A campus safety early warning system based on video analysis

Abstract

The application relates to the technical field of campus safety early warning, in particular to a campus safety early warning system based on video analysis, which comprises a video feature extraction module, a behavior recognition module, a risk assessment module and an early warning response module. By acquiring the angle change of edge pixel points in a continuous monitoring video frame sequence, combining the light intensity distribution map with the environmental variable direction matching, generating behavior disturbance area mapping information, locking the high-risk behavior section, and capturing short-time state fluctuations through a neural network prediction model, disturbance influence early warning information is generated. The application can significantly improve the behavior recognition accuracy and early warning response speed in complex scenes, enhance the comprehensiveness and reliability of campus safety management, reduce resource waste, and improve the operation stability and economy.

Description

Technical Field

[0001] This invention relates to the field of campus safety early warning technology, and in particular to a campus safety early warning system based on video analysis. Background Technology

[0002] The field of campus safety early warning systems involves using video analytics to identify behaviors and provide risk warnings from real-time monitoring data within a campus, and is an important component of intelligent safety management. This field encompasses multiple aspects, including video stream acquisition, behavior analysis algorithm design, multi-source data fusion, early warning decision-making, and visual management. In practical applications, campus safety early warning systems are typically deployed in key areas such as teaching buildings, playgrounds, and corridors to achieve comprehensive monitoring and timely response to abnormal student behavior, dangerous events, and potential hazards. Key technological developments in this field include improving video analysis accuracy, enhancing adaptability to complex scenarios, optimizing multi-dimensional data fusion capabilities, and improving system interoperability, thereby providing more accurate, real-time, and comprehensive security for the campus. Among these, video analytics-based campus safety early warning systems are a technical solution that intelligently analyzes behavioral patterns in campus surveillance videos to predict and warn of potential security risks. Their main purpose is to improve the efficiency and accuracy of campus safety management and reduce security risks caused by untimely handling of emergencies. By comprehensively modeling and analyzing video data, environmental data, and other relevant information, this system can provide early risk warnings for campus safety management, adapt to dynamic needs in diverse scenarios, thereby assisting managers in formulating more scientific risk response strategies, reducing the probability of safety incidents, and improving overall emergency response capabilities.

[0003] The existing campus safety intelligent monitoring and early warning system and method (publication number CN116486586B) uses a data acquisition module to determine the identity of people entering the school and a data analysis module to predict their activity range and stay time, thereby implementing an early warning response mode. This technical solution improves the accuracy of campus safety early warnings. However, this system mainly relies on single behavioral trajectory analysis and fails to fully integrate multi-dimensional information such as meteorological data and equipment status for comprehensive judgment, which may lead to one-sidedness in the early warning results. In addition, the behavior recognition capability of this solution is limited in complex scenarios, which may affect the timeliness and comprehensiveness of early warnings in multi-person interaction scenarios or rapidly changing emergencies. Another campus behavior comprehensive early warning system (publication number CN116895128B) collects campus monitoring data through an information receiving module and uses a behavior analysis module and a hidden danger early warning module to comprehensively analyze and alarm students' abnormal behavior. This technical solution realizes intelligent monitoring and early warning management of student behavior. However, this system has certain shortcomings in the optimization of video analysis algorithms, especially in complex lighting conditions or under occlusion, where the accuracy of behavior recognition may decrease. Furthermore, the plan does not fully consider its interoperability with other campus security management systems (such as fire monitoring systems and access control systems), which may result in low efficiency in transmitting early warning information and affect the overall emergency response capability.

[0004] The aforementioned problems indicate that existing video analytics-based campus security early warning systems still have certain shortcomings in areas such as multi-dimensional data fusion, behavior recognition capabilities in complex scenarios, and system interoperability. Therefore, this invention aims to further improve the accuracy, real-time performance, and comprehensiveness of early warnings by introducing multi-source data fusion technology, optimizing video analytics algorithms, and enhancing system interoperability, thereby better meeting the needs of modern campus security management. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a campus security early warning system based on video analysis.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a campus security early warning system based on video analysis, the system comprising:

[0007] The video feature extraction module acquires the key region contour of each frame in a continuous monitoring video frame sequence, compares the angle change value of edge pixels with the environmental variable vector, determines whether the change of edge slope direction has consistent features, and generates dynamic behavior feature trend values.

[0008] The behavior recognition module calls the dynamic behavior feature trend value, combines it with the light intensity distribution map provided by the environmental monitoring equipment, and generates behavior disturbance area mapping information based on the spatial projection relationship between the distribution map coverage area and the behavior path.

[0009] The risk assessment module obtains the layout data of key areas within the campus based on the behavior disturbance area mapping information, makes spatial overlap judgments between behavior paths and risk areas, identifies high-risk behavior segments, and generates a set of coordinates for risk path overlay blocks.

[0010] The early warning response module calls the risk path overlay block coordinate set, obtains the number index of the coverage area at the campus level, marks the difference item of the output behavior change trend, and inputs the difference item as the external disturbance feature into the early warning input sequence to generate early warning information on the impact of disturbance on campus safety.

[0011] As a further aspect of the present invention, the dynamic behavioral feature trend values ​​include edge pixel rotation amplitude, contour slope change density, environmental variable consistency level, and boundary direction mutation identification marker; the behavioral disturbance region mapping information includes overlapping region principal axis angle distribution, disturbance region coverage breadth index, environmental variable classification level, and projection region number mapping table; the risk path superimposed block coordinate set specifically includes region projection junction boundary points, disturbance region and region arrangement correspondence, risk path block index, and spatial overlap density factor; the disturbance impact on campus safety early warning information includes external disturbance interference feature sequence, regional response trend symbol sequence, neural network prediction value set, and early warning segment index structure.

