sampling generates a non-uniform seed point set with density varying with weights. Based on this seed point set, weighted Thiessen polygon subdivision is performed to generate a spatially geometrically constrained adaptive viewport topology mesh. High-weight regions generate small-area high-frequency sampling viewports due to the high-density distribution of seed points, while low-weight regions generate large-area low-frequency sampling viewports due to the low-density distribution of seed points. S4. Using the spatially geometrically constrained adaptive viewport topology mesh as a data projection container, the global pedestrian flow interaction entropy field data is resampled and mapped to each viewport unit: within small-area high-frequency sampling viewports, the vector characteristics of the entropy field data are preserved, and flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics; within large-area low-frequency sampling viewports, scalar homogenization of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the region, combining these layers to generate a multi-granularity data fusion-based dynamic visualization and interactive display interface for pedestrian flow.

[0006] Furthermore, the collected personnel trajectory time-series data is vectorized and analyzed. Trajectory segments are extracted through a sliding time window, and the angular dispersion and velocity magnitude decay rate between adjacent trajectory vectors are calculated to construct a local interaction entropy value reflecting the disorder of the group's micro-motions. The specific process is as follows: Trajectory segments within the sliding time window are traversed, and the instantaneous velocity vector of the trajectory segment at the current moment and the instantaneous velocity vector at the previous moment are extracted. Vector normalization is performed on the instantaneous velocity vector and the instantaneous velocity vector at the previous moment. The dot product of the two normalized vectors is calculated to obtain the cosine value of the direction angle. The variance of the cosine value of the direction angle within the sliding time window is used as the angular dispersion. At the same time, the first-order difference sequence of the velocity magnitude of the trajectory segment within the sliding time window is calculated, and the cumulative amplitude of the negative elements in the first-order difference sequence is used as the velocity magnitude decay rate. The angular dispersion and the velocity magnitude decay rate are weighted and summed to generate a disorder index for a single trajectory. The arithmetic mean of the disorder index of the trajectory within a unit physical space grid is calculated, and the arithmetic mean is used as the local interaction entropy value at the center of the grid.

[0007] Furthermore, the specific process of performing inverse distance weighted interpolation to map discrete local interaction entropy values ​​to physical space grid cells and generate global pedestrian flow interaction entropy field data is as follows: Establish a coordinate index matrix for the physical space grid cells, determine the geometric center coordinates of each grid cell to be interpolated, set a search radius with the geometric center coordinates as the center, and retrieve reference grid points within the search radius that already contain local interaction entropy values; calculate the Euclidean distance between the geometric center coordinates of the grid cell to be interpolated and the coordinates of the reference grid points, construct a weight decay function based on the reciprocal of the Euclidean distance, and use the weight decay function to calculate the interpolation weight coefficient for each reference grid point; perform a weighted linear combination of the local interaction entropy values ​​of the reference grid points within the search radius and the corresponding interpolation weight coefficients, assign the calculation result to the grid cell to be interpolated, and after completing the traversal calculation of the grid cells, output the global pedestrian flow interaction entropy field data.

[0008] Further, step S2 includes the following steps: performing multi-level Gaussian blurring on the global pedestrian flow interaction entropy field data, generating a first Gaussian blurred image and a second Gaussian blurred image using Gaussian kernel functions of different scales, performing pixel-level difference operations on the first Gaussian blurred image and the second Gaussian blurred image to generate a Gaussian difference response map; traversing the pixels of the Gaussian difference response map, comparing the response values ​​of the current pixel with its eight neighboring pixels, and marking points whose response values ​​are simultaneously greater than the response values ​​of the eight neighboring pixels as spatial extrema points; calculating the horizontal gradient component and vertical gradient component of each pixel in the Gaussian difference response map, synthesizing a gradient magnitude matrix, spatially superimposing the position information of the spatial extrema points with the gradient magnitude matrix, and mapping the superimposed result to a numerical range of 0 to 1 through linear normalization to generate a visual saliency weight distribution matrix.

[0009] Furthermore, the specific process of generating a non-uniform seed point set with density varying with weight by performing Monte Carlo importance sampling within the visible region, using the visual saliency weight distribution matrix as a probability density function, is as follows: Accumulate the visual saliency weight distribution matrix across the entire domain, use the accumulated result as a normalization factor, perform division normalization on the weight values ​​of the elements within the matrix, and construct a two-dimensional cumulative probability distribution function; generate a set of pseudo-random numbers following a uniform distribution, use each random value in the pseudo-random number sequence as an index key, perform a reverse lookup operation in the two-dimensional cumulative probability distribution function to locate the corresponding matrix coordinate position; extract the matrix coordinate positions locked by the reverse lookup operation, remove duplicate coordinates and boundary overflow coordinates, and record the remaining set of coordinate positions as a non-uniform seed point set with density varying with weight.

[0010] Furthermore, the specific process of generating a spatially geometrically constrained adaptive viewport topology mesh by performing weighted Thiessen polygon subdivision based on the seed point set is as follows: The coordinate position of each seed point in the non-uniform seed point set is read, and the weight value corresponding to this coordinate position in the visual saliency weight distribution matrix is ​​read as the growth potential parameter of the seed point; the screen pixels within the visible area are traversed, the Euclidean spatial proximity between the screen pixels and the seed points is quantified, and the corresponding growth potential parameter is introduced to perform potential field intensity modulation and bias correction operations on the Euclidean spatial proximity, constructing a power distance metric that characterizes the expansion competitiveness of the viewport region; the power distance metric of the seed point to the current screen pixel is compared, and the current screen pixel is classified into the index region to which the seed point with the smallest power distance metric belongs. After completing the regional classification of all screen pixels, the pixel boundary lines between different index regions are extracted, and a spatially geometrically constrained adaptive viewport topology mesh enclosed by polygonal boundaries is constructed.

[0011] Furthermore, using a spatially geometrically constrained adaptive viewport topology grid as the data projection container, the global human flow interaction entropy field data is resampled and mapped to each viewport cell. Within a small area of ​​high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and the specific process of performing flow field line integral convolution rendering to reconstruct the micro-trajectory dynamics is as follows: A Gaussian white noise texture image with the same size as the small area of ​​high-frequency sampling viewport is generated; each pixel within the high-frequency sampling viewport is traversed, and bidirectional streamline integration is performed starting from the current pixel, guided by the vector direction in the global human flow interaction entropy field data, to obtain the streamline trajectory coordinate sequence passing through the pixel; the gray value corresponding to the streamline trajectory coordinate sequence in the white noise texture image is read, the convolution weighted average of the corresponding gray value is calculated, and the convolution weighted average is assigned to the current pixel to generate a micro-trajectory dynamic rendering layer with fluid texture features.

