Urban safety monitoring and early warning system with unmanned aerial vehicle and bee colony cooperation

Abstract

The invention relates to the technical field of urban safety monitoring, and discloses an unmanned aerial vehicle and bee colony coordinated urban safety monitoring and early warning system, which comprises a bee colony initialization module used for acquiring historical risk heat data, dividing honeycomb units and determining the number of unmanned aerial vehicles in each layer; the route planning module is used for generating a whistle ring layer grid route and an inspection layer clustering route and managing the standby state of a near inspection layer; the risk modeling module is used for constructing a risk projection matrix, generating a spatial risk level and calculating a preliminary risk index; the bee gathering response module is used for activating the micro unmanned aerial vehicle based on the initial risk index, forming a collaborative bee gathering unit and calculating a secondary risk index; the early warning level judgment module is used for fusing the risk index, the target density and the speed information and determining an early warning level; and the scheduling processing module is used for calculating ground resource priorities and completing guidance processing. According to the invention, an efficient risk perception and linkage disposal system of urban air-ground cooperation is realized.

Description

Technical Field

[0001] This invention belongs to the field of urban safety monitoring technology, specifically relating to an urban safety monitoring and early warning system based on drone swarm collaboration. Background Technology

[0002] With increasingly complex urban spatial structures and a rising frequency of emergencies, traditional urban safety management methods relying on manual patrols or fixed monitoring equipment are no longer sufficient to meet the demands for efficient, real-time, and refined risk perception. In recent years, the development of drone technology has provided new solutions for urban public safety monitoring. While individual drones possess certain advantages in maneuverability and field of view, their monitoring capabilities remain limited by flight time, payload, and communication capabilities, leaving blind spots when facing large-scale, multi-dimensional, and dynamically evolving urban safety scenarios.

[0003] To overcome the limitations of individual drones, swarm drone collaborative technology has emerged. This technology utilizes multiple drones deployed in layers in the air and operating collaboratively to achieve high-altitude wide-area patrols, mid-level cruise monitoring, and low-altitude precision reconnaissance and response in urban spaces. However, existing swarm collaborative systems mostly focus on path planning and target identification, lacking risk-driven dynamic task allocation and layered linkage mechanisms. They struggle to dynamically adjust resource allocation based on the actual risk intensity of the monitored area, and lack systematic risk modeling and response grading capabilities, thus failing to achieve a closed-loop process from early warning identification to resource scheduling. Summary of the Invention

[0004] This invention provides an urban safety monitoring and early warning system based on drone swarm collaboration, which solves the technical problems in related technologies.

[0005] This invention provides an urban safety monitoring and early warning system based on drone swarm collaboration, comprising:

[0006] The swarm initialization module is used to acquire historical risk heat data of the monitoring area, divide the monitoring area into multiple swarm units according to the sentry layer, patrol layer and close observation layer, and determine the number of drones in each layer by combining the single frame coverage area, coverage overlap ratio and preset ratio coefficient.

[0007] The route planning module is used to control the fixed-wing UAVs in the sentry layer to construct sparse grid routes, control the rotary-wing UAVs in the patrol layer to perform clustering based on honeycomb units and generate patrol routes, and control the micro UAVs in the near observation layer to enter standby mode.

[0008] The risk modeling module is used to subdivide the honeycomb unit into sub-grids, collect multi-dimensional monitoring data, construct a risk projection matrix, generate spatial risk levels based on the degree of risk clustering in the matrix, and calculate a preliminary risk index by combining the anomaly intensity scores of each dimension.

[0009] The swarm response module is used to select the activation area based on the preliminary risk index, match micro-drones to form a candidate node set according to distance, establish node adjacency relationships to form a collaborative swarm unit, dynamically add nodes, and output the secondary risk index and swarm size.

[0010] The early warning classification module is used to integrate the secondary risk index with the target density and movement speed within the activated area, calculate the final risk score, and compare it with the preset risk threshold to determine the early warning level.

[0011] The dispatch and handling module is used to calculate priority scores based on the final risk score, the estimated arrival time of each ground resource, and the resource matching coefficient, and to select and dispatch the ground resource with the highest priority score.

[0012] Furthermore, the historical risk heat data consists of the historical risk heat values ​​of the basic grid of the monitoring area;

[0013] By collecting the drone's flight altitude and the camera's horizontal field of view, the tangent of the flight altitude and the camera's horizontal field of view is multiplied to obtain the ground coverage radius. Then, the area of ​​the circle is calculated using the ground coverage radius to obtain the ground area that can be covered by a single drone image as the single frame coverage area.

[0014] Furthermore, the weighted area is obtained by integrating the historical risk heat data of the monitoring area. The ratio of the weighted area to the product of the single-frame coverage area and the coverage overlap ratio of the sentry layer is calculated and rounded up to obtain the number of drones at the sentry layer. The number of drones at the sentry layer is obtained by multiplying the number of drones at the sentry layer by the first inter-layer ratio coefficient and the second inter-layer ratio coefficient, respectively.

