A city public safety situation awareness prediction method

Abstract

The application provides a city public safety situation awareness prediction method, relates to the field of information technology, and comprises the following steps: identifying the local bending degree of each contour line segment according to a boundary form, identifying potential irregular points according to the bending degree, marking the potential irregular points, integrating the spatial coordinates of the potential irregular points, and obtaining a curvature change distribution; identifying the depth of concave of the potential irregular points by analyzing the curvature change distribution, combining the depth value of the concave with the area of the concave region, determining the spatial volume size of each concave, and obtaining a concave spatial volume distribution; determining a high-risk area through the predicted pressure concentration degree and boundary deformation degree, determining the position of the force convergence total amount that exceeds a preset warning standard by counting the force convergence total amount of each concave of the high-risk area, and marking the position as a risk point.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for predicting urban public safety situation. Background Technology

[0002] In the field of urban public safety management, ensuring safety in crowded situations is crucial, especially during large-scale events or emergencies, where preventing stampedes becomes a core task for maintaining social stability. However, current safety assessment methods, when dealing with complex crowd gathering scenarios, primarily rely on crowd density to judge risk, assuming that higher density equates to higher danger. This approach often overlooks the impact of crowd distribution patterns on safety, leading to biased assessments of potential risks, particularly in complex crowd gathering scenarios where the accuracy of risk assessments is significantly reduced. Boundary shape, as a key factor determining the distribution of crowd pushing pressure, often exhibits irregularities that lead to pressure concentration in localized areas, potentially triggering stampedes. Because the dynamic evolution of boundary shapes is not fully considered, existing technologies struggle to capture hidden risks during the transition from regular to irregular shapes and cannot promptly identify pressure accumulation in locally concave areas. Specifically, in real-world scenarios, such as during a festival, the boundaries of crowd gathering areas may form a jagged distribution due to terrain constraints or crowd movement. At concave boundary areas, the force of crowd pushing tends to concentrate, causing a rapid increase in localized pressure, while the density in surrounding areas may not be high. In such situations, traditional density monitoring is completely unable to detect potential hazards in depressions, leading to misjudgments of risk by managers and even missing the optimal intervention window. Therefore, accurately identifying the irregular evolution of boundary shapes and their impact on local pressure distribution in dynamically changing crowd gathering scenarios has become a key issue in improving the ability to predict public safety situations. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for predicting urban public safety situational awareness, comprising: By acquiring crowd video streams through the monitoring system, images of the crowd gathering area are extracted, and boundary detection algorithms are used to separate the contour lines of the gathering boundary to obtain the boundary morphology including the spatial location of the contour lines. At the same time, the distance distribution and density distribution between individuals are extracted to obtain the basic data of crowd pressure gradient. Based on the boundary morphology, the local curvature of the contour line segment is identified, potential irregular points are marked, and their spatial coordinates are integrated to obtain the curvature change distribution. The concavity depth of potential irregular points is identified by the curvature change distribution, and the spatial volume of each concavity is determined by combining the concavity depth and the area of ​​the concavity region, thus obtaining the concavity spatial volume distribution. The direction of pushing force between individuals is identified by the basic data of the crowd pressure gradient. The number of people and their movement direction converging on each depression area are identified by the volume distribution of the depression space. The depression intensity in the direction of pushing force is evaluated to obtain the local pressure concentration distribution. The local pressure concentration distribution is analyzed over a continuous period of time. At the same time, the boundary morphology of the continuous frame images is compared to identify the rate of evolution of the boundary from a regular shape to an irregular shape. The increase in indentation depth and the expansion of indentation area per unit time are obtained to determine the trend of shape evolution and predict the degree of pressure concentration and boundary deformation in future periods. High-risk areas are determined by the degree of pressure concentration and the degree of boundary deformation. The total force convergence at each depression within the high-risk area is calculated, and locations where the total force convergence exceeds a preset warning standard are marked as risk points.

[0004] Furthermore, the process involves acquiring crowd video streams through a monitoring system, extracting images of the crowd gathering area, using a boundary detection algorithm to separate the crowd boundary contour segments, obtaining the boundary morphology including the spatial location of the contour lines, and simultaneously extracting the distance distribution and density distribution between individuals to obtain basic data on the crowd pressure gradient, including: The video stream data from the monitoring system is acquired and decomposed frame by frame to extract image frames containing crowds; image processing technology is used to identify the boundary between the crowd area and the background to obtain the image of the area where the crowd is gathered. For the image of the region, a boundary detection algorithm is used to extract boundary pixels, connectivity analysis is performed to generate complete clustered boundary contour segments, the image coordinate values ​​of the contour points are recorded, and the actual spatial positions are converted through camera calibration parameters to obtain the spatial coordinate sequence of the boundary shape; Identify the location points of each individual within the area enclosed by the spatial coordinate sequence, calculate the distance between adjacent individual location points, count the number of individual pairs within different distance ranges, and form the distance distribution characteristics between individuals; The local density value is calculated based on the number of individuals per unit area and the average spacing. The local density value is then interpolated to generate a continuous density field. The gradient values ​​of the density field in each direction of space are calculated to obtain the basic data of the population pressure gradient.

[0005] Furthermore, the step of identifying the local curvature of the contour line segment based on the boundary morphology, marking potential irregular points, and integrating their spatial coordinates to obtain the curvature variation distribution includes: Based on the spatial coordinate sequence of the boundary morphology, three consecutive coordinate points are selected to calculate the local curvature value, and the complete curvature sequence is obtained by sliding along the contour line. Perform a moving average calculation on the curvature sequence, mark the contour segments where the mean curvature exceeds the curvature threshold, and record the peak curvature points; The difference between the curvature peak point and the reference curvature is calculated, potential irregular points are identified and grouped and clustered to obtain the curvature variation distribution.

[0006] Furthermore, the step of identifying the concavity depth of potential irregular points through the curvature change distribution, and determining the spatial volume of each concavity by combining the concavity depth and the area of ​​the concavity region, to obtain the concavity spatial volume distribution, includes: Based on the curvature change distribution, identify the concave segments formed by continuous irregular points, connect the concave boundary points to form a reference baseline, calculate the vertical distance from each point on the concave contour to the reference baseline, and select the maximum vertical distance as the concave depth. A horizontal cutting surface is established according to the concave depth. The areas of adjacent cutting surfaces are added together to obtain the spatial volume value. The volume value of each concave location is recorded to form the spatial volume distribution of the concave area.

[0007] Furthermore, the step of identifying the direction of pushing force between individuals using the basic data of population pressure gradient, and combining this with the volume distribution of the depression space to identify the number of people converging on each depression area and their direction of movement, assessing the depression intensity in the direction of pushing force, and obtaining the local pressure concentration distribution includes: Based on the basic data of population pressure gradient, the pressure difference between adjacent individuals is calculated, converted into the magnitude of the pushing force, and the vector summation is used to obtain the direction and intensity of the combined pushing force at each location point. Identify the force vector pointing to the center of each depression area, count the number of individuals within the coverage area as the number of people gathered, and calculate the angle between the moving velocity vector and the line connecting the depression center based on the changes in the individual positions in consecutive frames. The summation of the individual pushing force intensity values ​​moving towards the depression, combined with the volume distribution of the depression space, is used to calculate the total pushing force per unit volume as the depression intensity value. A pressure recording lattice is established on the boundary contour plane. The pressure propagation attenuation coefficient is calculated based on the distance. The pressure contributions of all depression areas are superimposed to obtain the local pressure concentration distribution.

