Air-ground cooperative unmanned patrol system and method

Abstract

The application provides an air-ground cooperative unmanned patrol system and method, which comprises an unmanned vehicle, an unmanned aerial vehicle cluster and a ground control terminal. The ground control terminal is used for loading map information of a region to be inspected, combining a task target input by a user, confirming a patrol task mode and patrol task formation parameters through a multi-parameter decision algorithm, encapsulating the patrol task mode and patrol task formation parameters as instructions, and delivering the instructions to the unmanned vehicle through an ad hoc network. The unmanned vehicle is used for analyzing the instructions, scheduling the unmanned aerial vehicle cluster according to real-time power of each unmanned aerial vehicle in the nest, the patrol task mode and the patrol task formation parameters, and performing target detection and behavior recognition according to image data obtained by a vehicle-mounted pan-tilt head and image information obtained by the unmanned aerial vehicle. The unmanned aerial vehicle obtains image information of an object to be inspected and returns the image information to the unmanned vehicle, thereby improving patrol efficiency and reliability.

Description

Technical Field

[0001] This invention relates to the field of collaborative control and security inspection of unmanned systems, and in particular to a ground-air collaborative unmanned inspection system and method. Background Technology

[0002] Traditional security patrols mainly rely on human foot patrols or vehicle inspections, which suffer from low efficiency, limited field of vision, and slow response speed. Existing single unmanned vehicle or drone inspection solutions also have limitations: unmanned vehicles have poor maneuverability in complex terrain areas and their field of vision is easily obstructed; drones have short flight times and lack a power supply platform for continuous operation.

[0003] Although there are solutions combining drones and unmanned vehicles in related technologies, the coordination capabilities between drones and unmanned vehicles are insufficient. In multi-drone collaborative operations, the task inspection mode is generally fixed and cannot be flexibly selected according to the actual situation of the area to be inspected, resulting in low inspection efficiency and reliability.

[0004] Therefore, how to combine the advantages of drones and unmanned vehicles to achieve integrated air-ground, all-weather, and intelligent collaborative inspections, and to achieve flexible selection of task inspection modes in multi-drone collaborative operations, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, this invention innovatively proposes a ground-air collaborative unmanned patrol system and method, which effectively solves the problem of low patrol efficiency and reliability caused by the prior art, and effectively improves patrol efficiency and reliability.

[0006] The first aspect of this invention provides a ground-air collaborative unmanned patrol system, comprising: an unmanned vehicle, a drone swarm, and a ground control terminal. The ground control terminal is used to load map information of the area to be inspected, and, in conjunction with the task objectives input by the user, to confirm the patrol task mode and patrol task formation parameters through a multi-parameter decision algorithm. The patrol task mode and patrol task formation parameters are then encapsulated into instructions and sent to the unmanned vehicle via a self-organizing network. The unmanned vehicle is used to parse the instructions, schedule the drone swarm based on the real-time battery level of each drone in the drone nest, the patrol task mode, and the patrol task formation parameters, and perform target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the drones. The drones are used to acquire image information of the objects to be inspected and transmit the image information back to the unmanned vehicle.

[0007] Optionally, the map information of the area to be inspected is loaded, and combined with the task objectives input by the user, the patrol task mode and patrol task formation parameters are determined through a multi-parameter decision algorithm, including: The system loads map information of the area to be inspected, combines it with the user-inputted task objectives, and confirms the inspection task mode through a multi-parameter decision algorithm and decision rules. The map information includes the outlines, heights, and road network information of buildings within the inspection area. The patrol formation parameters are determined based on the patrol mission mode and road condition information.

[0008] Furthermore, based on the user-inputted task objectives, the patrol task mode is confirmed through a multi-parameter decision algorithm and decision rules, specifically including: Based on the map information of the area to be inspected, determine the area of ​​the inspection site, the terrain complexity index, and the density of obstructions; Based on the inspection site area, terrain complexity index, obstruction density, and decision rules, the inspection task mode is determined; among them, the inspection task modes include autonomous mixed patrol of drones and unmanned vehicles, autonomous patrol of drones, autonomous patrol of unmanned vehicles, and remote-controlled patrol.

[0009] Furthermore, the decision-making rules include: If the area of ​​the inspection site is greater than the preset area threshold, and the terrain complexity index is greater than the preset complexity threshold, or the density of obstructions is greater than the preset obstruction density threshold, then the inspection task mode is recommended to be a mixed autonomous patrol of drones and unmanned vehicles. If the area of ​​the inspection site is not greater than the preset area threshold, and the density of obstructions is greater than the preset obstruction density threshold, then the inspection task mode is recommended to be drone autonomous patrol. If the area of ​​the inspection site is greater than the preset area threshold, and the terrain complexity index is not greater than the preset complexity threshold, then the inspection task mode is recommended to be autonomous patrol by unmanned vehicles. If the user specifies close-range reconnaissance, the patrol mission mode is forcibly set to remote patrol.

[0010] Optionally, the method for calculating the terrain complexity index is as follows: Based on the DEM data of the map, the slope value of the grid points in the area to be inspected is calculated, and the slope factor in the area to be inspected is determined based on the slope value; where the slope factor is the proportion of the area in the area to be inspected where the slope value exceeds the maximum climbable slope of the unmanned vehicle. The vegetation factor within the area to be inspected is determined based on the vegetation coverage. The vegetation factor is the percentage of the area of ​​vegetation with a height exceeding a preset height threshold. The slope factor and vegetation factor in the area to be inspected are weighted and summed to obtain the terrain complexity index. The specific method for calculating the density of obstructions is as follows:

[0011] in The first in the area to be inspected The floor area of ​​each building, The first in the area to be inspected The height of the building For reference height, This represents the total area of ​​the area to be inspected.

[0012] Optionally, the patrol formation parameters determined based on the road condition information of the patrol mission mode specifically include: The number of drones to be taken off is determined based on the patrol mission mode and the area to be inspected. When the patrol mission mode is a mixed autonomous patrol of drones and unmanned vehicles or an autonomous patrol of drones, the formation type, formation distance parameters, and flight altitude of the drones are selected according to the current road conditions of the area to be inspected and the correspondence of formation parameters; the correspondence of formation parameters stores the correspondence between different road conditions and drone formations.