[0015] The first and second inter-layer scaling factors are determined using a particle swarm optimization algorithm, including:

[0016] The first inter-layer scaling factor, the second inter-layer scaling factor, and the risk enhancement factor are combined to form a particle vector. The particle vector search boundary is set, and N groups of particles are randomly generated to form a particle swarm. The maximum number of iterations is set.

[0017] A fitness function is constructed by weighting three indicators—coverage omission rate, spatial overlap rate, and clustering replacement delay—with preset weights. Specifically, the coverage omission rate is obtained by comparing the sum of historical risk heat values ​​of uncovered basic grids with the sum of historical risk heat values ​​of all basic grids in the monitoring area; the spatial overlap rate is obtained by comparing the overlapping area of ​​multi-layer UAV imaging with the overall monitoring area; and the clustering replacement delay is obtained by comparing the replacement time of clustered nodes with the total duration of clustering triggering.

[0018] The fitness of each particle is calculated based on the fitness function, and the particle swarm is iterated using a velocity update formula with a risk enhancement term. The risk enhancement term is obtained by multiplying the risk enhancement factor by the spatial gradient of the quadratic risk index.

[0019] When the current iteration count reaches the maximum iteration count, output the particle vector with the highest fitness to obtain the final first inter-layer scaling factor and the second inter-layer scaling factor.

[0020] Furthermore, the route planning module specifically includes:

[0021] S201, by dividing the coverage area of ​​a single frame of the sentry layer by the coverage overlap ratio, the coverage area reference value is obtained, the cruise spacing is determined according to the square root of the coverage area reference value, and a sparse grid flight path of the sentry layer is generated in the detection area based on the cruise spacing.

[0022] S202, take the center position of the patrol layer honeycomb unit as the clustering sample, cluster according to the number of patrol layer rotary-wing UAVs, connect the centers of the honeycomb units in the unified cluster in the order of nearest neighbors, and generate the corresponding patrol layer cruise route.

[0023] S203 incorporates the near-survey micro-drone into a standby pool, records the remaining battery power and location information, and generates a standby sequence.

[0024] Furthermore, the multi-dimensional monitoring data includes: optical image data, infrared thermal imaging data, gas concentration data, and acoustic waveform data; the monitoring values ​​for each dimension include: the average brightness value of the optical image data, the peak temperature difference of the infrared thermal imaging data, the concentration value of the gas concentration data, and the power spectrum energy value of the acoustic waveform data.

[0025] The process of constructing a risk projection matrix includes:

[0026] S301, perform threshold comparison on the monitoring values ​​of each dimension of each sub-grid. If all monitoring values ​​are within the safe threshold range, assign a safe label. If any monitoring value falls within the warning threshold range, assign a suspicious label. If any monitoring value exceeds the alarm threshold, assign a high-risk label. Record the high-risk label indication value. When the high-risk label indication value is 1, it indicates a high-risk label. When the high-risk label indication value is 0, it indicates a safe label or a suspicious label.

[0027] S302, calculate the cumulative statistical value of risk events for each subgrid based on historical risk heat data, and calculate the ratio of this cumulative statistical value to the maximum cumulative statistical value among all subgrids to obtain the historical risk heat weight;

[0028] S303, multiply the high-risk label indication value of each subgrid by the historical risk heat weight of the same subgrid to obtain the subgrid risk projection value;

[0029] S304, construct a risk projection matrix using the subgrid risk projection values ​​as matrix elements.

[0030] Furthermore, the baseline mean, maximum monitoring value, and minimum monitoring value are extracted based on the monitoring values ​​of each dimension of each subgrid. The absolute value of the difference between the current monitoring value and the baseline mean is divided by the difference between the maximum monitoring value and the minimum monitoring value to obtain the anomaly intensity score for each dimension.

[0031] Furthermore, spatial risk levels are generated based on the degree of risk clustering in the matrix, and a preliminary risk index is calculated by combining the anomaly intensity scores of each dimension, including:

[0032] S401, the risk projection matrix is ​​accumulated and superimposed to obtain the risk clustering degree of the subgrid;

[0033] S402, based on the preset high-risk clustering threshold and the preset medium-risk clustering threshold, the risk clustering degree of the subgrid is discretized into spatial risk level and quantified;

[0034] S403, the anomaly intensity scores of each dimension are weighted according to the preset dimension weights to obtain the mean anomaly intensity;

[0035] S404 combines the numerical spatial risk level with the mean anomaly intensity according to a preset fusion coefficient to obtain a preliminary risk index.

[0036] Furthermore, the swarm response module specifically includes:

[0037] S501 defines the subgrids whose initial risk index exceeds the preset activation threshold as activation areas, and selects micro-drones from the standby pool according to the preset response distance threshold to obtain a candidate node set containing candidate nodes;

[0038] S502, calculate the distance between candidate nodes to generate an adjacency matrix. When the distance does not exceed the preset communication radius, set the adjacency value to 1 and extract connected components to form cooperative clustering units;

[0039] S503, if the number of collaborative clustering units is less than the preset target number, candidate nodes are dynamically added according to the rule of priority of remaining power and second best distance, and the adjacency matrix is ​​updated in real time.