[0012] Furthermore, within a large-area low-frequency sampling viewport, scalar homogenization of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the region. The specific process for combining and generating a multi-granularity data fusion dynamic visualization interactive display interface for pedestrian flow is as follows: The scalar amplitude of the global pedestrian flow interaction entropy field data within the physical space grid unit covered by the large-area low-frequency sampling viewport is statistically analyzed, and the arithmetic mean of the scalar amplitudes is calculated as the unified pressure value of the viewport unit. The unified pressure value is input into a preset color lookup table for linear interpolation mapping to obtain the corresponding RGBA color channel parameters. The RGBA color channel parameters are used to fill the low-frequency sampling viewport, generating a color-gradient fill layer. An image synthesis buffer is established, using the color-gradient fill layer as the background layer and the micro-trajectory dynamic rendering layer as the foreground layer. The region ID of the spatial geometric constraint adaptive viewport topology grid is read as the mask channel, and the foreground layer is superimposed onto the background layer using an Alpha blending algorithm to output the multi-granularity data fusion dynamic visualization interactive display interface for pedestrian flow.

[0013] The interactive display interface generation system for dynamic visualization of pedestrian flow includes the following modules: an entropy field construction module, used to vectorize and parse the collected time-series data of pedestrian trajectories, extract trajectory segments through a sliding time window, calculate the angular dispersion and velocity magnitude attenuation rate between adjacent trajectory vectors, construct a local interactive entropy value reflecting the disorder of the micro-motions of the group, and perform inverse distance weighted interpolation to map the discrete local interactive entropy value to physical space grid cells, generating global pedestrian flow interactive entropy field data; a saliency weight calculation module, used to perform multi-scale Gaussian difference convolution operation on the global pedestrian flow interactive entropy field data, extract the spatial extreme points and edge gradient features of the local entropy values, and construct a visual saliency weight distribution matrix corresponding to the screen pixel coordinate system through normalization mapping; and a viewport topology generation module, used to use the visual saliency weight distribution matrix as a probability density function to perform Monte Carlo recalculation within the visible area. Priority sampling generates a non-uniform seed point set with density varying with weights. Based on this seed point set, weighted Thiessen polygon subdivision is performed to generate a spatially geometrically constrained adaptive viewport topology mesh. High-weight regions generate small-area high-frequency sampling viewports due to the high-density distribution of seed points, while low-weight regions generate large-area low-frequency sampling viewports due to the low-density distribution of seed points. The interface rendering generation module uses the spatially geometrically constrained adaptive viewport topology mesh as a data projection container to resample and map the global pedestrian flow interaction entropy field data to each viewport unit. Within the small-area high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics. Within the large-area low-frequency sampling viewport, scalar homogenization processing of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the region. These layers are combined to generate a multi-granularity data fusion-based dynamic visualization and interactive display interface for pedestrian flow.

[0014] The present invention has the following beneficial effects: (1) The interactive display interface generation method for dynamic visualization of pedestrian flow can accurately identify and highlight abnormal risk areas in group movement, solving the technical problem that traditional density heat maps cannot reflect the degree of micro-movement disorder. By performing vectorization analysis and sliding window calculation on the collected trajectory data, a local interactive entropy value containing the angular dispersion and velocity modulus attenuation rate is constructed, upgrading simple location information into physical field data that reflects the degree of disorder in group movement. Furthermore, multi-scale Gaussian difference convolution operation is used to extract the spatial extreme points and edge features of the entropy field, and the visual saliency weight is automatically calculated. This process enables the system to capture the congestion risk caused by pedestrian collision, stagnation or turbulence from the underlying logic through the surface density data, and automatically quantify the information carrying priority of different areas, providing a precise data-driven basis for subsequent interface generation, and ensuring that the monitoring focus is always locked on the area with the most potential risk.

[0015] (2) A dynamic visualization interactive display interface generation system for pedestrian flow realizes adaptive multi-granular reconstruction of the display interface, solving the contradiction between information overload and low rendering efficiency in large-scale scenes. Utilizing Monte Carlo importance sampling and weighted Thiessen polygon subdivision algorithms, a spatial geometric constraint mesh is dynamically generated based on visual saliency weights. This mechanism automatically generates high-density subdivided viewports in high-risk areas and low-density aggregated viewports in low-risk areas. Based on this, the method employs differentiated rendering strategies for different viewports: in subdivided viewports, microscopic trajectory dynamics are reconstructed through flow field line integral convolution rendering, intuitively presenting the direction and conflict of pedestrian flow; in aggregated viewports, a color-gradient fill layer is generated through scalar homogenization processing, providing a macroscopic situational background. This ensures the extreme clarity of microscopic details in key areas while significantly reducing visual interference and computational overhead in non-critical areas, significantly improving the efficiency of human-computer interaction and the information transmission effectiveness of the interface.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for generating a dynamic, visual, and interactive display interface for pedestrian flow according to the present invention.

[0018] Figure 2 This is a flowchart of the system for generating a dynamic, visual, and interactive display interface for pedestrian flow, as described in this invention. Detailed Implementation

[0019] This application's embodiments solve the technical problems of existing pedestrian flow monitoring interfaces, which struggle to simultaneously assess macroscopic situational awareness and display microscopic risk details, and are unable to distinguish between orderly high-density and disorderly congestion states, through a method and system for generating a dynamic, visual, and interactive display interface for pedestrian flow.

[0020] The overall concept of the solution in this application embodiment is as follows: First, the system starts with the underlying trajectory data, constructing an interactive entropy field that reflects the degree of microscopic motion disorder by calculating the angle change and velocity decay between adjacent trajectory vectors, thus transforming the flow of people in the physical world into energy field data that can be processed by a computer. Next, it uses the difference of Gaussians technique in image processing to identify feature regions in this entropy field and generate a visual saliency weight matrix to guide the interface layout. Subsequently, based on this weight matrix, it dynamically constructs a viewport topology grid with varying density through probability sampling and geometric subdivision algorithms, which serves as the skeleton of the interface. Finally, it automatically matches the rendering strategy according to the size attribute of the grid, performing complex vector flow field rendering in small grids to display details, and performing simplified scalar color filling in large grids to display an overview, ultimately synthesizing a dynamic interactive interface that has both a macroscopic global view and microscopic local precision.