[0008] Furthermore, the assessment of the indentation strength in the direction of the pushing force also includes: Calculate the distance between the centers of gravity of adjacent individuals based on the direction of the pushing force between individuals, mark the high aggregation state, extract the temporal movement trajectory of people around the depression area, and identify body posture parameters. By comparing the position difference vectors and the spacing reduction rate at adjacent time points, the trajectory deflection pattern and body tilt direction sequence are determined; Tracing back along the reverse extension line of the body tilt direction to the peripheral low-density region, connecting them to form a force transmission path, and performing kernel density estimation to obtain the force application circuit network; The main convergence channels are determined by connecting the peak path density in the applied force line network.

[0009] Furthermore, the analysis of continuous time-period changes in the localized pressure concentration distribution, combined with comparison of the boundary morphology in consecutive frame images, identifies the rate of boundary evolution from a regular shape to an irregular shape, obtains the increase in depression depth and the expansion of depression area per unit time, determines the shape evolution trend, and predicts the degree of pressure concentration and boundary deformation in future time periods, including: The pressure value sequence is obtained by sampling the localized pressure concentration distribution, and the pressure change rate is calculated; Extract boundary contour morphological feature data at the corresponding time point, calculate roundness and regularity, and measure the rate of increase in depression depth and the rate of area expansion. Linear regression was used to fit the rate change relationship, and the shape evolution trend was obtained based on the regression coefficient. The depth, area and corresponding pressure concentration in the future period were predicted by the security situation awareness model.

[0010] Furthermore, the rate at which the identification boundary evolves from a regular shape to an irregular shape also includes: Extract the boundary contour pixel coordinate sequence from consecutive frame images, calculate the displacement vector of the corresponding contour points in adjacent frames, and identify the inflection point position and the indentation advance distance; Calculate the change in the included angle between adjacent line segments at the turning point, determine whether a smooth turn becomes a sharp turn, and measure the minimum distance between the inner boundary of the depression and the outer normal boundary. Using the initial contour coordinate sequence as a reference, the displacement vectors of the current contour points are connected to form a deformation field, thus obtaining the boundary change path.

[0011] Furthermore, the process of determining high-risk areas based on the degree of pressure concentration and the degree of boundary deformation, calculating the total force convergence at each depression within the high-risk area, and marking locations where the total force convergence exceeds a preset warning standard as risk points includes: The pressure concentration degree and the boundary deformation degree are compared with the pressure threshold and deformation threshold, respectively. If both exceed the threshold, the area is identified as a high-risk area. The inward converging pushing force values ​​at each depression within the high-risk area are obtained and summed to obtain the total force convergence. When the total force convergence exceeds the preset warning standard, the spatial coordinates of the depression center are recorded and marked as a risk point.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for predicting urban public safety situational awareness. By real-time monitoring of crowd video streams, it extracts images of clustered areas and uses a boundary detection algorithm to separate contour segments. Combining this with inter-individual distance and density distribution, it constructs crowd pressure gradient data to identify localized pressure concentrations. This invention innovatively analyzes the curvature changes and concave space volume of contour segments to assess the direction of pushing force and the trend of personnel movement, predicting the rate of evolution of boundary morphology from regular to irregular and the future degree of pressure concentration, ultimately accurately locating high-risk areas and risk points. Through continuous time-period change analysis and safety situational awareness, this invention achieves dynamic monitoring and early warning of crowd gathering risks, significantly improving the efficiency of public place safety management, reducing the potential accident risks caused by pressure concentration, and providing a scientific basis and technical support for safety assurance in densely populated scenarios. Attached Figure Description

[0013] Figure 1 This is a flowchart of an urban public safety situational awareness and prediction method according to the present invention.

[0014] Figure 2 This is a schematic diagram of an urban public safety situational awareness and prediction method according to the present invention.

[0015] Figure 3 This is another schematic diagram of an urban public safety situational awareness and prediction method according to the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1-3 This embodiment of a method for predicting urban public safety situation may specifically include: S101. Obtain crowd video stream through monitoring system, extract crowd gathering area image, process crowd gathering area image with boundary detection algorithm, separate gathering boundary contour line segment to obtain boundary shape, the boundary shape includes the spatial position of contour line, and at the same time extract the distance distribution and density distribution between individuals in the image to obtain basic data of crowd pressure gradient.

[0018] The system acquires video stream data from the monitoring system, decomposes it frame-by-frame according to the video frame rate, and extracts image frames containing crowd clusters. Image processing techniques are used to identify the boundary between the crowd area and the background environment, and the crowd area is processed to obtain an image of the crowd clustered region. A boundary detection algorithm is used to identify the edges of the crowd cluster, extract boundary pixels, and perform connectivity analysis to generate a complete cluster boundary contour segment. The image coordinates of each contour point are recorded and converted into actual spatial positions using camera calibration parameters, resulting in a spatial coordinate sequence of the boundary morphology. Within the clustered area enclosed by the spatial coordinate sequence, the location points of each individual are identified, the distance between adjacent individual location points is calculated, and the number of individual pairs within different distance ranges is counted to form the distance distribution characteristics between individuals. Local density values ​​are calculated based on the number of individuals per unit area and the average spacing. Interpolation processing is performed on the local density values ​​to generate a continuous density field, and the gradient values ​​of the density field in each spatial direction are calculated. The relationship between the gradient values ​​and the individual spacing yields the basic data of the crowd pressure gradient.

[0019] Specifically, in one implementation, the monitoring system acquires real-time video streams via cameras deployed at high locations in the gathering area. The video streams are decoded at a standard frame rate of 25 frames per second. The system establishes a background model of the scene and separates the crowd activity area by comparing the current frame with the background model. The separated crowd area is then binarized to clearly distinguish it from the background area, and image frames containing crowd activity are extracted.

[0020] It should be noted that the boundary detection algorithm is based on the principle of image gradient analysis. It identifies abrupt changes in grayscale as boundary points by calculating the grayscale change rate of each pixel in the horizontal and vertical directions. The input to the algorithm is a binarized image of the crowd area, where the crowd area is represented by white pixels and the background area by black pixels. During processing, gradient calculations are performed on the image, and pixels with gradient magnitudes exceeding a set threshold are extracted. These points are located at the boundary between the crowd area and the background area. Connectivity analysis is then performed on the extracted boundary pixels, connecting adjacent boundary points into continuous contour segments. The output of the algorithm is a sequence of pixel coordinates for these contour segments, which describes the boundary morphology of the crowd gathering area.