[0040] S504: Calculate the ratio of the number of supplemented candidate nodes to the preset target size, multiply it by the preset gain coefficient and the preliminary risk index to obtain the secondary risk index, and output the number of supplemented candidate nodes as the cluster size.

[0041] Furthermore, the targets include: personnel and vehicles;

[0042] The secondary risk index is fused with the target density and movement speed within the activated area to calculate the final risk score, including:

[0043] S601: Count the number of targets in each sub-grid in the active region and calculate the target density based on the grid area, while also calculating the average movement speed of the targets;

[0044] S602, normalize the target density and movement speed using the historical maximum density and historical maximum speed respectively to obtain normalized density and normalized speed;

[0045] S603 calculates the final risk score by weighting and fusing the secondary risk index, normalized density, and normalized velocity.

[0046] Furthermore, the ground resources include: vehicles, equipment, and personnel;

[0047] By statistically analyzing the final risk scores of all subgrids in the activated area, the maximum value is taken as the risk extreme value. The estimated arrival time of each ground resource is obtained, and the priority score is calculated by weighting and fusing the risk extreme value, the estimated arrival time, and the preset resource matching coefficient.

[0048] The beneficial effects of this invention are as follows: By introducing a three-layer swarm architecture of sentinel layer, patrol layer, and close-range observation layer, this invention optimizes airspace division of labor and coverage efficiency, thereby improving risk perception capabilities in large-scale urban environments; it integrates historical risk heat and multi-dimensional real-time monitoring data, combined with spatial aggregation characteristics and multi-source anomaly intensity, to construct a highly accurate risk modeling mechanism; it utilizes a dynamic swarm mechanism and candidate node connectivity strategy to achieve efficient response and adaptive deployment of UAV swarms; and it combines ground resource scheduling and guidance mechanisms to streamline the entire chain of linkage from airborne early warning to ground response. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the modules of the UAV swarm collaborative urban safety monitoring and early warning system of the present invention. Detailed Implementation

[0050] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0051] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0052] like Figure 1 As shown, the urban safety monitoring and early warning system using drone swarm collaboration includes:

[0053] The swarm initialization module 101 is used to acquire historical risk heat data of the monitoring area, divide the monitoring area into multiple swarm units according to the sentry layer, patrol layer and close observation layer, and determine the number of drones in each layer by combining the single frame coverage area, coverage overlap ratio and preset ratio coefficient.

[0054] The route planning module 102 is used to control the fixed-wing UAV in the sentry layer to construct a sparse grid route, control the rotary-wing UAV in the patrol layer to perform clustering based on the honeycomb unit and generate patrol routes, and control the micro UAV in the near observation layer to enter the standby state.

[0055] The risk modeling module 103 is used to subdivide the honeycomb unit into sub-grids, collect multi-dimensional monitoring data, construct a risk projection matrix, generate a spatial risk level based on the degree of risk clustering in the matrix, and calculate a preliminary risk index by combining the anomaly intensity scores of each dimension.

[0056] The swarm response module 104 is used to select the activation area based on the preliminary risk index, match micro drones according to distance to form a candidate node set, establish node adjacency relationships to form a collaborative swarm unit, dynamically add nodes, and output the secondary risk index and swarm size.

[0057] The early warning classification module 105 is used to integrate the secondary risk index with the target density and movement speed in the activated area, calculate the final risk score, and compare it with the preset risk threshold to determine the early warning level.

[0058] The dispatch and disposal module 106 is used to calculate priority scores based on the final risk score, the expected arrival time of each ground resource, and the resource matching coefficient, select the ground resource with the highest priority score for dispatch, and guide it to the abnormal area by a micro drone to complete the disposal.

[0059] In one embodiment of the present invention, by connecting to multiple data sources such as the public security police system, fire dispatch records, and environmental monitoring networks, records related to safety risks within the monitoring area over the past three years are collected. These records include the occurrence time, geographical location, scope of impact, and severity level of events such as fires, public security incidents, gas leaks, and mass gatherings. The raw data is standardized, and risk weights are assigned according to the severity of the events. This data is then mapped to a 100m × 100m basic grid using GIS coordinates. The cumulative risk event weights for each basic grid are calculated and normalized by combining the population density, building density, and distribution of sensitive facilities within the grid. This yields the historical risk heat value for each basic grid, forming historical risk heat data.