[0013] Furthermore, the scheduling of the drone swarm based on the real-time battery level, patrol mission mode, and patrol mission formation parameters of each drone within the swarm specifically includes: Based on the number N drones to be launched, select the N drones with the highest battery levels from those with battery levels greater than the first battery threshold and launch them; if the number of drones with battery levels greater than the first battery threshold is less than N, then launch them from drones with battery levels greater than the second battery threshold. During the mission, the unmanned vehicle continuously monitors the remaining battery power of the aerial drones. When the battery power of any aerial drone falls below a preset third battery threshold, a drone rotation process is triggered. The first, second, and third battery thresholds decrease sequentially. Adjust the drone's mission type according to its battery level.

[0014] Optionally, target detection and behavior recognition models are set up in the drone and the unmanned vehicle respectively. When both the unmanned vehicle and the drone detect anomalies in the same area, the overall confidence scores of the drone and the unmanned vehicle are compared, and the detection result with the highest overall confidence score is selected as the final alarm basis. If the drone or the unmanned vehicle detects anomalies in the same area and the overall confidence score exceeds a preset overall confidence score threshold, an alarm is triggered. The calculation method for the overall confidence score is as follows: ; Among them, C total To determine the overall confidence level, C conf The original confidence level. The original confidence weights, The visual quality index. As the weight of the perspective quality index, Infrared characteristic intensity, The infrared feature intensity weights are M, where M is the normalized velocity of the target motion. Z represents the normalized velocity weights for the target motion, and Z represents the region sensitivity. The region sensitivity weight is denoted by ; where the viewpoint quality index is the proportion of the target imaging area to the total image area.

[0015] Optionally, the unmanned vehicle and the drone obtain their relative positions through an ad hoc network. When the relative position is greater than a preset distance threshold, the drone sets the landing target point to the current coordinates of the unmanned vehicle plus a preset return point offset. When the relative position is not greater than the preset distance threshold, the drone captures the QR code on the drone's nest platform and calculates the drone's position deviation and yaw angle deviation relative to the center of the QR code. The drone then performs the landing operation based on the position deviation and yaw angle deviation relative to the center of the QR code.

[0016] The second aspect of this invention provides a ground-air cooperative unmanned patrol method, implemented based on the ground-air cooperative unmanned patrol system provided in the first aspect of this invention, comprising: The ground control terminal loads map information of the area to be inspected, combines it with the task objectives input by the user, and uses a multi-parameter decision algorithm to confirm the patrol task mode and patrol task formation parameters. The patrol task mode and patrol task formation parameters are then encapsulated into instructions and sent to the unmanned vehicle through the self-organizing network. The unmanned vehicle interprets commands and schedules the drone swarm based on the real-time battery level, patrol mission mode, and patrol mission formation parameters of each drone in the drone nest. The drone acquires image information of the object to be inspected and transmits the image information back to the unmanned vehicle; The unmanned vehicle performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the drone.

[0017] The technical solution adopted in this invention has the following technical effects: 1. In the technical solution of this invention, the ground control terminal is used to load map information of the area to be inspected, and combined with the task target input by the user, confirms the inspection task mode and inspection task formation parameters through a multi-parameter decision algorithm. The inspection task mode and inspection task formation parameters are encapsulated into instructions and sent to the unmanned vehicle through a self-organizing network. The unmanned vehicle is used to parse the instructions, schedule the drone cluster according to the real-time battery level of each drone in the drone nest, the inspection task mode, and the inspection task formation parameters, and perform target detection and behavior recognition based on the image data obtained by the vehicle-mounted gimbal and the image information obtained by the drones. The drones obtain the image information of the object to be inspected and transmit the image information back to the unmanned vehicle, effectively solving the problem of low inspection efficiency and reliability caused by the existing technology, and effectively improving inspection efficiency and reliability.

[0018] 2. In the technical solution of the present invention, the area of ​​the inspection site, the terrain complexity index, and the density of obstructions are determined based on the map information of the area to be inspected. Based on the area of ​​the inspection site, the terrain complexity index, the density of obstructions, and the decision rules, the inspection task mode is determined. The inspection task formation parameters are determined according to the inspection task mode and road condition information. Different inspection task modes can be flexibly selected according to the actual situation of the area to be inspected, which improves the adaptability of the inspection to different terrains and task requirements.

[0019] 3. The technical solution of this invention adjusts the mission type of the drone according to its different battery levels. Through dynamic battery management and rotation strategy, it solves the problem of short battery life of a single drone and realizes uninterrupted relay patrol of "one vehicle and multiple drones", ensuring seamless mission coverage.

[0020] 4. In the technical solution of this invention, target detection and behavior recognition models are set up in the drone and the unmanned vehicle respectively. When both the unmanned vehicle and the drone detect anomalies in the same area, the comprehensive confidence scores of the unmanned vehicle and the drone are compared, and the detection result with the highest comprehensive confidence score is selected as the final alarm basis. If the unmanned vehicle or the drone detects anomalies in the same area and the comprehensive confidence score exceeds the preset comprehensive confidence score threshold, an alarm is triggered. Multi-source data from the vehicle's high-altitude view and the drone's low-altitude view are fused, and a decision-level fusion algorithm based on confidence score is adopted to significantly improve the detection rate of abnormal events in complex environments and reduce false alarms and missed alarms.

[0021] 5. In the technical solution of this invention, when the relative position is greater than a preset distance threshold, the UAV sets the landing target point to the current coordinates of the unmanned vehicle plus a preset return point offset; when the relative position is not greater than the preset distance threshold, the UAV performs the landing operation based on the position deviation and yaw angle deviation relative to the center of the nest QR code; through the dual guidance mechanism of long-range coarse guidance + close-range visual fine guidance, the UAV is ensured to land accurately on the unmanned vehicle dynamic platform.

[0022] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the communication between the unmanned vehicle and the drone cluster in the system of Embodiment 1 of the present invention; Figure 2This is a flowchart illustrating the method of Embodiment 1 in the present invention. Detailed Implementation

[0025] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0026] Example 1 like Figure 1 As shown, this invention provides a ground-air collaborative unmanned patrol system, comprising: an unmanned vehicle, a drone swarm, and a ground control terminal. The ground control terminal loads map information of the area to be inspected, combines it with user-inputted task objectives, and uses a multi-parameter decision algorithm to determine the patrol task mode and patrol task formation parameters. It then encapsulates the patrol task mode and formation parameters into instructions and sends them to the unmanned vehicle via a self-organizing network. The unmanned vehicle parses the instructions, schedules the drone swarm based on the real-time battery level of each drone in its pod, the patrol task mode, and the patrol task formation parameters, and performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the drones. The drones acquire image information of the objects to be inspected and transmit the image information back to the unmanned vehicle.