[0021] Please see Figure 1This invention provides a technical solution: a method for generating a dynamic and interactive display interface for pedestrian flow, comprising the following steps: S1. Vectorizing and parsing the collected time-series data of pedestrian trajectories, extracting trajectory segments through a sliding time window, calculating the angular dispersion and velocity magnitude attenuation rate between adjacent trajectory vectors, constructing a local interaction entropy value reflecting the degree of disorder in the micro-motion of the group, and performing inverse distance weighted interpolation to map the discrete local interaction entropy value to physical space grid cells to generate a global pedestrian flow interaction entropy field; S2. Performing multi-scale Gaussian difference convolution operation on the global pedestrian flow interaction entropy field data, extracting the spatial extreme points and edge gradient features of the local entropy value, and constructing a visual saliency weight distribution matrix corresponding to the screen pixel coordinate system through normalization mapping; S3. Using the visual saliency weight distribution matrix as a probability density function, performing Monte Carlo mapping within the visible area. Importance sampling generates a non-uniform seed point set with density varying with weight. Based on this seed point set, weighted Thiessen polygon subdivision is performed to generate a spatially geometrically constrained adaptive viewport topology mesh. High-weight regions generate small-area high-frequency sampling viewports due to the high-density distribution of seed points, while low-weight regions generate large-area low-frequency sampling viewports due to the low-density distribution of seed points. S4. Using the spatially geometrically constrained adaptive viewport topology mesh as a data projection container, the global pedestrian flow interaction entropy field data is resampled and mapped to each viewport unit: within the small-area high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics; within the large-area low-frequency sampling viewport, scalar homogenization processing of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the region, which is then combined to generate a multi-granularity data fusion-based dynamic visualization interactive display interface for pedestrian flow.

[0022] In this implementation scheme, step S1 is mainly used for data preprocessing and physical field construction. This step first uses a sliding time window mechanism to segment continuously collected personnel trajectory data to capture short-term motion change characteristics. In this process, local interaction entropy is a core concept. It is not entropy in traditional thermodynamics, but rather a quantitative index synthesized by calculating the angular dispersion between adjacent trajectory vectors (i.e., inconsistency in motion direction) and the velocity modulus decay rate (i.e., the degree of sudden drop in motion speed). It is specifically used to characterize the disorder and congestion trends of micro-movements in a group. Subsequently, using inverse distance weighted interpolation, the values ​​of blank areas are estimated based on the values ​​of known data points, expanding and filling the discrete trajectory point data into continuous grid data covering the entire physical space, thereby generating global human flow interaction entropy field data. The technical role of this step is to transform discrete, isolated individual location data into a continuous energy field that can describe the overall congestion pressure and disorder level of the area, providing a unified data basis for subsequent global analysis. Step S2 is mainly used for feature extraction and quantitative allocation of visual attention. This step performs multi-scale Gaussian difference convolution on the generated global entropy field data. This is an image processing technique that simulates a bandpass filter by subtracting two images with different degrees of Gaussian blur, effectively suppressing background noise and enhancing local features, thereby accurately extracting spatial extrema (such as congestion centers) and edge gradient features (such as crowd diffusion edges) in the entropy field. Based on this, a visual saliency weight distribution matrix is ​​constructed, a numerical matrix corresponding one-to-one with screen pixels. Its value directly reflects the information importance or visual attention priority of the area at the current moment. The technical role of this step is to automatically identify truly noteworthy risk areas in the monitoring screen through algorithms, transforming objective data features into rendering priority parameters in computer graphics. Step S3 is mainly used for dynamic reconstruction of the interface layout and mesh generation. This step uses Monte Carlo importance sampling technology, treating the above weight matrix as a probability density function. Dense random seed points are generated with high probability in areas with high weight, and sparse seed points are generated with low probability in areas with low weight. Subsequently, weighted Thiessen polygon subdivision is performed based on these non-uniformly distributed seed points. This algorithm divides the space into several polygonal regions, minimizing the weighted distance from each point within a region to its corresponding seed point. The resulting spatially geometrically constrained adaptive viewport topology mesh exhibits significant structural characteristics: high-weight regions automatically form small, densely packed, high-frequency sampling viewports, while low-weight regions automatically form large, sparsely packed, low-frequency sampling viewports. This step establishes a non-uniform, data-driven interface skeleton, laying the topological foundation for hierarchical rendering with varying precision. Step S4 is primarily used for differentiated rendering and final interface compositing. This step uses the mesh generated in the previous step as a container, determining the data presentation method based on the mesh's area attribute.For small-area, high-frequency sampling viewports, the system identifies them as high-risk or high-interest areas. Therefore, it preserves the vector characteristics of the entropy field data and employs flow field line integral convolution rendering technology. This method generates streamline textures by performing integral convolution on noise textures along the vector field direction, enabling extremely detailed reconstruction of the flow direction and dynamic trends of microscopic trajectories. For large-area, low-frequency sampling viewports, the system identifies them as background or low-risk areas and performs scalar homogenization processing. This involves calculating the average value within the area and mapping it to a single color, generating a color-gradient fill layer reflecting the average pressure value of the area. Finally, the two are combined. This step resolves the contradiction that a single visualization method cannot simultaneously capture both macroscopic trends and microscopic details, achieving the fusion display of multi-granular data. It highlights the dynamic details of key areas while preserving the overall macroscopic background.

[0023] Specifically, the process of vectorizing and parsing the collected personnel trajectory time-series data, extracting trajectory segments through a sliding time window, calculating the angular dispersion and velocity magnitude decay rate between adjacent trajectory vectors, and constructing a local interaction entropy value reflecting the disorder of the group's micro-motions is as follows: Trajectory segments within the sliding time window are traversed, and the instantaneous velocity vectors of the current and previous moments are extracted. Vector normalization is performed on these vectors, and the dot product of the two normalized vectors is calculated to obtain the cosine of the directional angle. The variance of the cosine of the directional angle within the sliding time window is used as the angular dispersion. Simultaneously, the first-order difference sequence of the velocity magnitude of the trajectory segment within the sliding time window is calculated, and the proportion of the cumulative amplitude of negative elements in the first-order difference sequence to the total sequence amplitude is used as the velocity magnitude decay rate. The angular dispersion and velocity magnitude decay rate are weighted and summed to generate a disorder index for a single trajectory. The arithmetic mean of the disorder index of the trajectories within a unit physical space grid is calculated, and the arithmetic mean is used as the local interaction entropy value at the grid center.