[0060] Based on the aforementioned historical risk heat map data, the monitoring area is divided into three vertically stratified layers according to the performance characteristics of different types of UAVs: the sentry layer, the patrol layer, and the close-range observation layer. The sentry layer is located at high altitudes and is equipped with fixed-wing UAVs, with an altitude range of 200 to 400 meters; the patrol layer is located at mid-altitudes and is equipped with rotary-wing UAVs, with an altitude range of 50 to 200 meters; the close-range observation layer is located at low altitudes and is composed of micro-UAVs, with an altitude range of 10 to 50 meters. Within each layer, the monitoring area is further divided into multiple honeycomb units according to a preset grid division scale and honeycomb pattern. The honeycomb unit has a hexagonal or square structure, and its side length and area can be set according to the single-frame coverage area of ​​the UAVs in that layer.

[0061] In one embodiment of the present invention, the flight altitude of the UAV and the horizontal field of view of the camera are collected, and the tangent of the flight altitude and the horizontal field of view of the camera are multiplied to obtain the ground coverage radius. Then, the circular area is calculated using the ground coverage radius to obtain the ground area that can be covered by a single image from the UAV as the single-frame coverage area; wherein, the formula for calculating the single-frame coverage area is: S represents the coverage area of ​​a single frame, h represents the flight altitude, tan represents the tangent function, and θ represents the horizontal field of view of the camera, which is determined according to the configuration of the camera carried by the UAV.

[0062] In one embodiment of the present invention, the weighted area is obtained by integrating the historical risk heat data of the monitoring area. The ratio of the weighted area to the product of the single-frame coverage area and the coverage overlap ratio of the sentry layer is calculated and rounded up to obtain the number of drones at the sentry layer. The number of drones at the sentry layer is obtained by multiplying the number of drones at the sentry layer by the first inter-layer ratio coefficient and the second inter-layer ratio coefficient. The number of drones at the patrol layer and the number of drones at the close observation layer are obtained respectively.

[0063] The formula for calculating the weighted area is: A w =∫∫w(x,y)dxdy, where A w The weighted area is represented by w(x,y), which represents historical risk heat data. It is a two-dimensional heat function that reflects the spatial distribution of the frequency and risk intensity of safety incidents within the monitoring area, where x and y represent geographical coordinates.

[0064] The formula for calculating the number of drones at the sentry layer is: Where, N sentry S represents the number of drones at the sentry layer. sentry This represents the coverage area of ​​a single frame at the sentinel layer, and ξ represents the coverage overlap ratio, which is set according to the monitoring accuracy requirements. Indicates to Round up.

[0065] The first and second inter-layer scaling factors are determined using a particle swarm optimization algorithm, including:

[0066] The first inter-layer scaling factor, the second inter-layer scaling factor, and the risk enhancement factor are used to form a particle vector. The particle vector search boundary is set, and N groups of particles are randomly generated to form a particle swarm. The maximum number of iterations is set. The risk enhancement factor has a value range of 0.1 to 1 and is used to adjust the strength of the risk enhancement term.

[0067] A fitness function is constructed by weighting three indicators—coverage omission rate, spatial overlap rate, and swarm replacement delay—with preset weights. The coverage omission rate is obtained by comparing the sum of historical risk heat values ​​of uncovered basic grids with the sum of historical risk heat values ​​of all basic grids in the monitoring area. The spatial overlap rate is obtained by comparing the overlapping area of ​​multi-layer UAV imaging with the overall monitoring area, representing the efficiency of vertical airspace resource allocation. The swarm replacement delay is obtained by comparing the swarm replacement time with the total duration of triggering swarming, reflecting the timeliness of the scheduling response. The swarm replacement time refers to the time taken for the selected micro-UAV node to move from its original standby position along the planned path to the edge of the swarm unit after triggering the swarm response.

[0068] The fitness of each particle is calculated based on the fitness function, and the particle swarm is iterated using a velocity update formula with a risk enhancement term. The risk enhancement term is obtained by multiplying the risk enhancement factor by the spatial gradient of the quadratic risk index. Its function is to guide the particles to tilt towards the high-risk gradient region during the optimization process, thereby improving the adaptability of parameter configuration to high-risk scenarios and avoiding the problem of missing drone resources in local high-risk areas while achieving the best average index.

[0069] When the current iteration count reaches the maximum iteration count, output the particle vector with the highest fitness to obtain the final first inter-layer scaling factor and the second inter-layer scaling factor.

[0070] The above steps maximize the monitoring coverage of high-risk areas and reduce the risk of missed detections by comprehensively optimizing the coverage omission rate, airspace overlap rate and swarm filling delay under limited UAV resources. The introduction of the product term of the risk enhancement factor and the gradient of the second risk index makes the algorithm tilt resource allocation towards high-risk areas during the iteration process, thereby improving the system's response speed to emergencies.

[0071] In one embodiment of the present invention, the route planning module specifically includes:

[0072] S201: By dividing the coverage area of ​​a single frame of the sentry layer by the coverage overlap ratio, a baseline value of the coverage area is obtained. This baseline value reflects the effective monitoring range of a single UAV under the consideration of coverage overlap requirements. The cruise spacing is determined based on the square root of the baseline value of the coverage area to ensure that the coverage areas between the flight paths have neither obvious blind spots nor excessive overlap. Based on the cruise spacing, a sparse grid flight path of the sentry layer is generated in the detection area.