[0027] Among them, the unmanned vehicle, serving as a ground inspection and energy supply mechanism, can include the following structure: 1. Industrial control computer: operation control software, multi-source data fusion algorithm, deep learning recognition model and task scheduling engine.

[0028] 2. Integrated navigation module: integrates an RTK-GPS / BeiDou dual-mode receiver and a high-precision inertial measurement unit (IMU) to provide centimeter-level positioning and attitude information.

[0029] 3. Tri-light gimbal: Includes a visible light camera (supporting 70x optical zoom), an uncooled infrared thermal imager, and a laser rangefinder, used for wide-area ground monitoring and target tracking.

[0030] 4. LiDAR: Used for obstacle detection and avoidance in autonomous navigation of unmanned vehicles.

[0031] 5. Vehicle-mounted self-organizing network radio: Adopts the MESH self-organizing network protocol to achieve high-bandwidth, low-latency data interaction with drones and control terminals.

[0032] 6. Onboard automated control unit (ACU), which includes the following structures: (1) Six independent take-off and landing platforms: Each platform corresponds to a UAV with a fixed number. The platform surface is printed with a high-contrast QR code (30cm×30cm in size, for visual guidance).

[0033] (2) Automatic door mechanism: The door is an opening and closing mechanism controlled by a servo motor. It opens automatically before takeoff and closes automatically after landing.

[0034] (3) Contact charging module: Each platform is equipped with a spring-loaded charging probe, which automatically connects after the drone lands in place to achieve constant current and constant voltage charging.

[0035] (4) CAN bus communication interface: After the UAV lands, it establishes wired communication with the unmanned vehicle through the contact point for data transmission and status query.

[0036] (5) Mechanical return and locking mechanism: The platform is equipped with an electric guide structure and an electric locking device. After the UAV lands and is guided back to its original position by the guide mechanism, the UAV landing gear is automatically locked to prevent the vehicle from shaking. At the same time, the platform integrates an automatic rotor folding mechanism to store the rotor blades in a safe position after the UAV is locked.

[0037] A patrol drone swarm can include several identical quadcopter drones, each drone containing: 1. Flight Controller: Runs UAV flight control algorithms and supports RTK positioning and vision-assisted navigation.

[0038] 2. Dual-light gimbal: Equipped with a visible light camera (supporting 4K video) and an infrared sensor for aerial reconnaissance.

[0039] 3. Onboard vision processing unit: Embedded GPU, running visual SLAM algorithm and QR code recognition algorithm.

[0040] 4. Airborne self-organizing network radio: It can network with unmanned vehicles and control terminals to transmit telemetry data such as images, location, and power in real time.

[0041] 5. Bottom charging contacts: Correspond to the contacts on the nest platform, automatically making electrical connection during landing.

[0042] 6. Identification Number: Each UAV has a fixed ID (e.g., UAV001~UAV006), which is bound to the UAV Nest platform.

[0043] The ground control terminal, serving as a ground mobile command center, includes: 1. High-performance ruggedized tablet PC based on an embedded operating system: Install dedicated control software.

[0044] 2. Self-organizing network handheld terminal: Establishes communication links with unmanned vehicles and drones.

[0045] 3. Functional modules: including high-precision map loading module, task planning module, multi-machine status monitoring module, load control module, time-sharing / cluster control module, and multi-source data fusion and display module.

[0046] The core patrol task mode in this invention may include: 1. Silent Monitoring Mode: Unmanned vehicles are stationed in key areas, while the three-light gimbal performs low-power monitoring, and the drones stand by charging in their nests. Suitable for densely populated, small-area areas, enabling long-term surveillance.

[0047] 2. Remote patrol mode: The operator remotely controls the unmanned vehicle or a single drone in real time through the terminal to perform close reconnaissance or specific target tracking tasks.

[0048] 3. Autonomous Patrol by Unmanned Vehicles: Unmanned vehicles autonomously travel along preset paths and conduct inspections using onboard gimbals, while drones remain silently stationed in their nests throughout the process. Suitable for flat, unobstructed open areas.

[0049] 4. Autonomous patrol by drones: The unmanned vehicle stays in place as a relay station, and one or more drones take off from the vehicle-mounted drone nest and conduct aerial inspections of the area around the unmanned vehicle or complex environments (such as narrow alleys or densely built-up areas) along a predetermined route.

[0050] 5. Autonomous vehicle and drone hybrid patrol mode: The unmanned vehicle travels along a planned route, while the drone swarm performs collaborative operations according to mission requirements. This includes two sub-modes: vehicle-accompanying flight (drones maintain a specific formation around the unmanned vehicle, providing a moving aerial view) and fixed-point blind spot filling / advanced reconnaissance (drones take off from the vehicle-mounted pod in shifts to conduct rapid close-range reconnaissance of blind spots, building-obscured areas, or predetermined points of interest in front of the unmanned vehicle, and transmit the intelligence back to the unmanned vehicle and terminal in real time).

[0051] The ground remote control terminal loads map information of the area to be inspected, and combines it with the user-inputted task objectives. A multi-parameter decision algorithm is then used to determine the patrol task mode and patrol formation parameters, including: S11: Load map information of the area to be inspected, and combine it with the task objectives input by the user. Then, confirm the inspection task mode through a multi-parameter decision algorithm and decision rules. The map information includes the outlines, heights, and road network information of buildings within the inspection area. S12, determine the patrol formation parameters based on the patrol mission mode and road condition information.

[0052] Specifically, step S11 includes: S111, based on the map information of the area to be inspected, determine the area of ​​the inspection site, the terrain complexity index, and the density of obstructions; Specifically, the input parameters for the decision-making algorithm are: the area A of the inspection site, the terrain complexity index T, and the density of obstructions. .

[0053] Terrain complexity index Defined as a weighted combination of slope factor and vegetation factor: ; in In typical scenarios, it is advisable to (Can be adjusted according to actual needs).

[0054] Slope factor Based on DEM data from a high-precision map, calculate the slope value (unit: degrees) of grid points within the region. Slope factor Defined as a slope exceeding the maximum climbable slope of an autonomous vehicle. Area percentage:

[0055] in 15° is acceptable. Let A be the area of ​​the inspection site.

[0056] vegetation factors Based on vegetation cover or lidar point cloud classification, calculate the percentage of vegetation areas with a height exceeding 2m: ; in, For the area of ​​vegetation with a height exceeding 2m, the final The larger the value, the more complex the terrain.