[0012] As a further aspect of the present invention, the video feature extraction module includes:

[0013] The image contour extraction submodule acquires the key region contour of each frame in a continuous monitoring video frame sequence, extracts the pixel boundary points with gray values ​​greater than a set threshold in each frame, establishes a contour point set, constructs a contour boundary line sequence based on the relative position of contour pixel values, and generates an image boundary line structure sequence.

[0014] The edge change calculation submodule calls the image boundary line structure sequence, extracts the coordinate values ​​of corresponding edge point pairs in the contour line sequence of adjacent frames, calculates the relative displacement vector, constructs an angle sequence set based on the angle between points, performs ratio calculation on the edge angle sequences of consecutive frames, calculates the mean value of edge change gradient, filters out segments whose angle changes exceed the change gradient threshold, and generates a set of edge direction change gradient values.

[0015] The direction consistency judgment submodule calls the edge direction change gradient value set, retrieves the average environmental variable vector recorded by the environmental monitoring equipment, performs projection calculation on the direction change gradient, compares the angle between the gradient and the direction of the environmental variable, filters the segments according to the set consistency threshold, obtains the proportion of segments whose edge change direction is consistent with the direction of the environmental variable, and obtains the dynamic behavior feature trend value.

[0016] As a further aspect of the present invention, the behavior recognition module includes:

[0017] The environmental variable direction extraction submodule calls the dynamic behavior feature trend value to obtain the light intensity distribution map recorded by the environmental monitoring equipment, identifies the boundary of the equal intensity area in the distribution map, determines the environmental variable distribution contour based on the set of iso-contour lines, extracts the boundary line direction distribution and performs vector fitting according to the position order to generate the environmental variable principal axis direction sequence.

[0018] The angle matching calculation submodule obtains the angle relationship between the vector segment and the edge slope direction segment marked in the trend value of dynamic behavior features based on the main axis direction sequence of the environmental variables. It compares the angle between the two sets of direction vectors in the spatial coordinate plane, filters vector pairs with angle deviation values ​​lower than the perturbation angle set threshold, calculates and obtains the regional direction deviation index value, and takes the block with deviation value lower than the perturbation angle threshold as the matching region to generate a direction matching perturbation distribution value set.

[0019] The spatial intersection construction submodule matches the disturbance distribution value set according to the direction, retrieves the behavior path information within the time period, establishes a path vector projection layer in a unified spatial coordinate system, locates the intersection block of the projection boundary between the behavior path and the marked disturbance area, extracts the boundary line index value of the overlapping block and generates a spatial geometric coverage group, and establishes behavior disturbance area mapping information.

[0020] As a further aspect of the present invention, the risk assessment module includes:

[0021] The regional layout extraction submodule obtains the behavior disturbance region mapping information, collects key regional layout data within the campus, extracts the two-dimensional spatial location index corresponding to the regional number grid, spatially locates the arrangement of rows and columns and numbering order of the regions in the actual site, and generates a set of regional spatial locations by combining the regional orientation angle value.

[0022] The disturbance path mapping submodule calls the set of spatial locations in the region, collects the distribution of environmental variables, the measured value of light intensity, and the azimuth angle of the surface of the region. Based on the path direction vector, it calculates the behavior path direction corresponding to the receiving surface of the region, projects the path direction onto the plane coordinates, determines whether there is an overlap with the coverage of the behavior disturbance area mapping information, obtains the spatial path segment that coincides with the behavior and the disturbance area, and generates path disturbance overlap distribution data.

[0023] The cross-block identification submodule extracts the set of coordinate points corresponding to the occluded segment based on the path disturbance overlap distribution data, identifies the numbered region to which the coordinate points belong in the regional arrangement space, performs cross-judgment between the regional boundary and the occluded path boundary, extracts the regional number index corresponding to the intersection point, determines the regional number and corresponding coordinate region of the high-risk behavior segment, and establishes a risk path superimposed block coordinate set.

[0024] As a further aspect of the present invention, the early warning response module includes:

[0025] The regional signal acquisition submodule calls the risk path overlay block coordinate set, extracts the corresponding regional number index at the campus level, collects the unit regional output status value and environmental variable count value of the regional monitoring record, constructs the operating status data group of each region at the current moment, and generates a set of regional operating signals;

[0026] The trend difference marking submodule obtains the state output values ​​of the region within the disturbance zone and the adjacent non-disturbance zone at the same time point based on the set of regional operation signals, compares the differences in the direction of change, identifies signal items with inconsistent directions, records the difference items in the form of symbol values, classifies and organizes them into a separate input factor sequence, and obtains a state trend difference vector group.

[0027] The disturbance information generation submodule calls the state trend difference vector group, inputs the neural network input sequence structure trained with historical periodic samples, inputs the vector group and the constructed input factor terms in parallel, and performs combined inference based on environmental variables, light intensity and behavioral path data input factors to obtain the predicted value distribution results at the corresponding time point, and generates early warning information on the impact of disturbances on campus safety.

[0028] As a further aspect of the present invention, the system further includes:

[0029] The trusted label assignment module calls the disturbance impact on campus safety early warning information, obtains the predicted value and the actual output value of the area at each prediction time point, compares the root mean square error between the predicted value and the actual output value, and compares it with the set confidence error threshold. The error segment above the threshold is divided into intervals according to the level division rules, and each level division segment is marked as a trusted level identifier to generate the trusted level information of the disturbance coverage area.

[0030] The perturbation coverage area output credibility level information specifically refers to the error interval level index, credibility label mapping result, RMSE fluctuation trend group, and credibility judgment result number table.

[0031] As a further aspect of the present invention, the trusted tag assignment module includes:

[0032] The error value calculation submodule calls the disturbance impact on campus safety early warning information, obtains the predicted output state sequence and the actual output state sequence of each region within the prediction time period, constructs the error difference set between the corresponding sequence indices, and combines the prediction state to normalize the error ratio and calculate the normalized state error index.