[0073] S202, using the center position of the inspection layer's honeycomb unit as the clustering sample, the K-Means clustering algorithm is used to divide the clustering sample into N groups. ins There are 1 cluster, and each cluster corresponds to the monitoring range of a drone. Within each cluster, the centers of the honeycomb units are connected in the order of nearest neighbors to form a closed path. That is, starting from the center of the cluster, the drone moves to the next honeycomb unit center that is closest to the current position each time, until all honeycomb units are covered and then it returns to the starting point, thus generating the patrol route of a single drone.

[0074] S203, the micro-drone at the near-field level is included in the standby pool, the remaining power and location information are recorded, and the standby order is generated; specifically, the standby order is generated according to the priority rules of remaining power from high to low and distance from the center of the monitoring area from near to far.

[0075] In one embodiment of the present invention, the multi-dimensional monitoring data includes: optical image data, infrared thermal imaging data, gas concentration data, and acoustic waveform data; the monitoring values ​​for each dimension include: the average brightness value of the optical image data, the peak temperature difference of the infrared thermal imaging data, the concentration value of the gas concentration data, and the power spectrum energy value of the acoustic waveform data; wherein, the optical image data is acquired by an RGB camera mounted on a UAV, the infrared thermal imaging data is acquired by a thermal imager, the gas concentration is acquired by a gas sensor, and the acoustic waveform data is acquired by a microphone array; the average brightness value is calculated after the optical image data is grayscaled; the peak temperature difference is extracted after the infrared thermal imaging data is noise-reduced and filtered; the concentration value of a specific gas, such as CO or NO2, is obtained after the gas concentration data is Kalman filtered; and the power spectrum energy value is calculated after the acoustic waveform data is Fourier transformed.

[0076] The process of constructing a risk projection matrix includes:

[0077] S301, perform threshold comparison on the monitoring values ​​of each dimension of each sub-grid. If all monitoring values ​​are within the safe threshold range, assign a safe label. If any monitoring value falls within the warning threshold range, assign a suspicious label. If any monitoring value exceeds the alarm threshold, assign a high-risk label. Record the high-risk label indication value. When the high-risk label indication value is 1, it indicates a high-risk label. When the high-risk label indication value is 0, it indicates a safe label or a suspicious label.

[0078] S302, calculate the cumulative statistical value of risk events for each subgrid based on historical risk heat data, and calculate the ratio of this cumulative statistical value to the maximum cumulative statistical value among all subgrids to obtain the historical risk heat weight; specifically, the cumulative statistical value of risk events for each subgrid in the past 3 years, such as the number of fires and the number of alarms, can be calculated and normalized based on the maximum cumulative statistical value of all subgrids.

[0079] S303, multiply the high-risk label indication value of each subgrid by the historical risk heat weight of the same subgrid to obtain the subgrid risk projection value;

[0080] S304, construct a risk projection matrix using the risk projection values ​​of the subgrids as matrix elements; this risk projection matrix reflects the real-time risk distribution of the monitoring area in numerical form.

[0081] This embodiment achieves multi-dimensional cross-validation of abnormal events by using threshold comparison of multi-dimensional data, reducing the false alarm rate of a single data source; the calculation of historical risk heat weights makes risk assessment both dynamic and real-time, as well as historical regularity; the binarization of high-risk label indication values ​​simplifies the fusion logic of complex multi-source data and improves computational efficiency; the risk projection matrix integrates spatially discrete risk information into a continuous mathematical expression, providing a direct quantitative basis for subsequent swarm scheduling.

[0082] In one embodiment of the present invention, the baseline mean, maximum monitoring value and minimum monitoring value are extracted based on the monitoring values ​​of each dimension of each sub-grid. The absolute value of the difference between the current monitoring value and the baseline mean is divided by the difference between the maximum monitoring value and the minimum monitoring value to obtain the anomaly intensity score of each dimension. Specifically, monitoring values ​​within a first preset time period are collected, the baseline mean, maximum monitoring value and minimum monitoring value are calculated, and then the anomaly intensity score of the current monitoring value is calculated to reflect the relative degree of deviation of the current monitoring value from the historical normal.

[0083] In one embodiment of the present invention, a spatial risk level is generated based on the degree of risk clustering in the matrix, and a preliminary risk index is calculated by combining the anomaly intensity scores of each dimension, including:

[0084] S401, the risk projection matrix is ​​accumulated and superimposed to obtain the risk clustering degree of the subgrid; by superimposing the elements of each row and column in the matrix and mapping them to the subgrid, the risk clustering degree of the subgrid reflecting the degree of risk clustering around the subgrid is obtained.

[0085] S402, the risk clustering degree of the subgrid is discretized into spatial risk levels and quantified according to the preset high-risk clustering threshold and the preset medium-risk clustering threshold; specifically, the spatial risk levels include: high-risk level, medium-risk level and low-risk level, and are represented by 3, 2 and 1 respectively.