[0057] Density of obstructions Defined as a weighted combination of building shading density and tree shading density: ; in Typical scenarios =0.7 (building obstruction density weight).

[0058] Building obstruction density Based on building outline and height data, calculate the percentage of projected area and weight it according to height:

[0059] in The first in the area to be inspected The floor area of ​​each building, The first in the area to be inspected The height of the building For reference height (e.g., 20m). This represents the total area to be inspected. (Final) The larger the value, the more severe the occlusion.

[0060] S112 determines the patrol task mode based on the inspection site area, terrain complexity index, obstruction density, and decision rules; among which, the patrol task modes include autonomous mixed patrol of drones and unmanned vehicles, autonomous patrol of drones, autonomous patrol of unmanned vehicles, and remote-controlled patrol.

[0061] Specifically, the decision-making rules include: R1: If the area of ​​the inspection site is greater than the preset area threshold; and the terrain complexity index is greater than the preset complexity threshold, or the density of obstructions is greater than the preset obstruction density threshold, then the inspection task mode is recommended to be a mixed autonomous patrol of drones and unmanned vehicles. Specifically, if A > area threshold (e.g., 10000㎡) and (T > complexity threshold or O > occlusion threshold), then “autonomous hybrid patrol” is recommended.

[0062] R2: If the area of ​​the inspection site is not greater than the preset area threshold, and the density of obstructions is greater than the preset obstruction density threshold, then the inspection task mode is recommended to be drone autonomous patrol. Specifically, if A (e.g., 5000㎡) If the area threshold is met and O > occlusion threshold Th_O_high, then "autonomous drone patrol" is recommended.

[0063] R3: If the area of ​​the inspection site is greater than the preset area threshold, and the terrain complexity index is not greater than the preset complexity threshold, then the inspection task mode is recommended to be autonomous patrol by unmanned vehicles. Specifically, if A (e.g., 2000㎡) Area threshold and T If the complexity threshold is considered, then "autonomous patrol by unmanned vehicles" is recommended.

[0064] R4: If the user specifies close-range reconnaissance, the patrol mission mode is forcibly set to remote patrol.

[0065] Specifically, all of the above thresholds are configurable parameters of the system and can be adjusted according to different application scenarios.

[0066] Specifically, step S12, which determines the patrol mission formation parameters based on the road condition information of the patrol mission mode, includes: S121, determine the number of drones to be taken off based on the patrol mission mode and the area to be inspected; Specifically, the number of drones N to be launched is automatically calculated based on the patrol mission mode and mission scope. For example, in mixed patrols, N = min(ceil(A / unit coverage area), 6), that is, one drone is allocated for every 2000㎡, where A is the total area to be inspected.

[0067] S122, when the patrol mission mode is autonomous mixed patrol of drones and unmanned vehicles or autonomous patrol of drones, the formation type, formation distance parameter and flight altitude of the drones are selected according to the current road conditions of the area to be inspected and the formation parameter correspondence; the formation parameter correspondence stores the correspondence between different road conditions and drone formations.

[0068] Specifically, for autonomous mixed patrols using drones and unmanned vehicles, or autonomous drone patrols, the formation is dynamically selected and formation type parameters are generated based on road conditions. Road condition types are determined by the path planning module, which segments the path. The correspondence rules for formation parameters are shown in the table below:

[0069] The formation distance parameter (relative distance) can be dynamically adjusted based on factors such as road width, curve radius, and safety clearance. Longitudinal distance D long Considering the reaction time and safe distance of the drone during emergency braking of the autonomous vehicle, take... ,in For drone speed, For response delay (e.g., 2 seconds). The minimum safe distance is 10m, with a typical value of 30~50m.

[0070] Lateral distance D lat According to road width and drone wingspan set up, Typical value is 20~30m.

[0071] in, To add extra buffer space when calculating lateral distance, to ensure that the drone maintains a sufficient safe distance from road edges, obstacles or other drones during flight.

[0072] Height parameters: Accompanying altitude of drones : Usually 30m is taken. If there are tall buildings on both sides (height H) obs ),but If it is a low shrub area, the height can be reduced to 15-20m.

[0073] If it is in independent patrol mode, its flight altitude Calculate based on the area of ​​the mission area and the camera's field of view to ensure that the coverage width meets the requirements, typically 50~100m.

[0074] Takeoff / Return Point: Based on the current position and path planning of the unmanned vehicle, set a dynamic takeoff point (such as 100m before the unmanned vehicle enters a complex area) and a return point (to facilitate the early return of the unmanned vehicle).

[0075] The ground control terminal encapsulates the mission plan (patrol mission mode and patrol mission formation parameters) into JSON format instructions and distributes them to the unmanned vehicles via the ad hoc network. The instructions include: json { "mission_id": "M001", "mode": "hybrid_patrol", "drone_count": 2, "formation": "line", "patrol_route": [[x1,y1], [x2,y2], ...], "uav_tasks": [ {"id": 1, "type": "escort", "offset_x": 50, "offset_y": 0, "height":30}, {"id": 2, "type": "area_scan", "polygon": [[...]], "height": 80} ] } The autonomous vehicle's main control computer parses commands and calls the corresponding function modules: Motion control module: Generates local paths based on patrol_route and combines LiDAR data and camera image information for real-time obstacle avoidance.

[0076] PTZ control module: Set automatic inspection preset positions (such as left and right swing scanning).

[0077] Nest control module: Prepares for the takeoff of a designated UAV, and obtains the relative position information of the UAV and the unmanned vehicle.

[0078] Specifically, the unmanned vehicle schedules the drone swarm based on the real-time battery level, patrol mission mode, and patrol mission formation parameters of each drone within the swarm, including: S21, based on the number N drones to be launched, select the N drones with the highest battery level from those with battery levels greater than the first battery level threshold and launch them; if the number of drones with battery levels greater than the first battery level threshold is less than N, then launch them from drones with battery levels greater than the second battery level threshold. Specifically, the unmanned vehicle executes a scheduling algorithm based on the real-time battery level (obtained via CAN bus), health status, patrol mission mode, and patrol mission formation parameters of each drone in the hive: Pre-flight allocation: Based on the N value, select the N drones with the highest battery level from those with ≥80% battery (first battery threshold) and assign them corresponding platform IDs. If there are insufficient drones with ≥80% battery level, supplement them from those with ≥50% battery level (second battery threshold) and adjust their mission type (reduce flight speed, shorten flight range).