[0021] Specifically, the binarization process employs an adaptive thresholding method, dynamically adjusting the binarization threshold based on the local brightness distribution of the image. This results in the crowd gathering area appearing as a white connected region, while the background area appears as black, forming a clear binarized mask image. Boundary extraction is performed on the binarized crowd area image to identify the boundary contour segments of the crowd gathering area. Through image processing techniques, connectivity analysis is performed along the boundary pixels to generate a complete and continuous boundary contour, and the image coordinates of the contour points are recorded. Using pre-acquired camera calibration parameters, including the camera intrinsic matrix and distortion coefficients, the two-dimensional image coordinates are converted into three-dimensional spatial coordinates. The conversion process is based on the camera imaging principle and combines actual site height information to determine the true spatial location of the boundary points on the ground. Within the crowd gathering area, the location of each individual is determined by identifying their characteristic contours (such as the head and shoulder area). The Euclidean distance between the center points of adjacent individuals is calculated, and the number of individuals per unit area is counted. The local density value is calculated based on the relationship between the number of individuals and the average spacing; a higher density value indicates a denser crowd in the area. Interpolation processing is performed on discrete density sampling points to generate a continuous and smooth density distribution field in space. Gradient calculations are performed on the density field. By calculating the rate of change of density in various spatial directions, the density gradient vector is obtained. The direction of the density gradient indicates the direction of the fastest increase in population density, and the magnitude of the gradient reflects the drastic degree of density change. Combined with individual spacing information, a correspondence between the density gradient and the crowd pushing pressure is established, yielding fundamental data on the pressure gradient that reflects the intensity and direction of the crowd pushing.

[0022] S102. Identify the local curvature of each contour line segment based on the boundary morphology, identify potential irregular points based on the curvature, mark them, integrate the spatial coordinates of the potential irregular points, and obtain the curvature change distribution.

[0023] Based on the spatial coordinate sequence of the boundary morphology, three consecutive coordinate points on the contour are selected. The radius of the arc formed by these three points is calculated. The local curvature value of the midpoint is determined based on the reciprocal relationship between curvature and radius. Coordinate point groups are selected sequentially along the contour line to calculate the complete curvature sequence. A fixed-length window moving average is applied to the curvature sequence to eliminate curvature fluctuations caused by local measurement errors. If the mean curvature value within the window exceeds a preset bending threshold, the contour segment is marked as a bending region, and the coordinate point corresponding to the maximum curvature value within the bending region is recorded as the curvature peak point. For each curvature peak point, the difference between its curvature value and the contour baseline curvature is calculated. When the difference exceeds an irregularity determination threshold, the point is identified as a potential irregularity point. A distance-based neighborhood clustering method is used to group the spatial coordinates of potential irregularities, and the number of irregularities and curvature intensity within each group are statistically analyzed to obtain the curvature variation distribution.

[0024] Specifically, in one implementation, the curvature of the boundary morphology is calculated by analyzing the geometric relationships of continuous coordinate points on the contour line. A continuous set of sampling points is selected on the contour line, and the curvature value is calculated based on the spatial relationship of the point set. Curvature κ is defined as a quantitative index of the degree of bending at that point; a larger value indicates more severe bending. When the sampling points are nearly collinear, the curvature tends to zero, indicating that the contour at that point is close to a straight line; when the sampling points form a small arc, the curvature value is larger, indicating a high degree of bending in the contour at that point.

[0025] Specifically, a fixed-length window is set to contain 5 to 7 consecutive sampling points. The window slides point by point along the curvature sequence, with each slide step being one sampling point. The local average curvature is obtained by summing the curvature values ​​at each point within the window and dividing by the number of points. This averaging process suppresses abrupt curvature changes caused by image noise or discrete sampling. A preset curvature threshold is set based on the characteristics of the site environment, typically ranging from 0.1 to 0.3 radians per meter. When the average curvature exceeds this threshold, it indicates a significant change in direction for that section of the contour.

[0026] It should be noted that the baseline curvature of the contour is obtained by statistically analyzing the median curvature of the entire boundary contour, representing the overall smoothness of the boundary. The difference between the peak curvature point and the baseline curvature reflects the degree to which that point deviates from the normal shape, and the irregularity judgment threshold is set at 2 to 3 times the baseline curvature. When the curvature of a point is significantly higher than that of the surrounding area, it indicates that a sharp corner or deep depression may have formed at that location, and these locations are prone to stress concentration during crowd pushing.

[0027] In one possible implementation, distance-based neighborhood clustering employs the principle of density reachability, setting a spatial distance threshold of 0.5 meters. When the Euclidean distance between two irregular points is less than this threshold, they are grouped into the same group. The number of irregular points within each group reflects the density of morphological anomalies in the region, and the curvature intensity is obtained by summing the curvature values ​​of all points within the group, reflecting the overall severity of curvature in the region.

[0028] Preferably, the curvature variation distribution is recorded in the form of a two-dimensional heat map, with the horizontal axis representing the position index along the contour line and the vertical axis representing the curvature value at the corresponding position. The distribution of the degree of curvature at various points along the boundary is intuitively displayed by the color intensity, enabling rapid location of potential risk areas.

[0029] S103. By analyzing the distribution of curvature changes, identify the concave depth of potential irregular points. Combine the concave depth value with the area of ​​the concave region to determine the spatial volume of each concave point and obtain the concave spatial volume distribution.

[0030] Based on the irregular point clusters marked in the curvature variation distribution, concave segments formed by continuous irregular points are identified. The contour points at both ends of the concave segments are extracted as concave boundary points. A straight line segment is formed by connecting these concave boundary points as a reference baseline. The vertical distance from each point on the concave contour to the reference baseline is calculated, and the maximum vertical distance value is selected as the concave depth of the concave region. Using this concave depth, horizontal cutting surfaces are established with a layer spacing of one-tenth of the depth. The area enclosed by the closed curve formed by the intersection of each cutting surface and the concave contour is calculated. The areas of adjacent cutting surfaces are accumulated using the trapezoidal area formula to obtain the spatial volume value of the concave region. Based on the spatial volume value and the projected area of ​​the concave region on the ground, the product of the volume value and the projected area of ​​each concave location is calculated. The volume values ​​of each concave location are recorded sequentially along the boundary contour line to form a concave spatial volume distribution reflecting the degree of concavity at different locations.

[0031] Specifically, in one implementation, the identification of concave segments is achieved by analyzing the continuity of irregular points in the curvature variation distribution. When the contour arc length between adjacent irregular points is less than a preset interval threshold, these points are determined to belong to the same concave segment. The determination of concave boundary points adopts a curvature abrupt change detection method, extending the search from the concave segment to both sides. When the curvature value of a point decreases to within the normal range, that point is marked as a concave boundary point.

[0032] Specifically, the establishment of the reference baseline avoids the use of complex fitting algorithms, directly connecting two concave boundary points to form a straight line segment. The vertical distance is calculated using the point-to-line distance formula, calculating the vertical projection length from each sampling point on the concave contour to the reference baseline. By iterating through the vertical distance values ​​of all sampling points, the maximum value is extracted as the feature depth of the concave region; this depth value directly reflects the severity of the concavity.

[0033] It should be noted that the interlayer spacing of the horizontal cut surfaces is adaptively set according to the depression depth, typically around one-tenth of the depth, to balance calculation accuracy and efficiency. The intersection of the horizontal cut surfaces and the depression contour forms a closed curve, and the area enclosed by this curve is calculated. The volume increment between two adjacent cut surfaces is obtained by multiplying the average area of ​​the upper and lower layers by the interlayer spacing, and the total volume is obtained by summing the volume increments of all layers. This layered accumulation method can accurately reflect the three-dimensional spatial characteristics of the depression, providing a quantitative basis for subsequent pressure assessment.

[0034] In one possible implementation, the projected area is calculated by vertically projecting the three-dimensional contour points of the recessed area onto the ground plane. The projected two-dimensional point set is used to determine the outer boundary using a convex hull algorithm, and then the projected area value is obtained using a polygon area calculation method. The product of the volume value and the projected area serves as a comprehensive indicator, reflecting not only the depth characteristics and lateral expansion of the recess but also comprehensively characterizing the overall severity of the recess. In engineering assessments, this can be used to determine the impact of the recess on structural safety and guide repair decisions.