[0086] S403, the anomaly intensity scores of each dimension are weighted according to the preset dimension weights to obtain the mean anomaly intensity;

[0087] S404, the numerical spatial risk level and the mean anomaly intensity are fused according to a preset fusion coefficient to obtain a preliminary risk index; wherein, the calculation formula of the preliminary risk index is: RI represents the preliminary risk index, β represents the preset fusion coefficient, and L represents the numerical spatial risk level. This represents the mean of the anomaly intensity.

[0088] This embodiment calculates the risk clustering degree of subgrids by accumulating and mapping rows and columns, capturing the spatial correlation of risks and avoiding isolated assessment of individual subgrid risks; the three-level spatial risk classification simplifies the expression of complex risk distributions and facilitates subsequent resource scheduling decisions; the setting of dimension weights allows for customized configuration of monitoring priorities for different monitoring scenarios such as chemical leaks and crowd gatherings; the introduction of fusion coefficients balances the influence of spatial risk background and real-time anomaly intensity, improving the stability of risk assessment.

[0089] In one embodiment of the present invention, the honeycomb aggregation response module specifically includes:

[0090] S501 defines the subgrids whose initial risk index exceeds the preset activation threshold as activation areas, calculates the straight-line distance between each drone and the center of the activation area, and selects drones whose straight-line distance is less than or equal to the preset response distance threshold to form a candidate node set.

[0091] S502, calculate the distance between candidate nodes to generate an adjacency matrix. When the distance does not exceed the preset communication radius, set the adjacency value to 1 and extract connected components to form a cooperative swarm unit. Here, a connected component represents a subset of all mutually reachable candidate nodes. Each connected component forms a cooperative swarm unit to ensure that candidate nodes within the unit can share data in real time.

[0092] S503 If the number of collaborative clustering units is less than the preset target number, candidate nodes are dynamically added according to the rule of priority of remaining power and second-best distance, and the adjacency matrix is ​​updated in real time; wherein, the preset target size is set according to the area of ​​the activated region and the spatial risk level.

[0093] S504: Calculate the ratio of the number of supplemented candidate nodes to the preset target size, multiply it by the preset gain coefficient and the preliminary risk index to obtain the secondary risk index, and output the number of supplemented candidate nodes as the cluster size.

[0094] This embodiment avoids invalid clustering in low-risk areas by setting a preset activation threshold, reducing resource waste. Communication radius and connectivity analysis ensure the collaborative ability of clustering units, avoiding the problem of physical aggregation but data isolation. The dynamic supplementation mechanism balances node endurance and response speed by sorting based on both power consumption and distance, ensuring that the clustering scale matches the spatial risk level.

[0095] In one embodiment of the present invention, the targets include: personnel and vehicles;

[0096] The secondary risk index is fused with the target density and movement speed within the activated area to calculate the final risk score, including:

[0097] S601 counts the number of targets in each sub-grid within the active region and calculates the target density based on the grid area, while simultaneously calculating the average moving speed of the targets. Among these methods, the HOG+SVM algorithm of optical images can be used to detect human contours, and the thermal radiation characteristics of infrared thermal imaging can be combined to distinguish between living and non-living entities. The geometric features of the lidar point cloud are used to identify vehicle contours, and the moving speed is calculated through positional changes between consecutive frames.

[0098] S602, normalize the target density and movement speed using the historical maximum density and historical maximum speed respectively to obtain normalized density and normalized speed;

[0099] S603 calculates the final risk score by weighting and fusing the secondary risk index, normalized density, and normalized velocity.

[0100] This embodiment eliminates the impact of differences in the basic environment of different monitoring areas by normalizing the target density and movement speed. The weighted fusion mechanism flexibly allocates weights according to the characteristics of the scene, which enhances the adaptability of risk assessment.

[0101] In one embodiment of the present invention, the ground resources include: vehicles, equipment, and personnel;

[0102] By statistically analyzing the final risk scores of all sub-grids in the activated area, the maximum value is taken as the risk extreme value. The estimated arrival time of each ground resource is then obtained. A priority score is calculated by weighting and fusing the risk extreme value, estimated arrival time, and preset resource matching coefficient. Specifically, the formula for calculating the priority score is as follows:

[0103] S j This represents the priority score of the j-th ground resource, where j represents the ground resource index, and w1, w2, and w3 represent the first, second, and third weight coefficients, respectively. RF max t represents the extreme value of risk. eta,j Let t represent the estimated arrival time of the j-th surface resource. max C represents the longest estimated arrival time of ground resources within the activated area. j This represents the preset resource matching coefficient, with a value ranging from 0 to 1, reflecting the degree of matching.

[0104] In this embodiment, the selection of risk extremes focuses on the most critical point in the activated region, ensuring that resources are prioritized for the highest risk location; the normalization of the estimated arrival time avoids response delays caused by excessively long resource distances; and the introduction of a preset resource matching coefficient gives special resources higher priority, reducing resource mismatch.