[0079] S22, During the mission, the unmanned vehicle continuously monitors the remaining battery power of the aerial drones. When the battery power of any aerial drone falls below a preset third battery threshold, the drone rotation process is triggered. The first, second, and third battery thresholds decrease sequentially. Specifically, the dynamic rotation strategy involves the autonomous vehicle continuously monitoring the remaining battery power of the drones in the air during the mission. The autonomous vehicle maintains a drone status table in real time, including ID, battery level, location, and mission type. When the battery level of any drone in the air falls below a preset low battery threshold (e.g., 30%, a third battery threshold), the autonomous vehicle triggers the rotation process by maintaining the drone status table, including ID, battery level, location, and mission type. (1) Confirmation and preparation: The unmanned vehicle selects the standby drone with the highest battery level (≥80%) from the drone status table and calculates and assigns the task area / formation parameters to be replaced.

[0080] (2) If the original UAV is performing an area scanning task, the remaining unscanned area is allocated to the new UAV. If the remaining area exceeds the single-run capability of the new UAV, the scanning path is replanned based on its battery power (e.g., the area is divided, and only the part that can be completed is allocated).

[0081] (3) If the original UAV is performing a chariot escort mission, the new UAV should fly to the original UAV's formation position. At the same time, the formation position should be adjusted according to the new UAV's battery status: if the battery is high (≥80%), it should completely take over the original position; if the battery is slightly low (e.g., 50%), it should be assigned to a formation position closer to the UAV (e.g., reduced longitudinal distance, lower altitude) to save energy. The formation position adjustment is based on the energy consumption model, which estimates the hovering power and flight power, calculates the energy required to fly from the current position to the new formation position and the power required to maintain that position, and ensures that the estimated remaining flight time is greater than the mission cycle.

[0082] (4) Command issuance: The unmanned vehicle simultaneously sends the CMD_PREPARE_RTH (prepare to return) command to the low-battery UAV or the CMD_LAND command to the target UAV via the self-organizing network (if the UAV is performing a mission, send CMD_ABORT_TASK first), and sends the CMD_TAKEOFF command to the standby UAV, or sends the CMD_TAKEOFF command to the standby UAV, with the route / formation parameters for taking over the mission attached; the nacelle automatically opens the hatch, and the standby UAV unlocks and takes off.

[0083] (5) Responsibility Confirmation: After the standby UAV takes off and arrives at the designated retrieval position, it sends a STATUS_TAKEOVER_READY status to the UAV. After the UAV confirms, it instructs the original low-battery UAV to execute the CMD_LAND command to ensure seamless mission coverage.

[0084] (6) Return and recovery: During the return process of the low-battery UAV, the UAV continuously sends its relative coordinates to the nest to assist in guidance. After landing, the nest closes the hatch, starts charging, and updates the status to "charging" via the CAN bus.

[0085] S23 adjusts the drone's mission type based on its battery level.

[0086] Specifically, when the battery level is ≥80%, a large-area scanning task is assigned, with a flight speed ≤8m / s, an altitude ≤80m, and the camera in wide-angle mode.

[0087] When the battery level is between 30% and 80%, assign a medium-range inspection task. The flight speed should be between 5m / s and the altitude between 50m. The camera should be in zoom mode to conduct a detailed inspection of key areas.

[0088] When the battery level is less than 30%, only short-range blind spot filling tasks (such as flying around the unmanned vehicle) will be assigned, with a flight speed of ≤3m / s and an altitude of ≤20m, and the vehicle will be forced to return to base within 5 minutes.

[0089] Preferably, target detection and behavior recognition models are respectively installed in the drone and the unmanned vehicle. Drones process one type of data source: Video stream captured by drone (dual-light gimbal, equipped with visible light camera and infrared sensor for aerial reconnaissance): also detected by YOLOv11, but due to the higher field of view of the drone, the detection of small targets is optimized (by adding a shallow feature layer).

[0090] Autonomous vehicles process three types of data sources: Vehicle-mounted PTZ video stream: The video stream is pulled via RTSP, preprocessed, and then input into the YOLOv11 model to detect targets such as people, vehicles, and fire points; skeletal key points are extracted from the personnel area and input into the ST-GCN model to identify behaviors such as fighting and falling; and high-temperature anomalies (fires) are detected through thermal imaging.

[0091] The video stream transmitted back by the drone was also detected by YOLOv11, but due to the higher field of view of the drone, the detection of small targets was optimized (by adding a shallow feature layer).

[0092] LiDAR point cloud: used to help determine the spatial distribution density when people gather.

[0093] When both the autonomous vehicle and the drone detect anomalies in the same area, the overall confidence scores of the autonomous vehicle and the drone are compared, and the detection result with the highest overall confidence score is selected as the final alarm basis. If either the autonomous vehicle or the drone detects anomalies in the same area, and the overall confidence score exceeds a preset overall confidence score threshold, an alarm is triggered. The calculation method for the overall confidence score is as follows: ; Among them, C total To determine the overall confidence level, C conf The original confidence level is (0~1). The original confidence weights, The visual quality index. As the weight of the perspective quality index, For infrared characteristic intensity (e.g., temperature anomalies normalized to (0~1)). The infrared feature intensity weights are M, where M is the normalized velocity of the target motion. The target motion normalized velocity weights (0~1) are used, and Z represents the region sensitivity. For region sensitivity weights; where, the viewpoint quality index is the normalized ratio (0~1) of the target imaging area to the total image area; region sensitivity, if the target is located in a preset high-sensitivity area (such as a restricted area, entrance / exit), then... ,otherwise .

[0094] Specifically, a confidence-based decision-level fusion is employed, incorporating a comprehensive score based on multiple influencing factors. A comprehensive confidence score is calculated for each detection result. Weights can be determined through expert experience or machine learning, and the sum of the weights must be 1. Example weights (normal daytime weather): =0.4, =0.2, =0.1, =0.2, =0.1. Adjustable at night. It is 0.3. It dropped to 0.1.

[0095] Alarm information (time, location, type, screenshot) is stored locally and uploaded to the remote command center via a 4G / 5G module (if equipped).

[0096] To achieve safe recovery of drones on mobile / stationary unmanned vehicles, a dual guidance mechanism can be adopted, namely long-range coarse guidance + short-range fine guidance. Long-range coarse guidance (relative distance > 10m): The unmanned vehicle and the drone exchange RTK-GPS / BeiDou differential data in real time through an ad hoc network to obtain centimeter-level relative positions (Δx, Δy, Δz). The drone flight controller sets the target point to the current coordinates of the unmanned vehicle plus a preset "home point offset" (e.g., 5m directly above), and uses a PID controller to navigate to the vicinity of that point.