[0033] The error segment division submodule calls the set confidence error threshold according to the normalized state error index, groups and classifies the index sequence according to the upper limit of the threshold, establishes index labels for the time period corresponding to the group, and sequentially numbers the error level corresponding to the group to obtain the error level mapping interval value.

[0034] The credibility level labeling submodule calls the error level mapping interval value, establishes the credibility level interval correspondence rule according to the time mapping table of the level interval and the region number, outputs the credibility label result set corresponding to the region in each time period, and establishes the perturbation coverage area to output credibility level information.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0036] In this invention, by acquiring the angle changes of edge pixels in a continuous monitoring video frame sequence and calculating the consistency characteristics of edge slope changes with the direction of environmental variables, the dynamic evolution of behavioral paths can be accurately identified, significantly improving the accuracy of behavior capture in complex scenes. By combining the direction of environmental variables in the illumination intensity distribution map for angle matching and spatial projection, the actual spatial location and range of the behavioral path disturbance can be clarified, achieving refined identification of the behavioral disturbance area. Based on the spatial mapping and cross-calculation of behavioral paths and regional layout, high-risk behavioral segments are specifically locked, and the disturbance location and region number are accurately located. The state fluctuations of specific region groups are clearly predicted, making the prediction results highly spatially specific. By establishing a neural network prediction model of disturbance impact, behavioral disturbance features are directly incorporated into the prediction input, effectively capturing subtle changes in short-term state fluctuations, greatly improving the accuracy of prediction response. On the basis of the early warning results, a credibility assessment is introduced. By labeling the root mean square error level, the degree of prediction deviation is effectively distinguished, improving the reliability of the early warning data in practical applications. This enables the campus safety management center to accurately adjust potential risks, reduce response errors, reduce resource waste, and enhance the operational stability and economy of the campus safety system. Attached Figure Description

[0037] Figure 1 This is a system flowchart of the present invention;

[0038] Figure 2 This is a flowchart illustrating the acquisition process of the video feature extraction module of the present invention.

[0039] Figure 3 This is a flowchart illustrating the acquisition process of the behavior recognition module of the present invention.

[0040] Figure 4 This is a flowchart illustrating the acquisition process of the risk assessment module in this invention.

[0041] Figure 5 This is a flowchart illustrating the acquisition process of the early warning response module of the present invention.

[0042] Figure 6 This is a flowchart illustrating the acquisition process of the trusted tag assignment module of this invention. Detailed Implementation

[0043] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0044] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0045] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0046] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0047] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0048] Please see Figure 1 This invention provides a technical solution: a campus security early warning system based on video analysis, the system comprising:

[0049] The video feature extraction module acquires the key region contour of each frame in a continuous monitoring video frame sequence, compares the angle change value of edge pixels with the environmental variable vector, determines whether the change of edge slope direction has consistent features, and generates dynamic behavior feature trend values.

[0050] The behavior recognition module calls dynamic behavior feature trend values, combines them with the light intensity distribution map provided by the environmental monitoring equipment, and generates behavior disturbance area mapping information based on the spatial projection relationship between the distribution map coverage area and the behavior path.

[0051] The risk assessment module obtains layout data of key areas on campus based on the behavioral disturbance area mapping information, makes spatial overlap judgments between behavioral paths and risk areas, identifies high-risk behavioral segments, and generates a set of coordinates of risk path superimposed blocks.

[0052] The early warning response module calls the risk path overlay block coordinate set, obtains the number index of the coverage area at the campus level, marks the difference items of the output behavior change trend, and inputs the difference items as external disturbance features into the early warning input sequence to generate early warning information on the impact of disturbance on campus safety.

[0053] The trusted label assignment module calls the early warning information on the impact of disturbances on campus safety, obtains the predicted value and the actual output value of the area at each prediction time point, compares the root mean square error between the predicted value and the actual output value, and compares it with the set confidence error threshold. The error segment that exceeds the threshold is divided into intervals according to the level division rules, and each level division segment is marked as a trusted level identifier, generating trusted level information for the disturbance coverage area.

[0054] The dynamic behavioral characteristic trend values ​​include edge pixel rotation amplitude, contour slope change density, environmental variable consistency level, and boundary direction mutation identification marker. The behavioral disturbance area mapping information includes the distribution of the principal axis angle of the overlapping area, the disturbance area coverage breadth index, environmental variable classification level, and projection area number mapping table. The risk path superimposed block coordinate set specifically includes the regional projection junction boundary point, the correspondence between the disturbance area and the region arrangement, the risk path block index, and the spatial overlap density factor. The disturbance impact on campus safety early warning information includes the external disturbance interference feature sequence, the regional response trend symbol sequence, the neural network prediction value set, and the early warning segment index structure. The disturbance coverage area output credibility level information specifically refers to the error interval level index, credibility label mapping results, RMSE fluctuation trend group, and credibility judgment result number table.

[0055] Please see Figure 2 The video feature extraction module includes:

[0056] The image contour extraction submodule acquires the key region contour of each frame in a continuous monitoring video frame sequence, extracts the pixel boundary points with gray values ​​greater than a set threshold in each frame, establishes a contour point set, constructs a contour boundary line sequence based on the relative position of contour pixel values, and generates an image boundary line structure sequence.