[0105] After determining the priority scores of each ground resource, the resource with the highest priority score is selected from all ground resources, and a dispatch order is generated and pushed to the corresponding ground resource terminal. Simultaneously, near-survey micro-UAVs within the area are activated to lock onto the core location of the anomaly zone in real time. Dynamic navigation information, including real-time latitude and longitude coordinates, optimal path, and images of the surrounding environment of the anomaly zone, is sent to the dispatched resource via air-to-ground communication links, ensuring the ground resource accurately reaches the target area. Upon arrival, the micro-UAVs continuously provide aerial perspective support, assisting ground personnel in quickly locating the anomaly point and updating on-site data in real time according to the progress of the response until the response is completed.

[0106] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0107] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A drone swarm collaborative urban safety monitoring and early warning system, characterized in that, include: The swarm initialization module is used to acquire historical risk heat data of the monitoring area, divide the monitoring area into multiple swarm units according to the sentry layer, patrol layer and close observation layer, and determine the number of drones in each layer by combining the single frame coverage area, coverage overlap ratio and preset ratio coefficient. The route planning module is used to control the fixed-wing UAVs in the sentry layer to construct sparse grid routes, control the rotary-wing UAVs in the patrol layer to perform clustering based on honeycomb units and generate patrol routes, and control the micro UAVs in the near observation layer to enter standby mode. The risk modeling module is used to subdivide the honeycomb unit into sub-grids, collect multi-dimensional monitoring data, construct a risk projection matrix, generate spatial risk levels based on the degree of risk clustering in the matrix, and calculate a preliminary risk index by combining the anomaly intensity scores of each dimension. The swarm response module is used to select the activation area based on the preliminary risk index, match micro-drones to form a candidate node set according to distance, establish node adjacency relationships to form a collaborative swarm unit, dynamically add nodes, and output the secondary risk index and swarm size. The early warning classification module is used to integrate the secondary risk index with the target density and movement speed within the activated area, calculate the final risk score, and compare it with the preset risk threshold to determine the early warning level. The dispatch and handling module is used to calculate priority scores based on the final risk score, the estimated arrival time of each ground resource, and the resource matching coefficient, and to select and dispatch the ground resource with the highest priority score.

2. The urban safety monitoring and early warning system based on UAV swarm collaboration according to claim 1, characterized in that, The historical risk heat data consists of the historical risk heat values ​​of the basic grid of the monitoring area; By collecting the drone's flight altitude and the camera's horizontal field of view, the tangent of the flight altitude and the camera's horizontal field of view is multiplied to obtain the ground coverage radius. Then, the area of ​​the circle is calculated using the ground coverage radius to obtain the ground area that can be covered by a single drone image as the single frame coverage area.

3. The urban safety monitoring and early warning system based on drone swarm collaboration according to claim 2, characterized in that, The weighted area is obtained by integrating the historical risk heat data of the monitoring area. The ratio of the weighted area to the product of the single frame coverage area and the coverage overlap ratio of the sentry layer is calculated and rounded up to obtain the number of drones at the sentry layer. The number of drones at the sentry layer is obtained by multiplying the number of drones at the sentry layer by the first inter-layer ratio coefficient and the second inter-layer ratio coefficient respectively. The number of drones at the patrol layer and the number of drones at the close observation layer are obtained. The first and second inter-layer scaling factors are determined using a particle swarm optimization algorithm, including: The first inter-layer scaling factor, the second inter-layer scaling factor, and the risk enhancement factor are combined to form a particle vector. The particle vector search boundary is set, and N groups of particles are randomly generated to form a particle swarm. The maximum number of iterations is set. A fitness function is constructed by weighting three indicators—coverage omission rate, spatial overlap rate, and clustering replacement delay—with preset weights. Specifically, the coverage omission rate is obtained by comparing the sum of historical risk heat values ​​of uncovered basic grids with the sum of historical risk heat values ​​of all basic grids in the monitoring area; the spatial overlap rate is obtained by comparing the overlapping area of ​​multi-layer UAV imaging with the overall monitoring area; and the clustering replacement delay is obtained by comparing the replacement time of clustered nodes with the total duration of clustering triggering. The fitness of each particle is calculated based on the fitness function, and the particle swarm is iterated using a velocity update formula with a risk enhancement term. The risk enhancement term is obtained by multiplying the risk enhancement factor by the spatial gradient of the quadratic risk index. When the current iteration count reaches the maximum iteration count, output the particle vector with the highest fitness to obtain the final first inter-layer scaling factor and the second inter-layer scaling factor.