[0097] Close-range precision guidance (relative distance ≤ 10m): The UAV's downward-facing camera captures a QR code on the platform. The onboard vision processing unit runs a QR code recognition algorithm (such as OpenCV's aruco module) to calculate the UAV's 3D positional deviation (Δx', Δy', Δz') and yaw angle deviation Δyaw relative to the center of the QR code. Since the QR code coordinate system and the platform coordinate system are already calibrated, this can be converted into the precise pose of the UAV relative to the platform's landing location.

[0098] The drone's flight control system makes fine adjustments based on this deviation and executes the landing procedure. During the process, visual information is updated at a frequency of 50Hz to ensure real-time performance.

[0099] Preferably, during the close-range fine guidance phase, if the onboard vision processing unit fails to effectively recognize the drone's QR code for one consecutive second (or a cumulative total of three frames), the visual guidance is deemed to have failed. The drone should immediately hover, ascend to an altitude of 5 meters, re-enter the long-range coarse guidance phase, and attempt to descend again. If three consecutive attempts fail, the drone should switch to an alternate landing point (an open area near the drone) according to the preset emergency procedure and issue an alarm.

[0100] Furthermore, once the drone's landing gear contacts the platform's guide slot, the electromechanical structure guides it to slide into the center, and the electric locking device automatically activates, securing the drone to the platform. The rotor automatic folding mechanism operates, retracting the blades into the correct position. The drone's landing platform descends back to its lowest position. The charging probe connects to the drone's contacts, and the drone's CAN bus queries the drone's battery information; if the battery level is less than 100%, charging begins. The drone's industrial control computer records the recovery results and updates the drone's status to "Ready (Charging)". The drone's hatch automatically closes, providing dust and rain protection.

[0101] In this solution, communication between the UAV swarm, unmanned vehicle, and ground control terminal can adopt the self-organizing MESH protocol, with an adaptive operating frequency band of 2.4GHz / 5.8GHz, supporting multi-hop relay, ensuring that data can still be relayed back through other nodes when the UAV flies out of line of sight.

[0102] During transmission, critical commands (takeoff, landing, emergency stop) are transmitted reliably using the TCP protocol; the video stream uses the UDP / RTP protocol, allowing for retransmission of key frames in case of packet loss.

[0103] During application, a unified JSON command set can be defined, including command type, target ID, parameters, timestamp, and checksum. Upon receiving a command, the autonomous vehicle or drone must reply with an ACK and, after execution, reply with the execution result.

[0104] During state synchronization, the unmanned vehicle periodically (1Hz) broadcasts its own status (position, speed, mode, hive door status, and platform occupancy); the drone periodically (5Hz) reports its own status (ID, location, battery level, mission stage, anomaly code, and current image acquisition mode). The ground control terminal subscribes to this information and updates its display in real time.

[0105] Preferably, if the drone loses contact with the unmanned vehicle for more than 10 seconds, the drone will automatically hover and attempt to reconnect. If it does not reconnect within 30 seconds, it will initiate an autonomous return-to-home procedure (flying to the last known location of the unmanned vehicle or a preset emergency landing point).

[0106] If the landing fails, meaning the drone's first landing attempt is unsuccessful (visual recognition failure or excessive deviation), it will automatically ascend 5 meters and attempt visual recognition again, with a maximum of 3 attempts. If all attempts fail, it will switch to an alternate landing point (an open area near the drone) and issue an alarm.

[0107] If the drone's battery level drops below the 10% emergency landing threshold, it will be forced to land at the nearest location, regardless of whether an instruction has been received (it will attempt to land if it is near an unmanned vehicle, otherwise it will seek an open area to land), and send its last location information.

[0108] In summary, the present invention has the following technical advantages over the prior art: (1) Multi-mode switching with strong adaptability: Through multiple modes, the system covers the entire scene from silent guarding to air-ground collaboration. Combined with rule-based decision-making algorithms, the system’s adaptability to different terrains and task requirements is greatly improved.

[0109] (2) Cluster scheduling, efficiency is doubled: Through dynamic power management and a rotation strategy with confirmation mechanism, the pain point of short battery life of single UAVs is solved, realizing uninterrupted relay patrol of "one vehicle and multiple UAVs" and ensuring seamless task coverage.

[0110] (3) Air-ground fusion for accurate identification: It integrates multi-source data from vehicle-mounted high-altitude vision and UAV low-altitude perspective, and adopts a confidence-based decision-level fusion algorithm to significantly improve the detection rate of abnormal events in complex environments and reduce false alarms and missed reports.

[0111] (4) High reliability recovery: The dual guidance mechanism of long-range RTK coarse guidance + close-range visual fine guidance, combined with visual redundancy processing, ensures that the UAV lands accurately on the dynamic platform with a success rate of ≥99.5%.

[0112] (5) Fully autonomous operation: From mission planning, takeoff inspection, rotation and automatic recovery and charging, the entire process does not require human intervention, realizing truly unmanned autonomous inspection.

[0113] (6) Robust anomaly handling: A comprehensive emergency mechanism is designed for situations such as communication interruption, landing failure, and power shortage to ensure the safety of the system and equipment.

[0114] In this invention, the ground control terminal loads map information of the area to be inspected, combines it with the user-inputted task objectives, and uses a multi-parameter decision algorithm to confirm the inspection task mode and formation parameters. The inspection task mode and formation parameters are then encapsulated into instructions and sent to the unmanned vehicle (UAV) via a self-organizing network. The UAV parses the instructions, schedules the UAV cluster based on the real-time battery level of each UAV in its nest, the inspection task mode, and the formation parameters, and performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the UAVs. The UAVs acquire image information of the objects to be inspected and transmit the image information back to the UAV. This effectively solves the problem of low inspection efficiency and reliability caused by existing technologies, and significantly improves inspection efficiency and reliability.

[0115] In the technical solution of this invention, the area of ​​the inspection site, the terrain complexity index, and the density of obstructions are determined based on the map information of the area to be inspected. Based on the area of ​​the inspection site, the terrain complexity index, the density of obstructions, and the decision rules, the inspection task mode is determined. The inspection task formation parameters are determined according to the inspection task mode and road condition information. Different inspection task modes can be flexibly selected according to the actual situation of the area to be inspected, which improves the adaptability of the inspection to different terrains and task requirements.