[0024] In this implementation plan, specifically, the collected time-series data of personnel trajectories is first preprocessed. To eliminate the influence of differences in individual movement speeds on direction determination, the system extracts the instantaneous velocity vectors of the trajectory segments at the current and previous moments and normalizes them to ensure uniform magnitude of all vectors, thus focusing on the extraction of directional features. Subsequently, the continuity of the movement direction is quantified by calculating the dot product of the normalized vectors, and the fluctuation of this value within the time window is further statistically analyzed, serving as a key indicator for measuring the stability of the trajectory direction. The formula for calculating the variance of the cosine of the direction angle is as follows: ;in, : The dispersion of the directional angle of the k-th trajectory within the current time window; The total number of time steps contained in the sliding time window; : The cosine of the directional angle of the k-th trajectory at time t; The arithmetic mean of the cosine values ​​of the directional angles within the time window. This formula intuitively reflects the turning frequency of an individual within a short period of time through statistical variance; the larger the variance, the more chaotic the movement direction. Simultaneously, to capture the "sudden stop" or "stagnation" characteristics caused by congestion, the system analyzes the first-order difference of the velocity modulus, focusing on the proportion of velocity decrease (i.e., the difference is negative). The formula for calculating the velocity modulus attenuation rate is as follows: ;in, : Velocity magnitude attenuation rate of the k-th trajectory; : The first-order difference of the velocity modulus at time t; This step prevents extremely small positive numbers with a denominator of zero. By quantifying the proportion of deceleration magnitude in the total velocity change, it effectively identifies hindered behavior. Finally, the two indicators are merged and mapped onto a spatial grid. The formula for calculating the local interaction entropy value at the grid center is as follows: ;in, : Local interaction entropy value of the target mesh cell; The number of trajectories falling within this grid cell; , : These are the weighting coefficients for the angular dispersion and the velocity modulus attenuation rate, respectively. These coefficients are pre-determined using the Analytic Hierarchy Process (AHP) based on the congestion type of the scenario. For example, the weight of velocity attenuation can be appropriately increased in exit scenarios.

[0025] Specifically, the process of performing inverse distance weighted interpolation to map discrete local interaction entropy values ​​to physical space grid cells and generate global pedestrian flow interaction entropy field data is as follows: Establish a coordinate index matrix for the physical space grid cells, determine the geometric center coordinates of each grid cell to be interpolated, set a search radius with the geometric center coordinates as the center, and retrieve reference grid points within the search radius that already contain local interaction entropy values; calculate the Euclidean distance between the geometric center coordinates of the grid cell to be interpolated and the coordinates of the reference grid points, construct a weight decay function based on the reciprocal of the Euclidean distance, and use the weight decay function to calculate the interpolation weight coefficient for each reference grid point; perform a weighted linear combination of the local interaction entropy values ​​of the reference grid points within the search radius and the corresponding interpolation weight coefficients, assign the calculation result to the grid cell to be interpolated, and after completing the traversal calculation of the grid cells, output the global pedestrian flow interaction entropy field data.

[0026] In this implementation scheme, specifically, in the process of generating global pedestrian flow interaction entropy field data, to address the problem of sparse and discontinuous grid data, a distance-based spatial interpolation algorithm is used to extend discrete entropy values ​​into a continuous field. First, a spatial index is established, and for each blank grid cell requiring interpolation, its search neighborhood is determined. The system does not simply average the surrounding points, but follows the first law of geography, namely, the closer the reference point, the greater its influence on the target point. Therefore, a weighting function based on the reciprocal of Euclidean distance is constructed. The formula for calculating the interpolation weighting coefficients is as follows: ;in, : The interpolation weight coefficient of reference grid point j for the grid cell i to be interpolated; : The x and y coordinates of the geometric center of the mesh element i to be interpolated; : The geometric center coordinates of the j-th valid reference grid point within the search radius; : Power parameter, usually with a value of 2, used to control the rate at which the weight decays with distance; A smoothing factor is used to prevent numerical overflow when the distance approaches zero. After obtaining the weight of each reference point, the final entropy value of the target point is calculated through a weighted linear combination. The formula for generating the global pedestrian flow interaction entropy field data is as follows: ;in, : The final global interactive entropy field value generated by the interpolated grid cell i; The total number of valid reference grid points within the search radius; : The known local interaction entropy value of the j-th reference grid point. Through the above steps, the system can smoothly fill in data gaps and generate an energy field with continuous gradient changes, ensuring the consistency of visual effects and the integrity of physical meaning in subsequent rendering processes.

[0027] Specifically, step S2 includes the following steps: performing multi-level Gaussian blurring on the global pedestrian flow interaction entropy field data; generating a first Gaussian blurred image and a second Gaussian blurred image using Gaussian kernel functions of different scales; performing pixel-level difference operations on the first Gaussian blurred image and the second Gaussian blurred image to generate a Gaussian difference response map; traversing the pixels of the Gaussian difference response map, comparing the response values ​​of the current pixel with its eight neighboring pixels, and marking points whose response values ​​are simultaneously greater than the response values ​​of the eight neighboring pixels as spatial extrema points; calculating the horizontal gradient component and vertical gradient component of each pixel in the Gaussian difference response map, synthesizing a gradient magnitude matrix, spatially superimposing the position information of the spatial extrema points with the gradient magnitude matrix, and mapping the superimposed result to a numerical range of 0 to 1 through linear normalization to generate a visual saliency weight distribution matrix.

[0028] In this implementation scheme, specifically, visual features are first extracted from the entropy field data of the entire pedestrian flow interaction. To simulate the sensitivity of the human eye to spatial features at different scales, the system treats the entropy field data as a grayscale image and extracts significant local features using a multi-scale Gaussian difference algorithm. This process is equivalent to constructing a bandpass filter, which can effectively remove high-frequency noise and suppress low-frequency background, thereby highlighting congestion patches with mid-frequency features. The formula for generating the Gaussian difference response map is as follows: ;in, The response value at coordinate position (x, y) in the difference of Gaussian response plot; The value of the entropy field data of the entire human flow interaction at the corresponding location; : Two-dimensional convolution operator; Standard deviation is The two-dimensional Gaussian kernel function; , Two Gaussian kernel standard deviation parameters at different scales are used. Then, the generated Gaussian difference response map is traversed, and non-maximum suppression is performed. The system compares the response values ​​of the current pixel with those of its eight neighboring pixels. If the current pixel's value is simultaneously and strictly greater than the values ​​of its eight neighboring pixels, it is marked as a spatial extremum. This step aims to accurately locate the geometric center of congested or chaotic areas, i.e., the focal point of visual attention. Next, to capture the edge contour features of crowd distribution, the system calculates the gradient magnitude of each point in the response map. The gradient reflects the drastic change in data and can effectively delineate the diffusion boundary of congested areas. The formula for calculating the gradient magnitude matrix is ​​as follows: ;in, Coordinates The magnitude of the synthesized gradient at that point; The response values ​​of the two adjacent points in the horizontal direction are used to calculate the horizontal gradient component. The response values ​​of the two adjacent points in the vertical direction are used to calculate the vertical gradient component. Finally, the extracted spatial extremum information and gradient magnitude information are fused to construct the final visual saliency weight distribution matrix. This process combines discrete attention points with continuous edge lines to form a complete visual attention map, and linear normalization ensures that the values ​​fall within the range of 0 to 1 to adapt to subsequent probability sampling algorithms. The formula for calculating the normalized map is as follows: ;in, Coordinates The final visual saliency weight value generated; : Spatial extreme point indicator function, which takes the value 1 if the point is marked as an extreme point, and takes the value 0 otherwise; : These are the fusion weighting coefficients for extreme point intensity and gradient magnitude, set using expert experience. Extreme points are usually given higher weights to highlight the central region. The minimum and maximum values ​​in the merged original data matrix.