[0057] First, two consecutive frames are captured from the video stream at a specific time interval, i.e., the t-th frame. With the image of frame t+1 The image resolution was set to 1920x1080 pixels, and the acquisition time interval was 1 / 30 second. The two frames were converted from the RGB color space to the grayscale space to obtain grayscale images. and Set a pixel grayscale value threshold The threshold The settings were based on statistical analysis of 1000 sets of campus surveillance sample images under different lighting conditions. The grayscale values ​​that could effectively separate moving human bodies from static backgrounds (such as the ground and walls) in 95% of the samples were selected. The specific calculation process was as follows: the sample images were processed using the Otsu method to obtain an initial threshold; the average of all initial thresholds was calculated and multiplied by a correction factor of 1.1; and the final setting was... For 135, iterate through... Each pixel in When its grayscale value If the value is greater than 135, it is identified as a key region pixel, and its eight neighboring pixels are further examined. If at least one neighboring pixel has a grayscale value not greater than 135, then the pixel is... If a point is identified as a contour point, then all such contour points constitute the set of contour points. For example, in In the set (500, 400), the gray value of pixel (500, 400) is 150. Since there is a pixel with a gray value of 110 in its eight neighborhood, (500, 400) is added to the set. According to the set The proximity relationship between each contour point in the two-dimensional spatial coordinate system, that is, any two points and Euclidean distance Less than the preset connection distance threshold When the value is set to 3 pixels, connect these two points. By traversing and connecting all pairs of points that meet the conditions, a series of non-intersecting contour boundary lines representing the outline of the key area are constructed. Generate a sequence of image boundary line structures.

[0058] The edge change calculation submodule calls the image boundary line structure sequence, extracts the coordinate values ​​of corresponding edge point pairs in the contour line sequence of adjacent frames, calculates the relative displacement vector, constructs an angle sequence set based on the angle between points, performs ratio calculation on the edge angle sequences of consecutive frames, calculates the mean value of edge change gradient, filters out segments whose angle changes exceed the change gradient threshold, and generates a set of edge direction change gradient values.

[0059] Calling the image boundary line structure sequence and Firstly, targeting Each contour point in ,exist Find the closest corresponding point in space. The search process is achieved through calculation. and This is achieved by finding the minimum Euclidean distance between all points, thus obtaining the edge point pairs. And calculate the relative displacement vector of the point pair. For example, if Its corresponding point in the next frame Then the relative displacement vector is Next, in the sequence of outline lines Above, take one point every 5 points, and form a group of three points with one point before and one point after it, for example, point... Based on these three points, construct two vectors. Calculate the angle between these two vectors. All these included angles constitute a set of angle sequences. Processed in the same way Obtain the set of angle sequences Then, the ratio of the angles corresponding to the spatial positions in the angle sequence sets of these two consecutive frames is calculated. This yields a sequence of ratios, and the arithmetic mean of this sequence is calculated, which is the mean of the marginal change gradient. For example, if three consecutive sets of corresponding angle values ​​are (120°, 122°), (95°, 94°), and (88°, 90°), then the ratios are 1.017, 0.989, and 1.023, respectively, with a mean of 1.010. A gradient threshold is then set. The threshold This analysis was based on 100 video clips of 10 common campus behaviors, including walking and running, and statistical analysis was conducted. The values ​​distributed within the range of [0.95, 1.05] accounted for 99%. To identify abnormal, violent movements, the threshold was set... Set to 1.15 and filter out the ratios. Segments greater than 1.15 or less than 1 / 1.15, for example Then the angle of this segment and its corresponding contour point coordinates and displacement vector Record it to generate a set of gradient values ​​for edge direction changes.

[0060] The direction consistency judgment submodule calls the edge direction change gradient value set, retrieves the average environmental variable vector recorded by the environmental monitoring equipment, performs projection calculation on the direction change gradient, compares the angle between it and the direction of the environmental variable, filters the segments according to the set consistency threshold, obtains the proportion of segments whose edge change direction is consistent with the direction of the environmental variable, and obtains the dynamic behavior characteristic trend value.

[0061] The gradient value set of edge direction changes contains a series of records consisting of coordinates, angles, and displacement vectors. Simultaneously, the average environmental variable vector recorded by environmental monitoring equipment linked to the field of view of the surveillance camera is retrieved. This vector represents the directionality of the main static environmental features within the field of view, such as the shadow boundary formed by light entering from the upper left corner, or the orientation of walls. This vector is obtained by performing principal component analysis (PCA) on the light intensity distribution map during a continuous 10-minute period without dynamic targets. The direction of its first principal component is taken as the principal axis direction of the environmental variable. It is assumed that this is obtained through calculation... Next, for each displacement vector in the set of gradient values ​​changing along the edge direction... (For example, (2,1) mentioned above) is normalized to obtain and with The projection operation involves calculating the dot product of the two vectors and then dividing by the product. The modulus length is used, but more directly, the angle comparison method is used here to calculate... and The angle between Set a consistency threshold The threshold is set based on the fact that when the direction of behavior is consistent with the direction of environmental features (such as walking along a wall), the included angle is usually small. This was determined by statistically analyzing 50 such cases. In 95% of cases, the temperature is below 20 degrees Celsius, therefore... Set the angle to 20 degrees, iterate through all displacement vectors in the set, and determine the angle for each vector. Count the number of vectors that meet the condition, whether the angle is less than 20 degrees. and the total number of vectors in the set. For example, if there are 50 vectors in the gradient value set, and calculations show that 35 vectors have an angle less than 20 degrees, then the percentage of calculated vectors is... This yields the trend values ​​of dynamic behavioral characteristics.