[0035] Preferably, the recording of the volume distribution of the concave space adopts a one-dimensional array structure, where the array index corresponds to the position number on the boundary contour line, and the array element value stores the volume value at the corresponding position. In this way, the correspondence between spatial volume information and position information is realized, which facilitates the subsequent identification of the specific location of high-risk concave areas.

[0036] S104. Identify the direction of pushing force between individuals using basic data on population pressure gradient, identify the number of people converging on the area and their direction of movement based on the volume distribution of the concave space, assess the concavity intensity in the direction of pushing force based on the number of people and their direction of movement, and obtain the local pressure concentration distribution.

[0037] Based on the pressure values ​​and gradient directions in the crowd pressure gradient baseline data, the pressure difference between adjacent individuals is calculated, and this pressure difference is converted into the magnitude of the pushing force. The transmission path of the pushing force is determined according to the gradient direction. By summing the vectors of the pushing forces of multiple adjacent individuals, the direction and intensity of the composite pushing force at each location point are obtained. For the direction of the composite pushing force, the force vector pointing towards the center of each depression area in the volume distribution of the depression space is identified. The number of individuals within the coverage area of ​​these force vectors is counted as the number of people converging. The movement velocity vector is calculated based on the changes in individual positions in consecutive frame images, and the angle between the movement direction and the line connecting the depression center is extracted to obtain the motion parameters of people converging towards the depression. According to the motion parameters, if the angle is less than a preset convergence judgment threshold, it is determined that the individual is moving towards the depression area. The pushing force intensity values ​​of each individual moving towards the depression are accumulated and summed. Combined with the volume of the depression space, the total pushing force per unit volume is calculated as the depression intensity value. Using the indentation intensity value, a uniformly distributed pressure recording point array is established on the boundary contour plane. The pressure propagation attenuation coefficient is calculated based on the distance from each recording point to the center of each indentation. The pressure contribution value of that point is obtained by multiplying the indentation intensity value by the attenuation coefficient. The pressure contributions of all indentation areas are superimposed to obtain a local pressure concentration distribution.

[0038] Specifically, in one implementation, the pressure gradient baseline data includes two key elements: the pressure value at each location and the gradient direction. The pressure value reflects the degree of compression caused by the crowd density at that location, while the gradient direction indicates the direction of the fastest pressure change. The pressure difference between adjacent individuals is obtained by calculating the difference in pressure values ​​between the two locations. This difference is multiplied by an estimated contact area between the individuals to convert it into the actual pushing force. The gradient direction determines the transmission path of the pushing force; along the downward gradient direction, pressure is transmitted from high-pressure areas to low-pressure areas.

[0039] Specifically, the vector summation process employs the parallelogram law, representing the pushing force generated by each individual force as a two-dimensional vector. The magnitude of the vector corresponds to the pushing force intensity, and the direction of the vector corresponds to the gradient direction. For a given location, the pushing force vectors of its eight neighboring locations are collected and superimposed one by one according to the vector addition rule to obtain the combined pushing force at that point. This synthesis reflects the combined effect of pushing forces from multiple directions. When the directions of multiple pushing forces are consistent, the combined force is significantly enhanced; when the directions are opposite, they cancel each other out.

[0040] It should be noted that the center of the depression region is determined based on the spatial volume distribution data of the depression, and the point with the largest volume value is selected as the geometric center of the depression. When identifying the force vector pointing towards the center of the depression, the angle between the direction of the combined pushing force and the line connecting that point to the center of the depression is calculated. When the angle is less than 90 degrees, it is determined that the pushing force has a tendency to converge towards the depression. The number of individuals within the statistical range is achieved by counting heads within the range of the force vector's action, which is defined as a circular area with a radius of 2 meters centered on the origin of the force vector.

[0041] In one possible implementation, the time interval between consecutive frames is set to 0.1 seconds. By comparing the position coordinate changes of the same object in two consecutive frames, the displacement vector is calculated, and then divided by the time interval to obtain the velocity vector. The direction of movement is extracted from the direction angle of the velocity vector, and the angle between the velocity vector and the line connecting the velocity vector and the center of the depression is calculated using the vector dot product formula. The convergence threshold is set to 30 degrees. This threshold is determined based on statistical analysis of a large number of real-world scenarios. When the angle is less than 30 degrees, it indicates that the object's movement path is basically pointing towards the depression area.

[0042] Preferably, the summation of the shoving force intensity values ​​is performed using a weighted summation method. Individuals closer to the center of the depression receive a larger weighting coefficient for their shoving force, with the coefficient set inversely proportional to the distance. The total shoving force per unit volume is calculated by dividing the accumulated total shoving force by the volume of the depression space. This depression intensity value comprehensively reflects the combined effect of three factors: the number of people, the intensity of the shoving, and the confinement of the space. A higher value indicates a higher risk in the depression area.

[0043] For example, pressure recording points are uniformly arranged on the plane containing the boundary profile at a grid spacing of 0.5 meters, and each recording point stores the pressure value at that location. The pressure propagation attenuation coefficient is expressed as a Gaussian function, specifically exp(-d 2 The formula is: d = σ / d, where d is the distance from the center of the depression in meters; σ is the attenuation parameter, set to 5 to 10 meters based on site characteristics. The calculation steps are as follows: First, calculate the distance d from the recording point to the center of the depression, then substitute it into the formula to calculate the attenuation coefficient; then multiply the depression intensity value by this coefficient to obtain the pressure contribution value at that point. The pressure contributions of multiple depressions are accumulated at each recording point to finally form an overall pressure distribution map. The farther away from the center of the depression, the smaller the attenuation coefficient, reflecting the physical law that the pressure influence weakens with increasing distance. Further, the pressure contribution value is calculated by using the depression intensity value as the source intensity, multiplying it by the corresponding attenuation coefficient to obtain the pressure contribution generated by the depression at a specific recording point. When multiple depression areas exist within the site, each depression will generate a pressure field on its surroundings, and these pressure fields are spatially superimposed. The superposition process is achieved by accumulating the pressure contributions of all depressions at each recording point, ultimately forming a two-dimensional pressure field map reflecting the overall pressure distribution.

[0044] Understandably, the concentrated distribution of localized pressure is represented by a heat map, with color depth corresponding to pressure intensity. Red areas indicate high-pressure concentration zones that require close monitoring and crowd control. By updating the pressure distribution map in real time, managers can quickly identify risk areas and promptly implement crowd control measures to prevent stampedes caused by excessive localized pressure concentration.

[0045] For example, at large events, when a deep depression forms at an entrance or exit, the pushing force of the surrounding crowd will converge towards the depression, causing a sharp increase in pressure in that area. The pressure concentration distribution map calculated using this method can accurately identify such high-risk areas, providing a quantitative basis for safety management.

[0046] Based on the direction of the pushing force between individuals, the spacing and aggregation of individuals are identified. The movement paths and body postures of people around the depression area are collected. The direction of change of movement path when the spacing decreases is analyzed. Based on the direction of change of movement path, the tilting direction of body posture after the aggregation intensifies is determined. Based on the tilting direction, the force line from the periphery to the center of the depression is determined. The flow channel of people converging towards the depression area along the force line is identified.