[0116] The technical solution of this invention adjusts the mission type of the drone according to its different battery levels. Through dynamic battery management and rotation strategy, it solves the problem of short battery life of a single drone and realizes uninterrupted relay patrol of "one vehicle and multiple drones", ensuring seamless mission coverage.

[0117] In this invention, target detection and behavior recognition models are set up in both the drone and the unmanned vehicle. When both the drone and the unmanned vehicle detect anomalies in the same area, the overall confidence scores of the drone and the unmanned vehicle are compared, and the detection result with the highest overall confidence score is selected as the final alarm basis. If the drone or the unmanned vehicle detects anomalies in the same area and the overall confidence score exceeds the preset overall confidence score threshold, an alarm is triggered. Multi-source data from the vehicle's high-altitude view and the drone's low-altitude view are fused, and a decision-level fusion algorithm based on confidence score is adopted to significantly improve the detection rate of abnormal events in complex environments and reduce false alarms and missed alarms.

[0118] In the technical solution of this invention, when the relative position is greater than a preset distance threshold, the UAV sets the landing target point to the current coordinates of the unmanned vehicle plus a preset return point offset; when the relative position is not greater than the preset distance threshold, the UAV performs the landing operation based on the position deviation and yaw angle deviation relative to the center of the nest QR code; through the dual guidance mechanism of long-range coarse guidance + close-range visual fine guidance, the UAV is ensured to land accurately on the unmanned vehicle dynamic platform.

[0119] Example 2 like Figure 2 As shown, the technical solution of the present invention also provides a ground-air cooperative unmanned patrol method, which is implemented based on a ground-air cooperative unmanned patrol system in Embodiment 1, and includes: S1, the ground control terminal loads the map information of the area to be inspected, combines it with the task objectives input by the user, and confirms the patrol task mode and patrol task formation parameters through a multi-parameter decision algorithm. The patrol task mode and patrol task formation parameters are then encapsulated into instructions and sent to the unmanned vehicle through the self-organizing network. S2, the unmanned vehicle parsing command, schedules the drone cluster based on the real-time battery level, patrol mission mode, and patrol mission formation parameters of each drone in the nest. S3: The drone acquires image information of the object to be inspected and transmits the image information back to the unmanned vehicle. S4: The unmanned vehicle performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the drone.

[0120] Preferably, it may also include step S5, where the image information acquired by the UAV is used for target detection and behavior recognition; Preferably, it may also include step S6, determining whether to issue an alarm based on the detection of targets in the same area by the unmanned vehicle and the drone.

[0121] Specifically, when both the autonomous vehicle and the drone detect anomalies in the same area, the overall confidence scores of the autonomous vehicle and the drone are compared, and the detection result with the highest overall confidence score is selected as the final alarm basis; if the autonomous vehicle or the drone detects anomalies in the same area and the overall confidence score exceeds the preset overall confidence score threshold, an alarm is triggered.

[0122] The execution process of each step is the same as that in Embodiment 1, and will not be described again in this embodiment.

[0123] In this invention, the ground control terminal loads map information of the area to be inspected, combines it with the user-inputted task objectives, and uses a multi-parameter decision algorithm to confirm the inspection task mode and formation parameters. The inspection task mode and formation parameters are then encapsulated into instructions and sent to the unmanned vehicle (UAV) via a self-organizing network. The UAV parses the instructions, schedules the UAV cluster based on the real-time battery level of each UAV in its nest, the inspection task mode, and the formation parameters, and performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the UAVs. The UAVs acquire image information of the objects to be inspected and transmit the image information back to the UAV. This effectively solves the problem of low inspection efficiency and reliability caused by existing technologies, and significantly improves inspection efficiency and reliability.

[0124] In the technical solution of this invention, the area of ​​the inspection site, the terrain complexity index, and the density of obstructions are determined based on the map information of the area to be inspected. Based on the area of ​​the inspection site, the terrain complexity index, the density of obstructions, and the decision rules, the inspection task mode is determined. The inspection task formation parameters are determined according to the inspection task mode and road condition information. Different inspection task modes can be flexibly selected according to the actual situation of the area to be inspected, which improves the adaptability of the inspection to different terrains and task requirements.

[0125] The technical solution of this invention adjusts the mission type of the drone according to its different battery levels. Through dynamic battery management and rotation strategy, it solves the problem of short battery life of a single drone and realizes uninterrupted relay patrol of "one vehicle and multiple drones", ensuring seamless mission coverage.

[0126] In this invention, target detection and behavior recognition models are set up in both the drone and the unmanned vehicle. When both the drone and the unmanned vehicle detect anomalies in the same area, the overall confidence scores of the drone and the unmanned vehicle are compared, and the detection result with the highest overall confidence score is selected as the final alarm basis. If the drone or the unmanned vehicle detects anomalies in the same area and the overall confidence score exceeds the preset overall confidence score threshold, an alarm is triggered. Multi-source data from the vehicle's high-altitude view and the drone's low-altitude view are fused, and a decision-level fusion algorithm based on confidence score is adopted to significantly improve the detection rate of abnormal events in complex environments and reduce false alarms and missed alarms.

[0127] In the technical solution of this invention, when the relative position is greater than a preset distance threshold, the UAV sets the landing target point to the current coordinates of the unmanned vehicle plus a preset return point offset; when the relative position is not greater than the preset distance threshold, the UAV performs the landing operation based on the position deviation and yaw angle deviation relative to the center of the nest QR code; through the dual guidance mechanism of long-range coarse guidance + close-range visual fine guidance, the UAV is ensured to land accurately on the unmanned vehicle dynamic platform.

[0128] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A ground-air cooperative unmanned patrol system, characterized in that, include: The system comprises an unmanned vehicle (UAV), a UAV swarm, and a ground control terminal. The ground control terminal loads map information of the area to be inspected, combines it with user-inputted task objectives, and uses a multi-parameter decision algorithm to determine the inspection task mode and formation parameters. It then encapsulates these parameters into instructions and sends them to the UAV via a self-organizing network. The UAV parses these instructions, schedules the UAV swarm based on real-time battery levels, inspection task mode, and formation parameters, and performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and the UAVs. The UAVs acquire image information of the objects to be inspected and transmit this information back to the UAV.