[0062] Please see Figure 3 The behavior recognition module includes:

[0063] The environmental variable direction extraction submodule calls the dynamic behavior feature trend value to obtain the light intensity distribution map recorded by the environmental monitoring equipment, identifies the boundary of the equal intensity area in the distribution map, determines the environmental variable distribution contour based on the set of iso-contour lines, extracts the boundary line direction distribution and performs vector fitting according to the position order to generate the environmental variable principal axis direction sequence;

[0064] The algorithm retrieves a dynamic behavior characteristic trend value of 0.7 and obtains a light intensity distribution map composed of real-time data uploaded by multiple light sensors deployed on campus. This map is presented in the form of a heatmap, where each pixel represents a light intensity value in lux. First, a contour line generation algorithm is applied to this distribution map to identify the boundaries of areas with equal light intensity; for example, the contour line formed by all points with an intensity of 500 lux is identified. The outline formed by points with an intensity of 600 lux These isopleths form the distribution profile of environmental variables; the longest of these is selected. (For example Along its direction, a tangent direction is extracted every 10 pixels, and these tangent directions are vectorized. For example, at the contour point (x, y), the tangent direction vector is... By analyzing all extracted tangent direction vectors By performing a weighted average (with weights determined by the length of the contour segment) or principal component analysis, a straight line representing the main trend of the environmental variable is fitted. This straight line is the principal axis direction of the environmental variable. This line is then sequentially divided into multiple vector segments to generate a sequence of principal axis directions for the environmental variable. .

[0065] The angle matching calculation submodule obtains the angle relationship between the vector segment and the edge slope direction segment marked in the trend value of dynamic behavior features based on the main axis direction sequence of environmental variables. It compares the angle between the two sets of direction vectors in the spatial coordinate plane, filters vector pairs with angle deviation values ​​lower than the perturbation angle set threshold, calculates and obtains the regional direction deviation index value, and takes the block with deviation value lower than the perturbation angle threshold as the matching region to generate a set of direction matching perturbation distribution values.

[0066] Obtain each vector segment in the sequence of principal axis directions of environment variables The angle relationship between the edge slope direction segment and the trend value of the dynamic behavior feature in the previous step is specifically defined by the displacement vectors in the set of edge direction change gradient values ​​that are judged to be consistent with the direction of the environmental variable. The directions they represent are compared sequentially with the vectors in the principal axis direction sequence within a unified campus planar spatial coordinate system. With each displacement vector that satisfies the consistency condition The angle between Set a perturbation angle threshold. The threshold was set based on the following considerations: the interaction between behavior and environmental features (such as stepping on the boundary of light and shadow) will produce highly correlated movement directions. Through the analysis of 200 such interaction samples, the vector angle... In 95% of cases, the temperature is less than 10 degrees, therefore it is set as follows: Set the temperature to 10 degrees and filter out all. Degree vector pair And for each main axis direction segment Calculate its matching The number of vector pairs is used as an indicator of the degree of directional deviation in the region, and all vector pairs with at least one matching vector pair are included. Main axis direction section The covered geographic blocks are marked as matching regions, and a set of directional matching perturbation distribution values ​​is generated.

[0067] The spatial intersection construction submodule matches the perturbation distribution value set according to the direction, retrieves the behavior path information within the time period, establishes a path vector projection layer in a unified spatial coordinate system, locates the intersection block of the projection boundary between the behavior path and the marked perturbation area, extracts the boundary line index value of the overlapping block and generates a spatial geometric coverage group, and establishes behavior perturbation area mapping information.

[0068] The direction matching perturbation distribution value set marks a series of matching areas on the campus map. At the same time, it retrieves the behavioral path information recorded by continuously locating the centroid of dynamic targets within the same time period. This path information is a series of coordinate points ordered by time. In a unified two-dimensional spatial coordinate system with geographic coordinate registration (such as a GIS map covering the entire campus), the behavior path is... Render it as a vector projection layer, and render each matching region marked by the direction matching perturbation distribution value set as another polygon layer. Using GIS spatial analysis tools, locate the blocks where the behavior path layer and the perturbation area layer spatially overlap, that is, the blocks where the path line segments and the boundaries of the perturbation area polygons intersect. Extract the unique identifiers of the boundary lines of these overlapping blocks in the GIS layer as index values, and combine these index values ​​and the corresponding overlapping geographic range coordinates to generate a spatial geometric overlay group containing multiple geometric objects and their attributes, and establish behavior perturbation area mapping information.

[0069] Please see Figure 4 The risk assessment module includes:

[0070] The regional layout extraction submodule obtains behavioral disturbance area mapping information, collects layout data of key areas within the campus, extracts the two-dimensional spatial location index corresponding to the area number grid, spatially locates the arrangement of rows and columns and numbering order of the areas in the actual site, and generates a set of area spatial locations by combining the area orientation angle value.

[0071] The behavioral disturbance area mapping information is essentially one or more specific geographic polygon areas marked on the campus map. At the same time, the layout data of key areas within the campus are collected from the campus management database, as shown in Table 1.

[0072] Table 1: Layout Data of Key Areas on Campus

[0073]

[0074] As shown in Table 1, this data table details the attributes of each key area. First, the number grid of each area, such as B-03, and its corresponding two-dimensional spatial location index, i.e., the center point coordinates (310, 620), are extracted. Based on these coordinates, the area is accurately located on the campus master plan. The spatial location is then determined by combining the arrangement of the areas (e.g., room number 3 in area B) and the numbering order. The orientation angle value of the area is also read. For example, the orientation angle of area B-03 is 180 degrees, which means that its entrance or main monitoring face faces due south. This information (number, coordinates, risk level, orientation angle) is integrated and processed to generate a structured set of regional spatial locations for subsequent modules to call.

[0075] The disturbance path mapping submodule calls the regional spatial location set, collects the distribution of environmental variables, light intensity measurement values, and the azimuth angle of the regional surface orientation, calculates the behavior path direction corresponding to the receiving surface of the region based on the path direction vector, projects the path direction onto the plane coordinates, determines whether there is an overlap with the coverage area of ​​the behavior disturbance area mapping information, obtains the spatial path segment that coincides with the behavior and the disturbance area, and generates path disturbance overlap distribution data.