[0047] Based on the direction of the pushing force between individuals, the actual distance between the centers of gravity of adjacent individuals is calculated. If the distance is less than a preset density threshold, it is marked as a high-cluster state. The continuous position coordinates of each person around the concave area are extracted from the video frame sequence to form a temporal movement trajectory. The angle between the line connecting the center of the shoulder and the center of the hip of each person is identified as a body posture parameter. By comparing the position difference vectors of the temporal movement trajectory at adjacent moments, the angle change value of the position difference vector is recorded when the distance between individuals decreases. The angle change value is correlated and matched with the rate of decrease in the distance between individuals to determine the trajectory deflection pattern in the dense state. Based on the rate of change of the body posture parameters over time, the sequence of body tilting directions during the intensification of clustering is obtained. Through the sequence of body tilting directions, the reverse extension line of the tilting direction is traced back to the peripheral low-density area. The peripheral starting point is connected to the center of the concave area to form a force transmission path. Kernel density estimation is performed on multiple transmission paths to obtain the path density distribution as the force application line network. Based on the areas in the force application line network where the path density exceeds a threshold, the density peak connection line is identified as the main convergence channel. The number of people passing through the channel cross-section per unit time and the average moving speed are calculated to determine the flow channel through which people converge towards the concave area along the force application line.

[0048] Specifically, in one implementation, the direction of the pushing force between individuals is determined by the pressure gradient direction obtained in the aforementioned steps, and the actual distance between the centers of gravity of adjacent individuals is calculated using Euclidean distance in the image coordinate system. The density threshold is set based on the average shoulder width of a human body, typically between 0.4 and 0.6 meters. When the distance between the centers of gravity of two individuals is less than this threshold, it indicates that physical contact has occurred. The video frame sequence is continuously recorded at a sampling rate of 25 frames per second, and the position of each person in the image is continuously locked using a target tracking algorithm. The position coordinates of 10 consecutive frames are concatenated to form a temporal movement trajectory.

[0049] Specifically, body posture parameters are obtained based on the biomechanical characteristics of the human body structure. The center point of the shoulder is determined by the midpoint between the left and right shoulder joints, and the center point of the hip is determined by the midpoint between the left and right hip joints; the line connecting these two points represents the trunk axis. The vertical direction is defined in the image coordinate system as the direction perpendicular to the ground, obtained through calibration. The angle between the trunk axis and the vertical direction reflects the degree of tilt of the body under pushing force; the larger the angle, the stronger the lateral pushing force. This tilt is a natural reaction of the body to resist the pushing force, and the tilt direction is opposite to the direction of the pushing force.

[0050] It should be noted that the position difference vector is obtained by calculating the difference in position coordinates between two adjacent moments. Its direction represents the instantaneous direction of movement, and its magnitude represents the speed of movement. When the distance between individuals is continuously decreasing, the angle of the position difference vector at each moment is recorded. The change in angle reflects the magnitude of the adjustment in the direction of movement. The rate of decrease in the distance between individuals is calculated using the derivative of distance with respect to time. Specifically, the rate is obtained by dividing the distance difference between two adjacent moments by the time interval; the time interval is set to 1 second, and the unit of distance is meters. Recording the angle change value in relation to the rate of decrease reveals a positive correlation between the two, meaning that the faster the distance decreases, the more drastic the change in the direction of movement.

[0051] In one possible implementation, the rate of change of body posture parameters over time is obtained by differentially analyzing a sequence of tilt angles. As aggregation intensifies, such as with increased crowd density or pushing force, the external pressure on the body increases, causing the tilt angle to exhibit a monotonically increasing trend with a positive rate of change. The tilt direction at each moment is recorded, forming a sequence of body tilt directions. This sequence contains dynamic evolution information about the pushing process, with each element in the sequence corresponding to the tilt direction angle value at a specific moment.

[0052] Preferably, the force transmission path is traced using a reverse reasoning method. The low-density area boundary is defined as the edge of an area where the population density is less than 2 people per square meter. Extending a line backward along the current slope direction outwards, the area where the extension line reaches a region with significantly reduced population density is identified as the initial source area of ​​the pushing force. The peripheral starting point is defined as the intersection of the extension line and the low-density area boundary. Connecting the peripheral starting points of multiple individuals with the depression center forms multiple force transmission paths. These paths reflect the transmission process of the pushing force converging from the periphery to the depression center.

[0053] For example, the kernel density estimation method is used to analyze the spatial distribution characteristics of force transmission paths and identify the main channels for pedestrian convergence. Multiple force transmission paths are obtained, each represented by a sequence of spatial coordinate points, reflecting the direction of the pushing force from the periphery to the center of the depression. In this embodiment, firstly, a uniformly distributed sampling grid is established on the boundary contour plane, with the grid spacing determined according to the site dimensions; secondly, for each sampling point in the grid, the shortest distance from that point to all paths is calculated, with closer distances indicating closer proximity to a particular path; thirdly, density contribution weights are calculated based on distance using a Gaussian kernel function, with closer distances resulting in larger weights, and the density contributions of all paths to that sampling point are summed to obtain the path density value for that point; finally, the density values ​​of all sampling points are arranged according to spatial location to form a continuous path density distribution map, ultimately yielding a force application path network. This force application path network is presented as a density field, with high density areas representing the convergence of multiple paths, corresponding to the main channels for pedestrian flow. By extracting the connecting lines of density peak areas, key pedestrian convergence paths can be identified. Furthermore, the identification of the main convergence channels is based on peak analysis of the path density distribution. A density threshold is set at 1.5 times the average density; areas exceeding this threshold are marked as high-density pathways. Connecting the center points of adjacent high-density areas forms density peak lines, which represent the main channels where pedestrian traffic converges. The width of these channels is determined based on the lateral span of the high-density areas, typically ranging from 1 to 3 meters.

[0054] Understandably, the cross-section of a passage is defined as a cross-sectional line perpendicular to the passage direction, and the number of people passing through this cross-section per unit time is counted. Using target detection and tracking algorithms, people crossing the cross-sectional line are identified, the crossing time is recorded, and the number of people passing through per unit time is calculated as the pedestrian flow. The average moving speed is obtained by averaging the velocity vector magnitudes of the passing people.

[0055] For example, when a music festival ends, audience members converge on the exit from all directions. If a depression forms at the exit, the surrounding crowd will move towards the depression along specific pathways. This method can accurately identify the location and flow of these pathways, enabling quantitative monitoring of the crowd convergence process.

[0056] S105. Perform continuous time-period change analysis on the local pressure concentration distribution. By comparing the boundary morphology of continuous frame images in the video stream, identify the rate of evolution of the boundary from a regular shape to an irregular shape, obtain the increase in indentation depth and the expansion of indentation area per unit time, determine the trend of shape evolution, and predict the degree of pressure concentration and boundary deformation in future time periods through security situation awareness.