[0029] Specifically, the process of generating a non-uniform seed point set with varying density as weights by performing Monte Carlo importance sampling within the visible region, using the visual saliency weight distribution matrix as a probability density function, is as follows: Accumulate the visual saliency weight distribution matrix across the entire domain, using the accumulated result as a normalization factor, and perform division normalization on the weight values ​​of the elements within the matrix to construct a two-dimensional cumulative probability distribution function; generate a set of pseudo-random numbers following a uniform distribution, using each random value in the pseudo-random number sequence as an index key, and perform a reverse lookup operation in the two-dimensional cumulative probability distribution function to locate the corresponding matrix coordinate position; extract the matrix coordinate positions locked by the reverse lookup operation, remove duplicate coordinates and boundary overflow coordinates, and record the remaining set of coordinate positions as a non-uniform seed point set with varying density as weights.

[0030] In this implementation scheme, during the generation of the non-uniform seed point set, the discrete weight matrix first needs to be transformed into a continuous probability model to facilitate random sampling by the computer. The system performs global numerical superposition of the visual saliency weight distribution matrix to obtain the total energy value of the matrix, and uses this as a basis to normalize the weight of each pixel, converting it into an occurrence probability. The formula for constructing the two-dimensional cumulative probability distribution function is as follows: ;in, Coordinates The two-dimensional cumulative probability value at the location; Coordinates in the visual saliency weight distribution matrix The weight value at the location; : These represent the height and width of the weight matrix, respectively. After construction, the system generates a sequence of pseudo-random numbers uniformly distributed between 0 and 1. For each random number in the sequence, a binary search is used to find the corresponding spatial coordinates in the CDF function. This process is mathematically equivalent to sampling the inverse transform of the probability density function. The positioning logic of the reverse search operation is as follows: ;in, : No. The coordinates of the seed points; The first in the pseudo-random number sequence A random number; The cumulative probability distribution function is the inverse function of the two-dimensional cumulative probability distribution function. Through this mechanism, regions with larger weight values ​​have a wider cumulative probability interval, thus increasing the probability of being "hit" by a random number. Finally, the system cleans the generated coordinate set, removing duplicate points and invalid points located at the screen edges. The resulting seed point set exhibits a spatial density distribution highly positively correlated with visual saliency.

[0031] Specifically, the process of generating a spatially geometrically constrained adaptive viewport topology mesh by performing weighted Thiessen polygon subdivision based on the seed point set is as follows: The coordinates of each seed point in the non-uniform seed point set are read, and the corresponding weight value in the visual saliency weight distribution matrix is ​​read as the growth potential parameter of the seed point; the screen pixels within the visible area are traversed, the Euclidean spatial proximity between the screen pixels and the seed points is quantified, and the corresponding growth potential parameter is introduced to perform potential field intensity modulation and bias correction operations on the Euclidean spatial proximity, constructing a power distance metric that characterizes the expansion competitiveness of the viewport region; the power distance metric of the seed point to the current screen pixel is compared, and the current screen pixel is classified into the index region to which the seed point with the smallest power distance metric belongs. After completing the regional classification of all screen pixels, the pixel boundary lines between different index regions are extracted, and a spatially geometrically constrained adaptive viewport topology mesh enclosed by polygonal boundaries is constructed.

[0032] In this implementation scheme, a weighted Thiessen polygon algorithm (PowerDiagram) is used to partition the screen space during the generation of the adaptive viewport topology mesh. Unlike ordinary Thiessen polygons, which only consider geometric distance, this scheme introduces the concept of "growth potential," meaning that the greater the weight of a seed point, the smaller the area it controls (for high-frequency sampling), or, according to the reverse logic, the greater the weight, the more compact the area. In this embodiment, to achieve the specific goal of "generating a small-area viewport with high weights," we use a weighted distance subtraction model. For any pixel on the screen, calculating which seed point it belongs to depends on a combined metric of the Euclidean distance from the pixel to the seed point and the seed point's weight. The formula for calculating the power distance metric is as follows: ;in, The power-law distance metric between the q-th seed point and any pixel p on the screen; The horizontal and vertical coordinates of screen pixel p; : The x and y coordinates of the q-th seed point; The visual saliency weight value corresponding to the q-th seed point, i.e., the growth potential parameter; The potential energy adjustment coefficient controls the sensitivity of the weights to the grid area. This coefficient is determined through prior simulation experiments and is usually a positive number. In this formula, subtracting the weight term means that seed points with larger weights have smaller calculated "power distances," thus having an advantage in region competition and attracting closer pixels. However, because seed points are extremely dense in high-weight areas, this competition can actually lead to a smaller area for each polygonal unit generated in the end. The system traverses all pixels in the image, compares the power distance values ​​of all seed points to the current pixel, and assigns the pixel to the seed point with the smallest value. The region classification logic is as follows: ;in, : The index of the grid region to which pixel p ultimately belongs; The set of all seed points. After classification, the boundaries of different index regions form polygonal boundary lines, thus constructing a spatially geometrically constrained adaptive viewport topology mesh with varying density, providing a precise geometric container for subsequent hierarchical rendering.

[0033] Specifically, using a spatially geometrically constrained adaptive viewport topology grid as the data projection container, the global human flow interaction entropy field data is resampled and mapped to each viewport cell. Within a small area of ​​high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and the flow field line integral convolution rendering is performed to reconstruct the micro-trajectory dynamics. The specific process is as follows: A Gaussian white noise texture image with the same size as the small area of ​​high-frequency sampling viewport is generated; each pixel within the high-frequency sampling viewport is traversed, and bidirectional streamline integration is performed starting from the current pixel, guided by the vector direction in the global human flow interaction entropy field data, to obtain the streamline trajectory coordinate sequence passing through the pixel; the gray value corresponding to the streamline trajectory coordinate sequence in the white noise texture image is read, the convolution weighted average of the corresponding gray value is calculated, and the convolution weighted average is assigned to the current pixel to generate a micro-trajectory dynamic rendering layer with fluid texture features.