[0076] The system retrieves a set of regional spatial locations and collects the distribution of environmental variables (such as light intensity maps), light intensity measurements, and surface orientation azimuth angles of each region, synchronized with the time of the behavior. For example, for region B-03, its orientation angle is 180 degrees. The current behavior path information displays the behavior path vector within a specific time period. The direction is (0.95, -0.31), which is southeast, according to the path direction vector. Surface normal vector of region B-03 (Given the 180-degree orientation angle, the normal vector direction is (0, -1)). Perform a dot product operation to determine whether the behavior is facing or moving away from the monitored area, and then set the behavior path. Project the path onto the plane coordinate system of the campus and determine whether there is spatial overlap between the coverage of the projected path and the coverage of the behavior disturbance area mapping information generated in the previous step. For example, the GIS system analysis shows that the path from coordinates (300,610) to (320,595) belongs to both the behavior path and the behavior disturbance area mapping information. Therefore, this spatial path segment is identified to obtain the spatial path segment where the behavior and the disturbance area overlap, and path disturbance overlap distribution data is generated.

[0077] The cross-block identification submodule extracts the set of coordinate points corresponding to the occluded segment based on the path disturbance overlap distribution data, identifies the numbered area to which the coordinate points belong in the area arrangement space, performs cross-judgment between the area boundary and the occluded path boundary, extracts the area number index corresponding to the intersection point, determines the area number and corresponding coordinate area of ​​the high-risk behavior segment, and establishes the risk path superimposed block coordinate set.

[0078] The overlapping distribution data of path disturbances clearly identifies the specific path segments that coincide with the disturbance behavior. The set of start and end coordinates of these segments is extracted, for example... The system identifies which region these coordinate points belong to within the spatial location set. Through coordinate matching, it is found that the path segment falls entirely within the 5-meter influence range of region B-03 (center point (310, 620)). Spatial intersection analysis is performed between the boundary polygon of region B-03 and the obstructed path segment to extract the coordinates of the intersection point. The region's index "B-03" is then extracted, thus determining the high-risk behavior zone as region B-03, and its corresponding coordinate region as the path segment. The scanned area is used to establish a set of coordinates for the risk path overlay blocks.

[0079] Please see Figure 5 The early warning response module includes:

[0080] The regional signal acquisition submodule calls the risk path overlay block coordinate set, extracts the corresponding regional number index at the campus level, collects the unit regional output status value and environmental variable count value of the regional monitoring record, constructs the operating status data group of each region at the current moment, and generates a set of regional operating signals.

[0081] The system calls upon the risk path overlay block coordinate set, which includes the high-risk area number "B-03" and its corresponding coordinate range. At the regional level of the campus security system, using the index "B-03", a data acquisition command is sent to the monitoring equipment or sensors (such as access control status sensors and infrared sensors) deployed in that area. This collects the unit area output status values ​​of that area at the current time and several previous time points. For example, the access control sensor in area B-03 has a status value of 0 (closed) at time t, and its values ​​are also 0 at times t-1 and t-2. At the same time, the system collects the environmental variable count values ​​of that area, such as light intensity of 350 lux and temperature of 22℃. These data are combined into a data record to construct the operating status data group of each area at the current time, such as {time: t, number: "B-03", access control status: 0, light intensity: 350, temperature: 22}, generating a set of area operating signals.

[0082] The trend difference marking submodule obtains the state output values ​​of the area within the disturbance zone and the adjacent non-disturbance zone at the same time point based on the set of regional operation signals, compares the differences in the direction of change, identifies signal items with inconsistent directions, records the difference items in the form of symbol values, classifies and organizes them into a separate input factor sequence, and obtains a state trend difference vector group.

[0083] Based on the set of regional operation signals, the state output values ​​of region "B-03" in the disturbance zone and the spatially adjacent non-disturbance zone region (e.g., "A-01") at the same time point t are obtained, as shown in Table 2.

[0084] Table 2: Comparison of Regional Status Trends

[0085]

[0086] As shown in Table 2, comparing the state values ​​of regions B-03 and A-01 in the time series (t-2, t-1, t), the state value sequence of region B-03 is (0, 0, 0), with a stable trend, while the pedestrian flow count sequence of region A-01 is (5, 5, 4), with a decreasing trend. Although the state of B-03 itself has not changed, its associated behavior (rapid approach) differs from the normal state trend of the surrounding areas (stable or decreasing pedestrian flow). This inconsistency caused by potential external disturbances is identified, and the difference is recorded in the form of a specific symbol value. For example, the difference vector element is defined as follows: 0 for the same trend, 1 for the opposite trend, and -1 for no trend. Here, the dynamic behavior trend (approach) of region B-03 is unrelated to the pedestrian flow trend (decreasing) of region A-01, and is marked as -1. All such difference items are classified and organized into a single input factor sequence to obtain the state trend difference vector group.

[0087] The disturbance information generation submodule calls the state trend difference vector group, inputs the neural network input sequence structure trained with historical periodic samples, inputs the vector group and the constructed input factor terms in parallel, and performs combination inference based on environmental variables, light intensity and behavioral path data input factors to obtain the predicted value distribution results at the corresponding time point and generate early warning information on the impact of disturbance on campus safety.

[0088] The state trend difference vector group is added as a new feature input to an early warning neural network model that has been trained with historical periodic samples (5000 hours of monitoring data including normal and abnormal events). In addition to the original regional state data, the input sequence structure of this neural network now includes this difference vector. This vector group, for example [-1], along with other existing input factors, such as environmental variables (light intensity 350 lux) and behavioral path data (path vector (0.95, -0.31), speed 5 m / s), is used as the input at the current time point and input into the input layer of the neural network. The network model extrapolates these combined factors based on its internal weights and activation functions to predict the possible changes in the state value of the region (B-03) at a very short time point in the future (e.g., t+1, t+2, t+3 seconds). For example, the model predicts that the probability of the access control status changing to 1 (open) at t+3 seconds is 0.85. The distribution results of this series of predicted values ​​are obtained to generate early warning information on disturbances affecting campus safety.