[0057] Localized pressure concentrations are sampled at fixed time intervals to obtain pressure value sequences within continuous time periods. The pressure change rate is calculated by differentiating pressure values ​​between adjacent time periods. Simultaneously, boundary contour images at corresponding moments are extracted from the video stream, and vertex coordinates and curvature values ​​on the contours are identified as morphological feature data. Based on these morphological feature data, the similarity between the boundary contour and a standard circle is calculated as circularity, and the deviation of the aspect ratio of the minimum bounding rectangle of the contour from a preset value is calculated as regularity. If circularity continuously decreases while regularity increases, the boundary is determined to be transitioning from a regular shape to an irregular shape. The depth difference between the deepest points of concavities in adjacent frames is divided by the inter-frame time to obtain the depth increase rate, and the inter-frame difference in the number of pixels in the concave region is calculated to obtain the area expansion rate. A time series of shape evolution is constructed using the depth increase rate and area expansion rate. Linear regression is used to fit the relationship between the rate and time, and the evolution is determined to be accelerating or decelerating based on the regression coefficients, thus obtaining the shape evolution trend. Based on the evolution trend and historical pressure change rate data, the correlation coefficient between pressure change and shape rate is calculated, and a safety situation awareness model is constructed. The current pressure state, shape evolution trend and historical patterns are used as model inputs. The model predicts the depth and area values ​​for future periods. The pressure concentration degree for the corresponding period is estimated based on the correlation coefficient. The predicted depth and area are combined into a boundary deformation degree score.

[0058] Specifically, in one implementation, sampling of locally concentrated pressure distribution periods uses an equal-interval method, with the time interval set to 1 second to ensure the capture of rapid pressure changes. The pressure value sequence records the pressure intensity value at each sampling point, in Pascals (Pa). By calculating the difference between the pressure value at the current moment and the previous moment, and then dividing by the 1-second time interval, the pressure change rate is obtained, in Pascals per second (Pa / s). The video stream frame rate is synchronized with the pressure sampling frequency; each pressure sampling corresponds to the extraction of one frame of image. Boundary contour regions are segmented from the image, and the coordinate positions of all pixels on the contour lines are extracted.

[0059] Specifically, the extraction process of morphological feature data includes two steps: contour vertex detection and local curvature calculation. Vertex detection uses a corner detection algorithm to identify locations where curvature abruptly changes on the contour; these vertices reflect key turning points in the boundary shape. Local curvature values ​​are obtained by fitting an arc within the neighborhood of each contour point and calculating the reciprocal of the radius of the fitted circle. Combining the vertex coordinate sequence and the corresponding curvature value sequence constitutes a complete feature dataset describing the boundary morphology.

[0060] It should be noted that the roundness calculation is based on the geometric similarity assessment between the outline and a standard circle. Its value reflects the degree to which the outline approximates a circle, ranging from 0 to 1. The closer the value is to 1, the closer the outline is to a circle; the closer the value is to 0, the greater the deviation from a circle. In practical applications, 0.9 is usually used as the threshold. A roundness of 0.9 or greater is considered to indicate that the outline is close to a circle, suitable for quality inspection or shape classification; values ​​less than 0.9 require further analysis of shape deviation.

[0061] In one possible implementation, regularity is evaluated using the minimum bounding rectangle feature of the contour. A rotating caliper algorithm is used to find the smallest rectangle that completely contains the contour, and the aspect ratio of this rectangle is calculated. A preset standard aspect ratio is determined based on the typical layout of the site, typically set to 1.5. Regularity is calculated as the absolute value of the difference between the aspect ratio and the standard value; a larger difference indicates a more irregular shape. When the circularity of consecutive frames shows a decreasing trend while the regularity shows an increasing trend, the decision boundary is evolving from a regular circle or ellipse to an irregular polygon.

[0062] Preferably, measuring the depth increase rate requires accurately locating the deepest point of the depression. The point with the largest distance from each depression contour point to the depression boundary baseline is identified as the deepest point. The depth values ​​of the deepest points in two adjacent frames are recorded, the depth difference is calculated, and divided by the inter-frame time interval of 0.04 seconds to obtain the instantaneous depth increase rate. The area expansion rate is calculated by statistically analyzing the change in the number of pixels within the depression region. Converting the number of pixels to the actual area requires considering the spatial resolution of the image.

[0063] For example, the construction of the shape evolution time series uses the depth increase rate and area expansion rate as two-dimensional feature vectors, with one feature vector corresponding to each sampling time point, arranged in chronological order to form a sequence. Linear regression models are fitted to the depth rate sequence and the area rate sequence respectively, with time as the independent variable and the rate value as the dependent variable. Regression coefficients are calculated using the least squares method; a positive regression coefficient indicates an increasing rate, i.e., the evolution process is accelerating, while a negative regression coefficient indicates a decreasing rate, i.e., the evolution process is decelerating. The absolute value of the regression coefficient reflects the degree of acceleration or deceleration. Furthermore, the correlation coefficient between pressure change and shape rate is calculated using the Pearson correlation coefficient. The pressure change rate and shape evolution rate at the same time in historical data are paired, and the covariance of the two variables is calculated divided by the product of their respective standard deviations. The correlation coefficient ranges from -1 to 1; a positive value indicates that increased pressure leads to faster deformation, and a negative value indicates the opposite relationship. Based on the calculated correlation coefficient, a linear mapping relationship from pressure to deformation is established.

[0064] In one possible implementation, the security situation awareness model is constructed based on time series analysis and situation extrapolation methods. The model's input data includes three types of elements: first, the current pressure concentration distribution value, including pressure intensity values ​​at each location; second, the shape evolution rate sequence over a continuous period, including historical records of depth increase rate and area expansion rate; and third, the correlation coefficient between pressure change and shape evolution, reflecting the mutual influence between the two. The model's processing involves three stages: first, trend fitting is performed on the historical evolution rate sequence to extract evolution acceleration parameters and determine whether the current state is accelerating or trending towards stability; second, based on the correlation coefficient between pressure and shape, a mapping function from pressure to deformation is established, which describes the deformation increment corresponding to a certain increase in pressure; finally, using the current state as a starting point, and following the extracted evolution trend and mapping relationship, the model extrapolates forward time by time, calculating the deformation and pressure parameters for each future time period. The model's output includes two parts: first, the predicted depth and area values ​​for future time periods, used to quantify the degree of boundary deformation; and second, the predicted pressure concentration value for the corresponding time period, used to assess the risk level. Through this situational awareness prediction, the system can provide an early warning window before dangerous situations occur.

[0065] Understandably, future time-period predictions are based on the assumption of a continuation of the current evolutionary trend. Using the trend equation derived from linear regression, future times are substituted into the equation to calculate the predicted depth and area values. Through the established mapping relationship between pressure and deformation, the predicted deformation rate is converted into a corresponding estimate of pressure concentration. The boundary deformation degree score uses a weighted combination method, with a depth weight of 0.6 and an area weight of 0.4. The normalized depth and area prediction values ​​are weighted and summed to obtain a deformation degree score ranging from 0 to 100.

[0066] For example, during large gatherings, when the boundary morphology of a certain area is detected to be continuously deteriorating, the system can predict the pressure and deformation development trend within the next 5 minutes by analyzing historical evolution patterns, providing early warning information to managers and enabling early risk prevention and control.

[0067] The positional changes of boundary contour line segments and the shape contour of the concave region are obtained from consecutive frame images. The turning positions of the boundary contour and the advancement state of the concavity are collected at different times. The deformation process of the boundary contour from a smooth turn to a sharp turn is analyzed. The compression state of the surrounding area when the concavity advances inward is evaluated, and the change path of the boundary from the initial form to the current form is determined.