4. The urban safety monitoring and early warning system based on UAV swarm collaboration according to claim 2, characterized in that, The route planning module specifically includes: S201, by dividing the coverage area of ​​a single frame of the sentry layer by the coverage overlap ratio, the coverage area reference value is obtained, the cruise spacing is determined according to the square root of the coverage area reference value, and a sparse grid flight path of the sentry layer is generated in the detection area based on the cruise spacing. S202, take the center position of the patrol layer honeycomb unit as the clustering sample, cluster according to the number of patrol layer rotary-wing UAVs, connect the centers of the honeycomb units in the unified cluster in the order of nearest neighbors, and generate the corresponding patrol layer cruise route. S203 incorporates the near-survey micro-drone into a standby pool, records the remaining battery power and location information, and generates a standby sequence.

5. The urban safety monitoring and early warning system based on drone swarm collaboration according to claim 1, characterized in that, The multi-dimensional monitoring data includes: optical image data, infrared thermal imaging data, gas concentration data, and acoustic waveform data; the monitoring values ​​for each dimension include: the average brightness value of the optical image data, the peak temperature difference of the infrared thermal imaging data, the concentration value of the gas concentration data, and the power spectrum energy value of the acoustic waveform data. The process of constructing a risk projection matrix includes: S301, perform threshold comparison on the monitoring values ​​of each dimension of each sub-grid. If all monitoring values ​​are within the safe threshold range, assign a safe label. If any monitoring value falls within the warning threshold range, assign a suspicious label. If any monitoring value exceeds the alarm threshold, assign a high-risk label. Record the high-risk label indication value. When the high-risk label indication value is 1, it indicates a high-risk label. When the high-risk label indication value is 0, it indicates a safe label or a suspicious label. S302, calculate the cumulative statistical value of risk events for each subgrid based on historical risk heat data, and calculate the ratio of this cumulative statistical value to the maximum cumulative statistical value among all subgrids to obtain the historical risk heat weight; S303, multiply the high-risk label indication value of each subgrid by the historical risk heat weight of the same subgrid to obtain the subgrid risk projection value; S304, construct a risk projection matrix using the subgrid risk projection values ​​as matrix elements.

6. The urban safety monitoring and early warning system based on UAV swarm collaboration according to claim 5, characterized in that, The baseline mean, maximum monitoring value, and minimum monitoring value are extracted from the monitoring values ​​of each dimension of each subgrid. The absolute value of the difference between the current monitoring value and the baseline mean is divided by the difference between the maximum monitoring value and the minimum monitoring value to obtain the anomaly intensity score for each dimension.

7. The urban safety monitoring and early warning system based on drone swarm collaboration according to claim 6, characterized in that, Spatial risk levels are generated based on the degree of risk clustering in the matrix, and a preliminary risk index is calculated by combining the anomaly intensity scores of each dimension, including: S401, the risk projection matrix is ​​accumulated and superimposed to obtain the risk clustering degree of the subgrid; S402, based on the preset high-risk clustering threshold and the preset medium-risk clustering threshold, the risk clustering degree of the subgrid is discretized into spatial risk level and quantified; S403, the anomaly intensity scores of each dimension are weighted according to the preset dimension weights to obtain the mean anomaly intensity; S404 combines the numerical spatial risk level with the mean anomaly intensity according to a preset fusion coefficient to obtain a preliminary risk index.

8. The urban safety monitoring and early warning system based on UAV swarm collaboration according to claim 7, characterized in that, The honeycomb response module specifically includes: S501 defines the subgrids whose initial risk index exceeds the preset activation threshold as activation areas, and selects micro-drones from the standby pool according to the preset response distance threshold to obtain a candidate node set containing candidate nodes; S502, calculate the distance between candidate nodes to generate an adjacency matrix. When the distance does not exceed the preset communication radius, set the adjacency value to 1 and extract connected components to form cooperative clustering units; S503, if the number of collaborative clustering units is less than the preset target number, candidate nodes are dynamically added according to the rule of priority of remaining power and second best distance, and the adjacency matrix is ​​updated in real time. S504: Calculate the ratio of the number of supplemented candidate nodes to the preset target size, multiply it by the preset gain coefficient and the preliminary risk index to obtain the secondary risk index, and output the number of supplemented candidate nodes as the cluster size.

9. The urban safety monitoring and early warning system based on UAV swarm collaboration according to claim 8, characterized in that, The targets include: personnel and vehicles; The secondary risk index is fused with the target density and movement speed within the activated area to calculate the final risk score, including: S601: Count the number of targets in each sub-grid in the active region and calculate the target density based on the grid area, while also calculating the average movement speed of the targets; S602, normalize the target density and movement speed using the historical maximum density and historical maximum speed respectively to obtain normalized density and normalized speed; S603 calculates the final risk score by weighting and fusing the secondary risk index, normalized density, and normalized velocity.

10. The urban safety monitoring and early warning system based on UAV swarm collaboration according to claim 9, characterized in that, The ground resources include: vehicles, equipment, and personnel; By statistically analyzing the final risk scores of all subgrids in the activated area, the maximum value is taken as the risk extreme value. The estimated arrival time of each ground resource is obtained, and the priority score is calculated by weighting and fusing the risk extreme value, the estimated arrival time, and the preset resource matching coefficient.

Images