[0089] Please see Figure 6 The trusted tag assignment module includes:

[0090] The error value calculation submodule calls the early warning information on the impact of disturbances on campus safety, obtains the predicted output state sequence and the actual output state sequence of each region within the prediction time period, constructs the error difference set between the corresponding sequence indices, and combines the prediction state to normalize the error ratio and calculate the normalized state error index.

[0091] The early warning information regarding disturbances affecting campus safety includes a sequence of predicted access control status values ​​for area B-03 at three future time points (t+1, t+2, t+3), with predicted probabilities of {0.2, 0.5, 0.85}. To calculate the error, predicted values ​​with probabilities greater than 0.5 are quantized as 1, and those with probabilities less than 0 are quantized as 0, resulting in a predicted status sequence of {0, 1, 1}. Simultaneously, the actual output status sequences of the access control sensors in area B-03 are obtained at the actual times t+1, t+2, and t+3, assuming the actual sequences are {0, 0, 1}. A set of error differences between corresponding sequence indices is constructed, i.e. The error is proportionally normalized based on the predicted state; here, the root mean square error (RMSE) is used directly for calculation. The normalized state error index is obtained through calculation.

[0092] The error segment division submodule calls the set confidence error threshold based on the normalized state error index, groups and classifies the index sequence according to the upper limit of the threshold, establishes index labels for the time period corresponding to the group, and sequentially numbers the error level corresponding to the group to obtain the error level mapping interval value.

[0093] Based on the normalized state error index of 0.577, a pre-defined multi-level confidence error threshold is applied. This threshold is set based on statistical analysis of the prediction errors of 1000 historical warning events. The error values ​​are sorted from smallest to largest and divided according to quartiles, defining three thresholds: (Low error, high confidence level) (Medium error, medium confidence level) (High error, low confidence) The calculated error index of 0.577 is compared with these upper thresholds. 0.577 is greater than 0.5 and less than 0.8, therefore the error is classified as... and Within the defined interval, a time period index label is established for this group, namely the prediction time from t+1 to t+3. The error level corresponding to this group is numbered. For example, [0,0.2] is level 1, (0.2,0.5] is level 2, (0.5,0.8] is level 3, (0.8,~] is level 4. The current error of 0.577 belongs to level 3, and the error level mapping interval value is 3.

[0094] The credibility level labeling submodule calls the error level mapping interval value, establishes the credibility level interval correspondence rule based on the time mapping table of the level interval and the region number, outputs the credibility label result set corresponding to the region in each time period, and establishes the perturbation coverage area to output credibility level information.

[0095] Call the error level mapping interval value "3", and according to the pre-established mapping table of error level with region and time, as shown in Table 3.

[0096] Table 3: Trust Level Interval Mapping Table

[0097]

[0098] As shown in Table 3, a rule for corresponding confidence level intervals is established, mapping the error level "3" to the confidence level identifier "low confidence". The confidence label result corresponding to the warning information of output area B-03 in the prediction time period t+1 to t+3 is "low confidence". This result is combined with the area number and time period information to form a complete record and establish the output confidence level information of the disturbance coverage area.

[0099] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A campus security early warning system based on video analytics, characterized in that, The system includes: The video feature extraction module acquires the key region contour of each frame in a continuous monitoring video frame sequence, compares the angle change value of edge pixels with the environmental variable vector, determines whether the change of edge slope direction has consistent features, and generates dynamic behavior feature trend values. The video feature extraction module includes: The image contour extraction submodule acquires the key region contour of each frame in a continuous monitoring video frame sequence, extracts the pixel boundary points with gray values ​​greater than a set threshold in each frame, establishes a contour point set, constructs a contour boundary line sequence based on the relative position of contour pixel values, and generates an image boundary line structure sequence. The edge change calculation submodule calls the image boundary line structure sequence, extracts the coordinate values ​​of corresponding edge point pairs in the contour line sequence of adjacent frames, calculates the relative displacement vector, constructs an angle sequence set based on the angle between points, performs ratio calculation on the edge angle sequences of consecutive frames, calculates the mean value of edge change gradient, filters out segments whose angle changes exceed the change gradient threshold, and generates a set of edge direction change gradient values. The direction consistency judgment submodule calls the edge direction change gradient value set, retrieves the average environmental variable vector recorded by the environmental monitoring equipment, performs projection calculation on the direction change gradient, compares the angle between it and the direction of the environmental variable, performs segment filtering based on the set consistency threshold, obtains the proportion of segments whose edge change direction is consistent with the direction of the environmental variable, and obtains the dynamic behavior characteristic trend value. The behavior recognition module calls the dynamic behavior feature trend value, combines it with the light intensity distribution map provided by the environmental monitoring equipment, and generates behavior disturbance area mapping information based on the spatial projection relationship between the distribution map coverage area and the behavior path. The behavior recognition module includes: The environmental variable direction extraction submodule calls the dynamic behavior feature trend value to obtain the light intensity distribution map recorded by the environmental monitoring equipment, identifies the boundary of the equal intensity area in the distribution map, determines the environmental variable distribution contour based on the set of iso-contour lines, extracts the boundary line direction distribution and performs vector fitting according to the position order to generate the environmental variable principal axis direction sequence. The angle matching calculation submodule obtains the angle relationship between the vector segment and the edge slope direction segment marked in the trend value of dynamic behavior features based on the main axis direction sequence of the environmental variables. It compares the angle between the two sets of direction vectors in the spatial coordinate plane, filters vector pairs with angle deviation values ​​lower than the perturbation angle set threshold, calculates and obtains the regional direction deviation index value, and takes the block with deviation value lower than the perturbation angle threshold as the matching region to generate a direction matching perturbation distribution value set. The spatial intersection construction submodule matches the disturbance distribution value set according to the direction, retrieves the behavior path information within the time period, establishes a path vector projection layer in a unified spatial coordinate system, locates the intersection block of the projection boundary between the behavior path and the marked disturbance area, extracts the boundary line index value of the overlapping block and generates a spatial geometric coverage group, and establishes behavior disturbance area mapping information. The risk assessment module obtains the layout data of key areas within the campus based on the behavior disturbance area mapping information, makes spatial overlap judgments between behavior paths and risk areas, identifies high-risk behavior segments, and generates a set of coordinates for risk path overlay blocks. The early warning response module calls the risk path overlay block coordinate set, obtains the number index of the coverage area at the campus level, marks the difference item of the output behavior change trend, and inputs the difference item as the external disturbance feature into the early warning input sequence to generate early warning information on the impact of disturbance on campus safety.