[0068] The pixel coordinate sequence of the boundary contour is extracted from consecutive frame images. The displacement vectors of corresponding contour points between adjacent frames are calculated to identify the inner and outer boundary lines of the concave region. The position of the turning point is marked by calculating the change in the angle between contour line segments. The spatial offset distance of the turning point and the advancing distance of the concave boundary towards the center are measured before and after the time interval, obtaining position change data containing offset and advancing amounts. Based on the turning point information in the position change data, the angle value of the adjacent line segments at each turning point is calculated. When the angle value changes from greater than a preset obtuse angle threshold to less than a preset acute angle threshold, it is determined that the point changes from a smooth turning point to a sharp turning point. The increase in the number of sharp turning points per unit time is counted. The minimum distance between the inner boundary of the concave region and the outer normal boundary is measured. If this distance continues to decrease, it is determined that the surrounding area is in a state of compression. Using the compression state and the turning point angle change sequence, the contour coordinate sequence at the initial time is used as a reference, and the contour coordinate sequence at the current time is used as a comparison target. The displacement trajectory between corresponding contour points is calculated, and the displacement vectors of each point are connected to form a deformation field, obtaining the change path of the boundary from the initial form to the current form.

[0069] Specifically, in one implementation, the pixel coordinate sequence of the boundary contour is extracted from each frame of the image using an edge detection algorithm, and each contour point contains horizontal and vertical coordinate information. Corresponding points between adjacent frames are determined through nearest neighbor matching, and the displacement vector is obtained by calculating the coordinate difference of the same contour point in two frames. The identification of concave regions is based on the concavity and convexity of the contour; the inner boundary line is defined as the contour segment with the deepest concavity, and the outer boundary line is the connecting line at the concavity opening.

[0070] Specifically, the marking of inflection points employs curvature analysis. At each contour point, the direction angle of adjacent line segments is calculated, and an inflection point is marked when the change in direction angle exceeds a preset threshold. Spatial offset distance is calculated using the Euclidean distance formula to determine the positional difference of the same inflection point at different times. The measurement of advancing distance uses the geometric center of the concave region as a reference point, calculating the change in distance from the boundary points to the center; a decrease in distance indicates inward advancement.

[0071] It should be noted that the obtuse angle threshold is set at 120 degrees, and the acute angle threshold is set at 60 degrees. These two thresholds are determined based on the typical deformation characteristics of the human body in crowded environments. When the included angle at a certain turning point changes from greater than 120 degrees to less than 60 degrees, it indicates that the contour at that point changes from a gentle transition to a sharp turn. The number of sharp turning points is counted in a 5-second time window, calculating the number of newly added sharp turning points within the window.

[0072] Preferably, the minimum distance is measured by uniformly sampling several points on the inner boundary of the depression, calculating the vertical distance from each sampling point to the outer normal boundary, and taking the minimum value of all distances. When the minimum distance of three consecutive frames shows a decreasing trend, it is determined that the surrounding area is in a state of continuous compression, which indicates a further increase in crowd density.

[0073] In one possible implementation, the deformation field is constructed by establishing a correspondence between initial contour points and current contour points. The specific process is as follows: First, a feature matching algorithm is used to establish the correspondence between the initial and current contour points. Second, the displacement vector from each initial point to its corresponding current point is calculated, forming a vector field. Then, radial basis function interpolation is used to fill the intermediate region, calculating displacement vectors at undefined positions to ensure the continuity of the deformation field. Finally, all vectors are arranged according to their spatial positions to form a continuous deformation field covering the entire boundary region, reflecting the deformation characteristics at each location. The displacement vector from each initial point to its corresponding current point constitutes a vector field, and the direction and magnitude of the vector at each position in the field reflect the deformation characteristics at that location. All displacement vectors are arranged according to their spatial positions, and interpolation is used to fill the intermediate region, forming a continuous deformation field covering the entire boundary region.

[0074] For example, the change path is obtained by connecting the contour shapes at key moments. Each node on the path represents an intermediate state of shape evolution, realizing the complete evolution process recording from the initial regular shape to the current irregular shape.

[0075] S106. By predicting the degree of pressure concentration and the degree of boundary deformation, high-risk areas are identified. By statistically analyzing the total force convergence at each depression in the high-risk areas, the locations where the total force convergence exceeds the preset warning standard are identified and marked as risk points.

[0076] The predicted pressure concentration and boundary deformation values ​​are compared with preset pressure and deformation thresholds, respectively. A high-risk area is only identified when both values ​​simultaneously exceed their respective thresholds. The inward converging pushing force values ​​at each depression within the high-risk area are obtained and accumulated to obtain the total force convergence. Based on the comparison between this total force convergence and a preset warning standard, when the total exceeds the warning standard, the spatial coordinates of the depression's center are recorded, and this location is marked as a risk point on the monitoring interface.

[0077] Specifically, in one implementation, the pressure threshold and deformation threshold are determined based on the site's carrying capacity and historical safety data. When both the pressure concentration and boundary deformation of a certain area exceed the threshold, the area is identified as a high-risk area. The pushing force values ​​at each depression within the high-risk area are extracted from the aforementioned pressure distribution data. The total force convergence is obtained by summing the pushing forces at all depressions within the area. A preset warning standard is set based on the site's safe bearing capacity. When the total force convergence exceeds this standard, it indicates a high safety risk at that location. The system automatically records the spatial coordinates of the depression center and marks it prominently on the monitoring interface, triggering an early warning mechanism to remind management personnel to take timely evacuation measures. In another implementation, the pressure threshold and deformation threshold are determined based on statistical data from historical stampede incidents. The pressure threshold is set at a pushing force of 500 Newtons per square meter, and the deformation threshold is set at a boundary irregularity exceeding 0.7. When both the pressure concentration and boundary deformation values ​​of a certain area exceed these two thresholds, the area is identified as a high-risk area.

[0078] Specifically, the pushing force is extracted from the pressure distribution data. The pushing force at each depression is obtained by integrating the pressure values ​​of all grid points within that region. The total force is calculated using a simple summation method, that is, by directly adding the pushing force values ​​of all depressions within the high-risk region, without considering the influence of regional weights or spatial distribution. The calculation process is as follows: first, the range of the high-risk region is determined; then, the pressure integral values ​​of each depression region are summed to obtain the final total.

[0079] It should be noted that the preset warning standard is determined based on the site's capacity and safety factor, and is usually set at 80% of the site's design bearing capacity. When the total force exceeds this standard, the system automatically records the center coordinates of the corresponding depression.

[0080] Preferably, risk points are marked with flashing red circles on the monitoring interface, with the size of the circle proportional to the degree of risk, and an audible and visual alarm is triggered to remind managers to take immediate evacuation measures.