[0034] In this implementation scheme, for small-area, high-frequency sampling viewports identified as high-risk by the system, texture reconstruction is performed using flow field line integral convolution (LIC) technology to visually present the complex micro-motion trajectories of pedestrian flow within the area. First, a Gaussian white noise image is generated as the computational base map, with each pixel assigned a random grayscale value to provide a high-frequency texture carrier for the flow field. Then, the core step is to "smear" or integrate this random noise along the streamline direction. For any pixel within the viewport, bidirectional streamline tracing is performed using the vector direction in the global pedestrian flow interaction entropy field data as the tangent direction, thereby obtaining the streamline trajectory passing through that point. Mathematically, this process is represented as a convolution operation along the vector field on the noise texture. The formula for generating the dynamic rendering layer of the micro-trajectory is as follows: ;in, : The output grayscale value at pixel p in the micro-trajectory dynamic rendering layer; The convolution length of the streamline integral determines the length of the visual trailing. The convolution kernel function, usually a Hanning window or a Gaussian function, is used to assign weights to noise points at different locations on the streamline, with greater weights for points closer to the center point p. The input grayscale function for Gaussian white noise texture images; : Streamline trajectory function, representing the integral distance from pixel p along the vector field direction. The position coordinates at that time. Through the above calculations, the originally chaotic white noise pixels are smoothed and elongated along the direction of the crowd flow vector, thus visually forming a fluid texture with a strong sense of direction and continuity, which can clearly reveal the squeezing direction and turbulence pattern of the crowd in the congested area.

[0035] Specifically, within a large-area low-frequency sampling viewport, scalar homogenization of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the area. The specific process for combining and generating a dynamic visual interactive display interface for pedestrian flow with multi-granular data fusion is as follows: The scalar amplitude of the global pedestrian flow interaction entropy field data within the physical space grid unit covered by the large-area low-frequency sampling viewport is statistically analyzed, and the arithmetic mean of the scalar amplitudes is calculated as the unified pressure value of the viewport unit. The unified pressure value is input into a preset color lookup table for linear interpolation mapping to obtain the corresponding RGBA color channel parameters. The RGBA color channel parameters are used to fill the low-frequency sampling viewport, generating a color-gradient fill layer. An image synthesis buffer is established, using the color-gradient fill layer as the background layer and the micro-trajectory dynamic rendering layer as the foreground layer. The region ID of the spatial geometric constraint adaptive viewport topology grid is read as the mask channel, and the foreground layer is superimposed onto the background layer using an Alpha blending algorithm to output the dynamic visual interactive display interface for pedestrian flow with multi-granular data fusion.

[0036] In this implementation scheme, for large-area, low-frequency sampling viewports determined by the system to be low-risk, scalar homogenization is employed to reduce visual load and provide a macroscopic situational background. The system statistically analyzes the entropy values ​​within all physical grid cells covered by the viewport, calculates their arithmetic mean, thereby ignoring minor local fluctuations and extracting a uniform pressure value representing the overall congestion level of the area. The formula for calculating the uniform pressure value is as follows: ;in, The uniform pressure value calculated by the idx-th low-frequency sampling viewport unit; The total number of pixels or grid cells covered by this viewport unit; The set of spatial regions of this viewport unit; The scalar amplitude in the global pedestrian flow interaction entropy field data. After obtaining a uniform pressure value, the system uses a pre-set color lookup table to map it to a specific RGBA color, forming a color level fill layer. Finally, in the image synthesis stage, the Alpha blending algorithm is used to fuse the fine microscopic trajectory layer with the macroscopic color level fill layer, ensuring a natural visual transition between the two. The calculation formula for image synthesis is as follows: ;in, : The pixel color vector of the final output interface; The foreground color of the dynamic rendering layer for microscopic trajectories; : The background color of the level fill layer; The mask channel value is generated based on a spatially geometrically constrained adaptive viewport topology mesh. In high-frequency sampling viewport areas, this value is 1, and in low-frequency sampling viewport areas, it is 0. Transition values ​​can be set at the boundaries to achieve anti-aliasing. This step, through layer overlay technology, achieves an adaptive visualization effect where key areas show details, while background areas show overall situational awareness, effectively resolving the information expression conflict problem under massive data.

[0037] Please see Figure 2 The system for generating a dynamic and interactive display interface for pedestrian flow includes the following modules: an entropy field construction module, used to vectorize and parse the collected time-series data of pedestrian trajectories, extract trajectory segments through a sliding time window, calculate the angular dispersion and velocity magnitude attenuation rate between adjacent trajectory vectors, construct a local interactive entropy value reflecting the disorder of the micro-motions of the group, and perform inverse distance weighted interpolation to map the discrete local interactive entropy values ​​to physical space grid cells, generating a global pedestrian flow interactive entropy field; a saliency weight calculation module, used to perform multi-scale Gaussian difference convolution operations on the global pedestrian flow interactive entropy field data, extract the spatial extreme points and edge gradient features of the local entropy values, and construct a visual saliency weight distribution matrix corresponding to the screen pixel coordinate system through normalization mapping; and a viewport topology generation module, used to use the visual saliency weight distribution matrix as a probability density function to perform Monte Carlo recalculation within the visible area. Priority sampling generates a non-uniform seed point set with density varying with weights. Based on this seed point set, weighted Thiessen polygon subdivision is performed to generate a spatially geometrically constrained adaptive viewport topology mesh. High-weight regions generate small-area high-frequency sampling viewports due to the high-density distribution of seed points, while low-weight regions generate large-area low-frequency sampling viewports due to the low-density distribution of seed points. The interface rendering generation module uses the spatially geometrically constrained adaptive viewport topology mesh as a data projection container to resample and map the global pedestrian flow interaction entropy field data to each viewport unit. Within the small-area high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics. Within the large-area low-frequency sampling viewport, scalar homogenization processing of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the region. These layers are combined to generate a multi-granularity data fusion-based dynamic visualization and interactive display interface for pedestrian flow.

