A complex terrain-oriented unmanned aerial vehicle swarm cooperative search and positioning method

Abstract

This application relates to the field of emergency search and rescue technology, and discloses a method for collaborative search and positioning of UAV swarms in complex terrain. The method includes: S1: The ground control station retrieves 3D terrain data of a target urban mountain park from a pre-stored park geographic information database, and simultaneously receives initial information about a missing target sent by a reporting terminal via a mobile communication network. The two types of data are then correlated and stored in a local task database. S2: The ground control station calls a terrain analysis module to process the 3D terrain data from S1, calculating the real-time slope and vegetation coverage of each area. Through refined area division based on slope and vegetation coverage, combined with precise adaptation of UAV type and sub-area characteristics, targeted allocation of search resources is achieved, avoiding resource waste in conventional road sections and complex areas. Furthermore, through global path planning and multi-UAV collaborative coverage, the effective search coverage rate in complex terrain is improved, significantly shortening the discovery time of missing targets.

Description

Technical Field

[0001] This invention relates to the field of emergency search and rescue technology, specifically to a method for collaborative search and positioning of unmanned aerial vehicle (UAV) swarms in complex terrain. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarm collaborative search and positioning technology is a technical means that uses multiple UAVs to work together to quickly detect and locate targets in a specific area. It is widely used in emergency search and rescue, field security, geological exploration and other fields.

[0003] In existing technologies, collaborative search and localization of UAV swarms mostly employs preset path cruising or simple area division. Typically, a ground control station pre-plans the search area, distributing tasks evenly among the UAVs, relying on GPS positioning and single sensors such as vision and infrared for target detection. The collaborative approach is primarily centralized control, where the ground station receives data and issues commands uniformly, with limited information exchange between UAVs. This method is generally sufficient for scenarios with relatively flat terrain and minimal obstruction (such as open plains and conventional sites), and its operation is simple and easy to implement.

[0004] However, in complex terrains (such as urban mountain parks and suburban woodlands, which combine regular roads with steep slopes and dense forests), existing technologies have significant limitations: on the one hand, the area division is coarse and does not take into account terrain features (such as slope and vegetation coverage) for differentiated processing, resulting in repeated searches by drones on flat road sections and insufficient coverage on complex road sections (steep slopes and dense forests), leading to unreasonable resource allocation; on the other hand, the types of drones and sensor configurations lack specificity, and general-purpose equipment is difficult to adapt to the detection needs of complex terrains (such as visual sensor failure under dense forest cover and the accumulation of positioning errors in steep slope terrain), resulting in low overall search efficiency and difficulty in meeting the needs of rapid target discovery in scenarios such as emergency search and rescue. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a collaborative search and positioning method for UAV swarms in complex terrain, solving the problem of slow target detection in scenarios such as emergency search and rescue.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for cooperative search and localization of unmanned aerial vehicle (UAV) swarms in complex terrain, comprising the following steps:

[0007] S1: The ground control station retrieves the three-dimensional terrain data of the target city mountain park from the pre-stored park geographic information database, and at the same time receives the initial information of the missing target sent by the reporting terminal through the mobile communication network, and stores the two types of data together in the local task database;

[0008] S2: The ground control station calls the terrain analysis module to process the three-dimensional terrain data in S1, calculates the real-time slope value and vegetation coverage of each area, and divides the areas that simultaneously meet the conditions of slope <15° and vegetation coverage <30% into regular road section areas, and divides the areas that meet the conditions of slope ≥15° or vegetation coverage ≥30% into non-regular road section areas. In the non-regular road section areas, dense vegetation sub-areas with coverage ≥50% and steep slope sub-areas with slope ≥25° are further marked, thereby obtaining the area division results. Based on the area division results, the overall area is generated, and the area boundary and grid origin geodetic coordinates are determined.

[0009] S3: Based on the area division results in S2, the ground control station configures the types of drones in the drone swarm into three categories and sends task deployment instructions to the drone swarm: instructing the first type of drones to go to the dense vegetation sub-area, instructing the second type of drones to go to the steep slope sub-area, and instructing the third type of drones to cover the regular road section area. After receiving the instructions, each drone sends a confirmation signal back to the ground control station through the preset frequency band.

[0010] S4: Based on the area, area boundary, and drone type allocated to each drone in S3, the ground control station generates a global search path for each drone and sends it out separately. Each drone in the drone swarm performs a search task within its allocated area, collecting environmental data and avoiding obstacles while transmitting the data back to the ground control station.

[0011] S5: During the search process based on S4, when any UAV detects a suspected target, it extracts the features of the suspected target and matches them with the initial information of the missing target in S1. If the match is successful, a suspected signal is triggered, and the ground control station instructs the UAVs in the mission cluster to gather towards the missing target to form a multi-angle detection array.

[0012] S6: Based on the multi-angle detection array formed by S5, each UAV in the detection array transmits detection data to the ground control station from different angles. After verifying that it is a lost target, the sensor data from different positions and perspectives are fused to calculate the target's absolute coordinates and eliminate positioning errors caused by terrain obstruction.

[0013] Preferably, the three-dimensional terrain data includes historical slope and vegetation distribution data, and the initial information of the missing target includes the last known location and physical characteristics.

[0014] Preferably, the terrain analysis module in S2 specifically includes:

[0015] S2.1: 3D Terrain Data Preprocessing

[0016] The 3D terrain data retrieved from S1 is preprocessed as follows:

[0017] Point cloud filtering: A statistical filtering algorithm is used to remove noise points and retain effective shape points;

[0018] Rasterization: The target area is divided into raster cells with a size of 1m × 1m, denoted as .