2. The campus security early warning system based on video analysis according to claim 1, characterized in that: The dynamic behavioral feature trend values ​​include edge pixel rotation amplitude, contour slope change density, environmental variable consistency level, and boundary direction mutation identification marker. The behavioral disturbance area mapping information includes the distribution of the principal axis angle of the overlapping area, the disturbance area coverage breadth index, environmental variable classification level, and projection area number mapping table. The risk path superimposed block coordinate set specifically includes the regional projection junction boundary point, the correspondence between the disturbance area and the region arrangement, the risk path blocking block index, and the spatial overlap density factor. The disturbance impact on campus safety early warning information includes the external disturbance interference feature sequence, the regional response trend symbol sequence, the neural network prediction value set, and the early warning segment index structure.

3. The campus security early warning system based on video analysis according to claim 1, characterized in that, The risk assessment module includes: The regional layout extraction submodule obtains the behavior disturbance region mapping information, collects key regional layout data within the campus, extracts the two-dimensional spatial location index corresponding to the regional number grid, spatially locates the arrangement of rows and columns and numbering order of the regions in the actual site, and generates a set of regional spatial locations by combining the regional orientation angle value. The disturbance path mapping submodule calls the set of spatial locations in the region, collects the distribution of environmental variables, the measured value of light intensity, and the azimuth angle of the surface of the region. Based on the path direction vector, it calculates the behavior path direction corresponding to the receiving surface of the region, projects the path direction onto the plane coordinates, determines whether there is an overlap with the coverage of the behavior disturbance area mapping information, obtains the spatial path segment that coincides with the behavior and the disturbance area, and generates path disturbance overlap distribution data. The cross-block identification submodule extracts the set of coordinate points corresponding to the occluded segment based on the path disturbance overlap distribution data, identifies the numbered region to which the coordinate points belong in the regional arrangement space, performs cross-judgment between the regional boundary and the occluded path boundary, extracts the regional number index corresponding to the intersection point, determines the regional number and corresponding coordinate region of the high-risk behavior segment, and establishes a risk path superimposed block coordinate set.

4. The campus security early warning system based on video analysis according to claim 1, characterized in that, The early warning response module includes: The regional signal acquisition submodule calls the risk path overlay block coordinate set, extracts the corresponding regional number index at the campus level, collects the unit regional output status value and environmental variable count value of the regional monitoring record, constructs the operating status data group of each region at the current moment, and generates a set of regional operating signals; The trend difference marking submodule obtains the state output values ​​of the region within the disturbance zone and the adjacent non-disturbance zone at the same time point based on the set of regional operation signals, compares the differences in the direction of change, identifies signal items with inconsistent directions, records the difference items in the form of symbol values, classifies and organizes them into a separate input factor sequence, and obtains a state trend difference vector group. The disturbance information generation submodule calls the state trend difference vector group, inputs the neural network input sequence structure trained with historical periodic samples, inputs the vector group and the constructed input factor terms in parallel, and performs combined inference based on environmental variables, light intensity and behavioral path data input factors to obtain the predicted value distribution results at the corresponding time point, and generates early warning information on the impact of disturbances on campus safety.

5. The campus security early warning system based on video analysis according to claim 1, characterized in that, The system also includes: The trusted label assignment module calls the disturbance impact on campus safety early warning information, obtains the predicted value and the actual output value of the area at each prediction time point, compares the root mean square error between the predicted value and the actual output value, and compares it with the set confidence error threshold. The error segment above the threshold is divided into intervals according to the level division rules, and each level division segment is marked as a trusted level identifier to generate the trusted level information of the disturbance coverage area. The perturbation coverage area output credibility level information specifically refers to the error interval level index, credibility label mapping result, RMSE fluctuation trend group, and credibility judgment result number table.

6. The campus security early warning system based on video analysis according to claim 5, characterized in that, The trusted tag assignment module includes: The error value calculation submodule calls the disturbance impact on campus safety early warning information, obtains the predicted output state sequence and the actual output state sequence of each region within the prediction time period, constructs the error difference set between the corresponding sequence indices, and combines the prediction state to normalize the error ratio and calculate the normalized state error index. The error segment division submodule calls the set confidence error threshold according to the normalized state error index, groups and classifies the index sequence according to the upper limit of the threshold, establishes index labels for the time period corresponding to the group, and sequentially numbers the error level corresponding to the group to obtain the error level mapping interval value. The credibility level labeling submodule calls the error level mapping interval value, establishes the credibility level interval correspondence rule according to the time mapping table of the level interval and the region number, outputs the credibility label result set corresponding to the region in each time period, and establishes the perturbation coverage area to output credibility level information.

Images