[0038] In this implementation scheme, the entropy field construction module serves as the data processing foundation of the entire system. Its primary responsibility is to transform the raw, discrete temporal data of human trajectories acquired by the front-end acquisition devices into a field theory model that can be continuously computed by the computer. This module first cleans and vectorizes the trajectory data, then uses a sliding time window to extract effective segments to capture dynamic change characteristics. Subsequently, it calculates the local interaction entropy value, reflecting the degree of microscopic motion disorder, through a core algorithm. More importantly, this module performs spatial interpolation operations, extending isolated values ​​existing only at trajectory points to the entire physical space grid, thereby outputting global human flow interaction entropy field data, providing physically meaningful underlying data support for subsequent visual analysis. The significant weight calculation module acts as the system's visual attention analysis engine. Its core function is to automatically determine the display priority of the monitoring screen based on the feature distribution of the data. This module receives global human flow interaction entropy field data and uses feature extraction algorithms from the field of image processing to identify congestion centers (spatial extreme points) and diffusion edges (edge ​​gradient features). Through this process, the module transforms the objective physical entropy field into a visual saliency weight distribution matrix corresponding one-to-one with screen pixels, quantifying the information carrying value of each pixel region on the screen. This guides subsequent interface layout generation, ensuring that computing resources are concentrated on high-risk areas that truly require attention. The viewport topology generation module is the interface's structural builder, responsible for dynamically generating a non-uniform display grid based on saliency weights. This module uses probabilistic statistical methods to densely place seed points in high-weight areas and sparsely place seed points in low-weight areas, constructing weighted Thiessen polygons based on these points. This process essentially reconstructs the screen's topology in real time, automatically generating a spatially geometrically constrained adaptive viewport topology grid. The grid structure output by this module directly determines the spatial division logic of the interface, allowing high-risk areas to be finely segmented to accommodate more details, while low-risk areas are coarsely aggregated to display the macroscopic landscape. The interface rendering generation module, as the final visual synthesizer, is responsible for mapping data content onto the aforementioned topology grid and outputting the visualization. This module automatically switches rendering strategies based on the geometric area of ​​the viewport unit: for small viewports, it uses a vector rendering engine to reconstruct the flow direction and dynamic details of microscopic trajectories; for large viewports, it uses a scalar rendering engine to generate background color blocks reflecting average pressure. Finally, the module uses image compositing technology to seamlessly fuse these two different granularities of rendering results, outputting an interactive display interface that provides both a macroscopic global view and microscopic local precision for monitoring personnel to perform decision analysis.

[0039] In summary, this application has at least the following effects: A method and system for generating a dynamic, interactive display interface for pedestrian flow visualization has been developed. By constructing a global interactive entropy field that integrates microscopic motion features, it overcomes the limitation of traditional single-density statistics in identifying dynamic congestion risks. It innovatively introduces a weighted Thiessen polygon subdivision mechanism driven by visual saliency, achieving spatial geometric constraint-based adaptive reconstruction of the display interface. Combined with a multi-granularity fusion rendering strategy based on flow field line integral convolution and scalar homogenization, it effectively solves the technical contradiction between information overload and loss of key details in large-scale complex scenarios. While significantly reducing rendering computational overhead, it achieves efficient simultaneous interactive display of macroscopic situational awareness and microscopic trajectory dynamics, greatly improving the decision-making efficiency and response speed of public safety monitoring.

[0040] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0041] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0042] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0043] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0044] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0045] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for generating a dynamic, visual, interactive display interface for pedestrian flow, characterized in that: Includes the following steps: S1. The collected time-series data of human trajectories is vectorized and parsed. Trajectory segments are extracted by sliding time windows. The discrepancy of the angle between adjacent trajectory vectors and the velocity magnitude attenuation rate are calculated. A local interaction entropy value reflecting the degree of disorder of the micro-movement of the group is constructed. Inverse distance weighted interpolation is performed to map the discrete local interaction entropy value to physical space grid cells to generate global human flow interaction entropy field data. S2. Perform multi-scale Gaussian difference convolution operation on the global human flow interaction entropy field data, extract the spatial extreme points and edge gradient features of local entropy values, and construct a visual saliency weight distribution matrix corresponding to the screen pixel coordinate system through normalization mapping. S3. Using the visual saliency weight distribution matrix as the probability density function, Monte Carlo importance sampling is performed within the visible area to generate a non-uniform seed point set whose density varies with the weight. Based on this seed point set, weighted Thiessen polygon subdivision is performed to generate a spatially geometrically constrained adaptive viewport topology mesh. High-weight regions generate small-area high-frequency sampling viewports due to the high-density distribution of seed points, while low-weight regions generate large-area low-frequency sampling viewports due to the low-density distribution of seed points. S4. Using a spatially geometrically constrained adaptive viewport topology grid as the data projection container, the global pedestrian flow interaction entropy field data is resampled and mapped to each viewport unit: In the small-area high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics; in the large-area low-frequency sampling viewport, scalar homogenization processing of the entropy field data is performed to generate a color-level fill layer that reflects the average pressure value of the region, and the combination generates a multi-granular data fusion dynamic visualization interactive display interface for pedestrian flow.

2. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 1, characterized in that: The specific process of vectorizing and parsing the collected personnel trajectory time-series data, extracting trajectory segments through a sliding time window, calculating the angular dispersion and velocity magnitude attenuation rate between adjacent trajectory vectors, and constructing a local interaction entropy value reflecting the degree of disorder in the micro-motions of the group is as follows: Trajectory segments within the sliding time window are traversed, and the instantaneous velocity vectors of the trajectory segments at the current time and the previous time are extracted. Vector normalization is performed on the instantaneous velocity vectors of the current time and the previous time. The dot product of the two normalized vectors is calculated to obtain the cosine value of the direction angle. The variance of the cosine value of the direction angle within the sliding time window is calculated as the angle dispersion. Simultaneously, the first-order difference sequence of the velocity magnitude of the trajectory segment within the sliding time window is calculated, and the cumulative amplitude of the negative elements in the first-order difference sequence is used as the velocity magnitude attenuation rate. The dispersion of the included angle and the velocity magnitude attenuation rate are weighted and summed to generate a disorder index for a single trajectory. The arithmetic mean of the disorder index of the trajectory within a unit physical space grid is then calculated, and the arithmetic mean is used as the local interaction entropy value at the grid center.

3. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 2, characterized in that: The specific process of performing inverse distance weighted interpolation to map discrete local interaction entropy values ​​to physical space grid cells and generate global pedestrian flow interaction entropy field data is as follows: Establish a coordinate index matrix for physical space grid cells, determine the geometric center coordinates of each grid cell to be interpolated, set the search radius with the geometric center coordinates as the center, and retrieve reference grid points within the search radius that already have local interaction entropy values. Calculate the Euclidean distance between the geometric center coordinates of the grid cell to be interpolated and the coordinates of the reference grid point, construct a weight decay function based on the reciprocal of the Euclidean distance, and use the weight decay function to calculate the interpolation weight coefficient of each reference grid point; The local interaction entropy value of the reference grid point within the search radius is weighted linearly combined with the corresponding interpolation weight coefficient. The calculation result is assigned to the grid cell to be interpolated. After completing the traversal calculation of the grid cell, the global human flow interaction entropy field data is output.

4. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 1, characterized in that: Step S2 includes the following steps: Multi-level Gaussian blurring processing is performed on the entropy field data of the entire human flow interaction. The first Gaussian blurred image and the second Gaussian blurred image are generated by using Gaussian kernel functions of different scales. The first Gaussian blurred image and the second Gaussian blurred image are then subjected to pixel-level difference operation to generate a Gaussian difference response map. Traverse the pixels of the Gaussian difference response map, compare the response values ​​of the current pixel with those of its eight neighboring pixels, and mark the points whose response values ​​are simultaneously greater than those of its eight neighboring pixels as spatial extrema. Calculate the horizontal and vertical gradient components of each pixel in the Gaussian difference response map, synthesize the gradient magnitude matrix, spatially superimpose the location information of the spatial extrema with the gradient magnitude matrix, and map the superposition result to the numerical range of 0 to 1 through linear normalization to generate the visual saliency weight distribution matrix.

5. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 1, characterized in that: The specific process of generating a non-uniform seed point set with varying density as weights by performing Monte Carlo importance sampling within the visible region, using the visual saliency weight distribution matrix as the probability density function, is as follows: The visual saliency weight distribution matrix is ​​accumulated across the entire domain. The accumulated result is used as a normalization factor. The weight values ​​of the elements in the matrix are normalized by division, and a two-dimensional cumulative probability distribution function is constructed. Generate a set of pseudo-random number sequences that follow a uniform distribution. Use each random value in the pseudo-random number sequence as an index key and perform a reverse lookup operation in the two-dimensional cumulative probability distribution function to locate the corresponding matrix coordinate position. Extract the matrix coordinate positions locked by the reverse lookup operation, remove duplicate coordinates and boundary overflow coordinates, and record the remaining coordinate position set as a non-uniform seed point set with density varying with weight.

6. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 5, characterized in that: The specific process of generating a spatially geometrically constrained adaptive viewport topology mesh by performing weighted Thiessen polygon subdivision based on the seed point set is as follows: Read the coordinates of each seed point in the non-uniform seed point set, and read the weight value of the corresponding coordinates in the visual saliency weight distribution matrix as the growth potential parameter of the seed point; Traverse the screen pixels within the visible area, quantify the Euclidean spatial proximity between the screen pixels and the seed point, introduce the corresponding growth potential energy parameters to perform potential field intensity modulation and bias correction operations on the Euclidean spatial proximity, and construct a power distance metric that characterizes the expansion competitiveness of the viewport area. Compare the power distance metric values ​​of the seed point to the current screen pixel, classify the current screen pixel to the index region of the seed point with the smallest power distance metric value, and after completing the region classification of all screen pixels, extract the pixel boundary lines between different index regions and construct a spatial geometric constraint adaptive viewport topology mesh enclosed by polygonal boundaries.

7. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 1, characterized in that: Using a spatially geometrically constrained adaptive viewport topology mesh as the data projection container, the global human flow interaction entropy field data is resampled and mapped to each viewport cell. Within a small area of ​​high-frequency sampling viewport, the vector characteristics of the entropy field data are preserved, and the flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics. The specific process is as follows: Generate a Gaussian white noise texture image that is the same size as the small high-frequency sampling viewport; Traverse each pixel within the high-frequency sampling viewport, using the vector direction in the global human flow interaction entropy field data as a guide, perform bidirectional streamline integration starting from the current pixel to obtain the streamline trajectory coordinate sequence passing through the pixel; Read the gray values ​​corresponding to the streamline trajectory coordinate sequence in the white noise texture image, calculate the convolution weighted average of the corresponding gray values, assign the convolution weighted average to the current pixel, and generate a microscopic trajectory dynamic rendering layer with fluid texture features.

8. The method for generating a dynamic, visual, interactive display interface for pedestrian flow according to claim 7, characterized in that: Within a large-area low-frequency sampling viewport, scalar homogenization of the entropy field data is performed to generate a color-gradient fill layer reflecting the average pressure value of the area. The specific process of combining and generating a dynamic visualization and interactive display interface for pedestrian flow through multi-granularity data fusion is as follows: The scalar amplitude of the global human flow interaction entropy field data within the physical space grid cell covered by the large-area low-frequency sampling viewport is statistically analyzed, and the arithmetic mean of the scalar amplitudes is calculated as the uniform pressure value of the viewport cell. A uniform pressure value is input into a preset color lookup table for linear interpolation mapping to obtain the corresponding RGBA color channel parameters. The RGBA color channel parameters are then used to fill the low-frequency sampling viewport to generate a color level fill layer. An image compositing buffer is established, the color level fill layer is used as the background layer, the micro-trajectory dynamic rendering layer is used as the foreground layer, the region ID of the spatial geometric constraint adaptive viewport topology grid is read as the mask channel, and the foreground layer is superimposed on the background layer through the Alpha blending algorithm to output a dynamic visualization and interactive display interface for pedestrian flow with multi-granular data fusion.

9. A system for generating a dynamic, visual, interactive display interface for pedestrian flow, applied to the method for generating a dynamic, visual, interactive display interface for pedestrian flow as described in any one of claims 1-8, characterized in that, Includes the following modules: The entropy field construction module is used to perform vectorized analysis on the collected time-series data of human trajectories. It extracts trajectory segments through a sliding time window, calculates the angular dispersion and velocity magnitude attenuation rate between adjacent trajectory vectors, constructs local interaction entropy values ​​that reflect the degree of disorder in the micro-movement of the group, and performs inverse distance weighted interpolation to map the discrete local interaction entropy values ​​to physical space grid cells to generate global human flow interaction entropy field data. The saliency weight calculation module is used to perform multi-scale Gaussian difference convolution operation on the global human flow interaction entropy field data, extract the spatial extreme points and edge gradient features of local entropy values, and construct the visual saliency weight distribution matrix corresponding to the screen pixel coordinate system through normalization mapping. The viewport topology generation module uses the visual saliency weight distribution matrix as a probability density function to perform Monte Carlo importance sampling within the visible area to generate a non-uniform seed point set whose density varies with the weight. Based on this seed point set, weighted Thiessen polygon subdivision is performed to generate a spatially geometrically constrained adaptive viewport topology mesh. High-weight regions generate small-area high-frequency sampling viewports due to the high-density distribution of seed points, while low-weight regions generate large-area low-frequency sampling viewports due to the low-density distribution of seed points. The interface rendering generation module uses a spatially geometrically constrained adaptive viewport topology grid as a data projection container to resample and map the global human flow interaction entropy field data to each viewport unit: in small-area high-frequency sampling viewports, the vector characteristics of the entropy field data are preserved, and flow field line integral convolution rendering is performed to reconstruct the microscopic trajectory dynamics; in large-area low-frequency sampling viewports, scalar homogenization processing of the entropy field data is performed to generate a color-level fill layer that reflects the average pressure value of the region, and the combination generates a multi-granular data fusion dynamic visualization interactive display interface for human flow.

Images