2. The ground-air cooperative unmanned patrol system according to claim 1, characterized in that, Load the map information of the area to be inspected, and combine it with the user-inputted task objectives. A multi-parameter decision algorithm is then used to determine the patrol task mode and specific patrol task formation parameters, including: The system loads map information of the area to be inspected, combines it with the user-inputted task objectives, and confirms the inspection task mode through a multi-parameter decision algorithm and decision rules. The map information includes the outlines, heights, and road network information of buildings within the inspection area. The patrol formation parameters are determined based on the patrol mission mode and road condition information.

3. The ground-air cooperative unmanned patrol system according to claim 2, characterized in that, Based on the user-inputted task objectives, the patrol task mode is determined through a multi-parameter decision algorithm and decision rules, specifically including: Based on the map information of the area to be inspected, determine the area of ​​the inspection site, the terrain complexity index, and the density of obstructions; Based on the area of ​​the inspection site, the terrain complexity index, the density of obstructions, and the decision-making rules, the inspection task mode is determined; among them, the inspection task modes include autonomous mixed patrol of drones and unmanned vehicles, autonomous patrol of drones, autonomous patrol of unmanned vehicles, and remote-controlled patrol.

4. The ground-air cooperative unmanned patrol system according to claim 3, characterized in that, Decision-making rules include: If the area of ​​the inspection site is greater than the preset area threshold, and the terrain complexity index is greater than the preset complexity threshold, or the density of obstructions is greater than the preset obstruction density threshold, then the inspection task mode is recommended to be a mixed autonomous patrol of drones and unmanned vehicles. If the area of ​​the inspection site is not greater than the preset area threshold, and the density of obstructions is greater than the preset obstruction density threshold, then the inspection task mode is recommended to be drone autonomous patrol. If the area of ​​the inspection site is greater than the preset area threshold, and the terrain complexity index is not greater than the preset complexity threshold, then the inspection task mode is recommended to be autonomous patrol by unmanned vehicles. If the user specifies close-range reconnaissance, the patrol mission mode is forcibly set to remote patrol.

5. The ground-air cooperative unmanned patrol system according to claim 3, characterized in that, The specific method for calculating the terrain complexity index is as follows: Based on the DEM data of the map, the slope value of the grid points in the area to be inspected is calculated, and the slope factor in the area to be inspected is determined based on the slope value; where the slope factor is the proportion of the area in the area to be inspected where the slope value exceeds the maximum climbable slope of the unmanned vehicle. The vegetation factor within the area to be inspected is determined based on the vegetation coverage. The vegetation factor is the percentage of the area of ​​vegetation with a height exceeding a preset height threshold. The slope factor and vegetation factor in the area to be inspected are weighted and summed to obtain the terrain complexity index. The specific method for calculating the density of obstructions is as follows: in The first in the area to be inspected The floor area of ​​each building, The first in the area to be inspected The height of the building For reference height, This represents the total area of ​​the area to be inspected.

6. The ground-air cooperative unmanned patrol system according to claim 2, characterized in that, The specific parameters for determining the patrol formation based on road condition information and patrol mission mode include: The number of drones to be taken off is determined based on the patrol mission mode and the area to be inspected. When the patrol mission mode is a mixed autonomous patrol of drones and unmanned vehicles or an autonomous patrol of drones, the formation type, formation distance parameters, and flight altitude of the drones are selected according to the current road conditions of the area to be inspected and the correspondence of formation parameters; the correspondence of formation parameters stores the correspondence between different road conditions and drone formations.

7. The ground-air cooperative unmanned patrol system according to claim 1, characterized in that, The scheduling of the drone swarm based on the real-time battery level, patrol mission mode, and patrol mission formation parameters of each drone within the swarm specifically includes: Based on the number N drones to be launched, select the N drones with the highest battery levels from those with battery levels greater than the first battery threshold and launch them; if the number of drones with battery levels greater than the first battery threshold is less than N, then launch them from drones with battery levels greater than the second battery threshold. During the mission, the unmanned vehicle continuously monitors the remaining battery power of the aerial drones. When the battery power of any aerial drone falls below a preset third battery threshold, a drone rotation process is triggered. The first, second, and third battery thresholds decrease sequentially. Adjust the drone's mission type according to its battery level.

8. The ground-air cooperative unmanned patrol system according to claim 1, characterized in that, Target detection and behavior recognition models are set up in both drones and unmanned vehicles. When both drones and unmanned vehicles detect anomalies in the same area, the overall confidence scores of the drones and unmanned vehicles are compared, and the detection result with the highest overall confidence score is selected as the final alarm basis. If either the drone or the unmanned vehicle detects anomalies in the same area and the overall confidence score exceeds a preset overall confidence threshold, an alarm is triggered. The calculation method for the overall confidence score is as follows: ； Among them, C total To determine the overall confidence level, C conf The original confidence level. The original confidence weights, The quality index of the viewpoint. As the weight of the perspective quality index, Infrared characteristic intensity, The infrared feature intensity weights are M, where M is the normalized velocity of the target motion. Z represents the normalized velocity weights for the target motion, and Z represents the region sensitivity. The region sensitivity weight is denoted by ; where the viewpoint quality index is the proportion of the target imaging area to the total image area.

9. The ground-air cooperative unmanned patrol system according to claim 1, characterized in that, The autonomous vehicle and the drone obtain their relative positions through an ad hoc network. When the relative position is greater than a preset distance threshold, the drone sets the landing target point to the current coordinates of the autonomous vehicle plus a preset home point offset. When the relative position is not greater than the preset distance threshold, the drone captures the QR code on the drone's nest platform and calculates the drone's position deviation and yaw angle deviation relative to the center of the QR code. Based on the position deviation and yaw angle deviation relative to the center of the QR code, the drone performs the landing operation.

10. A ground-air cooperative unmanned patrol method, characterized in that, Based on any one of claims 1-9, the ground-air cooperative unmanned patrol system includes: The ground control terminal loads map information of the area to be inspected, combines it with the task objectives input by the user, and uses a multi-parameter decision algorithm to confirm the patrol task mode and patrol task formation parameters. The patrol task mode and patrol task formation parameters are then encapsulated into instructions and sent to the unmanned vehicle through the self-organizing network. The unmanned vehicle interprets commands and schedules the drone swarm based on the real-time battery level, patrol mission mode, and patrol mission formation parameters of each drone in the drone nest. The drone acquires image information of the object to be inspected and transmits the image information back to the unmanned vehicle; The unmanned vehicle performs target detection and behavior recognition based on image data acquired by the vehicle-mounted gimbal and image information acquired by the drone.

Images