[0081] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting urban public safety situational awareness, characterized in that, include: By acquiring crowd video streams through the monitoring system, images of the crowd gathering area are extracted, and boundary detection algorithms are used to separate the contour lines of the gathering boundary to obtain the boundary morphology including the spatial location of the contour lines. At the same time, the distance distribution and density distribution between individuals are extracted to obtain the basic data of crowd pressure gradient. Based on the boundary morphology, the local curvature of the contour line segment is identified, potential irregular points are marked, and their spatial coordinates are integrated to obtain the curvature change distribution. The concavity depth of potential irregular points is identified by the curvature change distribution, and the spatial volume of each concavity is determined by combining the concavity depth and the area of ​​the concavity region, thus obtaining the concavity spatial volume distribution. By identifying the direction of pushing force between individuals using the aforementioned crowd pressure gradient baseline data, and combining this with the volume distribution of the depression space to identify the number of people converging on each depression area and their movement direction, the depression intensity along the pushing force direction is assessed, resulting in a localized pressure concentration distribution, wherein: Based on the basic data of population pressure gradient, the pressure difference between adjacent individuals is calculated, converted into the magnitude of the pushing force, and the vector summation is used to obtain the direction and intensity of the combined pushing force at each location point. Identify the force vector pointing to the center of each depression area, count the number of individuals within the coverage area as the number of people gathered, and calculate the angle between the moving velocity vector and the line connecting the depression center based on the changes in the individual positions in consecutive frames. The summation of the individual pushing force intensity values ​​moving towards the depression, combined with the volume distribution of the depression space, is used to calculate the total pushing force per unit volume as the depression intensity value. A pressure recording lattice is established on the boundary contour plane. The pressure propagation attenuation coefficient is calculated based on the distance. The pressure contributions of all depression areas are superimposed to obtain the local pressure concentration distribution. The local pressure concentration distribution is analyzed over a continuous period of time. At the same time, the boundary morphology of the continuous frame images is compared to identify the rate of evolution of the boundary from a regular shape to an irregular shape. The increase in indentation depth and the expansion of indentation area per unit time are obtained to determine the trend of shape evolution and predict the degree of pressure concentration and boundary deformation in future periods. High-risk areas are determined by the degree of pressure concentration and the degree of boundary deformation. The total force convergence at each depression within the high-risk area is calculated, and locations where the total force convergence exceeds a preset warning standard are marked as risk points.

2. The urban public safety situational awareness and prediction method according to claim 1, characterized in that, The process involves acquiring crowd video streams through a monitoring system, extracting images of the crowd gathering area, using a boundary detection algorithm to separate the crowd boundary contour segments, obtaining the boundary morphology including the spatial location of the contour lines, and simultaneously extracting the distance distribution and density distribution between individuals to obtain basic data on the crowd pressure gradient, including: The video stream data from the monitoring system is acquired and decomposed frame by frame to extract image frames containing crowds; image processing technology is used to identify the boundary between the crowd area and the background to obtain the image of the area where the crowd is gathered. For the image of the region, a boundary detection algorithm is used to extract boundary pixels, connectivity analysis is performed to generate complete clustered boundary contour segments, the image coordinate values ​​of the contour points are recorded, and the actual spatial positions are converted through camera calibration parameters to obtain the spatial coordinate sequence of the boundary shape; Identify the location points of each individual within the area enclosed by the spatial coordinate sequence, calculate the distance between adjacent individual location points, count the number of individual pairs within different distance ranges, and form the distance distribution characteristics between individuals; The local density value is calculated based on the number of individuals per unit area and the average spacing. The local density value is then interpolated to generate a continuous density field. The gradient values ​​of the density field in each direction of space are calculated to obtain the basic data of the population pressure gradient.

3. The urban public safety situational awareness and prediction method according to claim 1, characterized in that, The step of identifying the local curvature of the contour line segment based on the boundary morphology, marking potential irregular points, and integrating their spatial coordinates to obtain the curvature variation distribution includes: Based on the spatial coordinate sequence of the boundary morphology, three consecutive coordinate points are selected to calculate the local curvature value, and the complete curvature sequence is obtained by sliding along the contour line. Perform a moving average calculation on the curvature sequence, mark the contour segments where the mean curvature exceeds the curvature threshold, and record the peak curvature points; The difference between the curvature peak point and the reference curvature is calculated, potential irregular points are identified and grouped and clustered to obtain the curvature variation distribution.

4. The urban public safety situational awareness and prediction method according to claim 1, characterized in that, The step of identifying the concavity depth of potential irregular points through the curvature change distribution, and determining the spatial volume of each concavity by combining the concavity depth and the area of ​​the concavity region, to obtain the concavity spatial volume distribution includes: Based on the curvature change distribution, identify the concave segments formed by continuous irregular points, connect the concave boundary points to form a reference baseline, calculate the vertical distance from each point on the concave contour to the reference baseline, and select the maximum vertical distance as the concave depth. A horizontal cutting surface is established according to the concave depth. The areas of adjacent cutting surfaces are added together to obtain the spatial volume value. The volume value of each concave location is recorded to form the spatial volume distribution of the concave area.

5. The urban public safety situational awareness and prediction method according to claim 1, characterized in that, The assessment of indentation strength in the direction of the pushing force also includes: Calculate the distance between the centers of gravity of adjacent individuals based on the direction of the pushing force between individuals, mark the high aggregation state, extract the temporal movement trajectory of people around the depression area, and identify body posture parameters. By comparing the position difference vectors and the spacing reduction rate at adjacent time points, the trajectory deflection pattern and body tilt direction sequence are determined; Tracing back along the reverse extension line of the body tilt direction to the peripheral low-density region, connecting them to form a force transmission path, and performing kernel density estimation to obtain the force application circuit network; The main convergence channels are determined by connecting the peak path density in the applied force line network.

6. The urban public safety situational awareness and prediction method according to claim 1, characterized in that, The analysis of continuous time-period changes in the localized pressure concentration distribution, combined with comparison of the boundary morphology in consecutive frame images, identifies the rate of boundary evolution from regular to irregular shapes, obtains the increase in depression depth and the expansion of depression area per unit time, determines the shape evolution trend, and predicts the degree of pressure concentration and boundary deformation in future time periods, including: The pressure value sequence is obtained by sampling the localized pressure concentration distribution, and the pressure change rate is calculated; Extract boundary contour morphological feature data at the corresponding time point, calculate roundness and regularity, and measure the rate of increase in depression depth and the rate of area expansion. Linear regression was used to fit the rate change relationship, and the shape evolution trend was obtained based on the regression coefficient. The depth, area and corresponding pressure concentration in the future period were predicted by the security situation awareness model.

7. The urban public safety situational awareness and prediction method according to claim 6, characterized in that, The rate at which the identification boundary evolves from a regular shape to an irregular shape also includes: Extract the boundary contour pixel coordinate sequence from consecutive frame images, calculate the displacement vector of the corresponding contour points in adjacent frames, and identify the inflection point position and the indentation advance distance; Calculate the change in the included angle between adjacent line segments at the turning point, determine whether a smooth turn becomes a sharp turn, and measure the minimum distance between the inner boundary of the depression and the outer normal boundary. Using the initial contour coordinate sequence as a reference, the displacement vectors of the current contour points are connected to form a deformation field, thus obtaining the boundary change path.

8. The urban public safety situational awareness and prediction method according to claim 1, characterized in that, The process involves determining high-risk areas based on the degree of pressure concentration and the degree of boundary deformation, calculating the total pressure convergence at each depression within the high-risk area, and marking locations where the total pressure convergence exceeds a preset warning standard as risk points. This includes: The pressure concentration degree and the boundary deformation degree are compared with the pressure threshold and deformation threshold, respectively. If both exceed the threshold, the area is identified as a high-risk area. The inward converging pushing force values ​​at each depression within the high-risk area are obtained and summed to obtain the total force convergence. When the total force convergence exceeds the preset warning standard, the spatial coordinates of the depression center are recorded and marked as a risk point.

Images