[0019] For row index, (for column indexes);

[0020] S2.2: Raster cell feature extraction;

[0021] S2.3: Multi-threshold region classification.

[0022] Preferably, S2.2 specifically includes:

[0023] S2.21: Calculate each grid cell slope :

[0024] The plane equation Z = ax + by + c is fitted using the least squares method.

[0025] Where: Z is the elevation value, x, y are the plane coordinates, and abc are the plane parameters;

[0026] Then through

[0027]

[0028] Convert plane inclination to slope value;

[0029] S2.22: Calculate each grid cell vegetation coverage :

[0030] The Normalized Difference Vegetation Index (NDVI) is calculated using remote sensing spectral information from three-dimensional terrain data.

[0031]

[0032] in For near-infrared reflectivity, Reflectance in the red light band (range -1 to 1);

[0033] vegetation coverage Obtained from NDVI conversion:

[0034] when hour, ;

[0035] when hour, ;

[0036] when hour, .

[0037] Preferably, the multi-threshold region classification specifically includes:

[0038] The regular road section area is for all those that meet the requirements. and The set of grid cells, in which The first slope threshold (valued at 15°) The first vegetation coverage threshold (value is 30%);

[0039] Unconventional road sections are all areas that meet the requirements. or A set of grid cells;

[0040] The densely vegetated sub-region is a non-standard road section area that meets the requirements. The set of grid cells, in which The second vegetation coverage threshold (value is 50%).

[0041] The steep slope area is a non-standard road section area that meets the requirements. The set of grid cells, in which This is the second slope threshold (valued at 25°);

[0042] in : The slope threshold that distinguishes between regular and non-regular road sections;

[0043] : The vegetation coverage threshold that distinguishes between regular and unregular road sections;

[0044] : Slope threshold for steep slope areas in unconventional road sections;

[0045] : The vegetation coverage threshold for densely vegetated sub-regions in unconventional road sections.

[0046] Preferably, the type configuration specifically includes: the first type of drone is equipped with a thermal imaging sensor and a high-definition camera; the second type of drone is equipped with a lidar and an inertial navigation module; the third type of drone is equipped with only a high-definition camera; and a unique device identifier and communication frequency band are assigned to each of the three types of drones.

[0047] Preferably, the process of calculating the absolute coordinates of the target in S6 is implemented by associating the region division algorithm in S2 with the grid cells, specifically including the following steps:

[0048] A1: Determine the grid cell where the missing target is located: Based on the location information transmitted back by the UAV from the detection array, match it to the grid cell divided in S2. Get the slope of the grid. Vegetation coverage and the geodetic coordinates of the grid origin;

[0049] A2: Extract the initial coordinates for region adaptation:

[0050] If the target is located in a steep slope area , (The second slope threshold in S2), using the lidar point cloud data of the second type of UAV as a reference, extracts the three-dimensional coordinates of the target relative to the lidar. ;

[0051] If the target is located in a densely vegetated sub-region ( , (The second vegetation coverage threshold in S2), fused with the thermal imaging plane coordinates of the first type of UAV. LiDAR elevation data of the second type of UAV , forming initial three-dimensional coordinates ;

[0052] If the target is located in a regular road section area ( and , , (As the first threshold in S2), the initial coordinates are obtained by fusing visual positioning data from a third type of UAV with data from the Chinese BeiDou satellite system. ;

[0053] A3: Coordinate correction based on regional features:

[0054] Steep slope areas: Utilize the slope of this grid in S2 ,right Terrain tilt correction is performed using the following formula: Planar coordinates corrected to ;

[0055] Densely vegetated sub-regions: based on thermal imaging plane coordinates Perform filtering. The higher the value, the larger the filter window;

[0056] A4: Convert to absolute coordinates: using the formula , , Calculate the absolute coordinates of the target;

[0057] in These are the corrected relative coordinates. For grid The coordinates of the origin.

[0058] Preferably, the elimination of positioning errors caused by terrain occlusion in S6 specifically involves the following logic, which is tied to the region division result in S2:

[0059] For steep slope areas Calling the plane equation used in S2 for slope calculation of this area To compensate for the drift error of the inertial navigation module of the second type of UAV, the compensation amount is related to the plane parameters. , The tilt is positively correlated;

[0060] Targeting densely vegetated sub-regions Based on S2 used for calculation NDVI value The reflection intensity in the filtered lidar point cloud is lower than The points retain valid data on penetration through vegetation;

[0061] For the boundary area between conventional and unconventional road sections: using the area boundary defined in S2 as a benchmark, the positioning data of UAVs on both sides are weighted and fused, and the grid data on both sides of the boundary are... , The smaller the difference, the more balanced the weights.

[0062] This invention provides a method for cooperative search and localization of unmanned aerial vehicle (UAV) swarms in complex terrain. It offers the following advantages:

[0063] 1. This invention achieves targeted allocation of search resources by using fine-grained regional division based on slope and vegetation coverage, combined with precise matching of UAV type and sub-region characteristics, avoiding resource waste in conventional road sections and complex areas. At the same time, through global path planning and multi-UAV collaborative coverage, the effective search coverage rate of complex terrain is improved, significantly shortening the discovery time of lost targets.

[0064] 2. The coordinate correction mechanism of the region feature association of the present invention, combined with multi-source sensor data fusion, effectively eliminates the positioning error caused by terrain occlusion, and controls the target positioning accuracy in complex areas within 1 meter, providing accurate target location guidance for search and rescue personnel. Attached Figure Description

[0065] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0066] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Example:

[0068] Please see the appendix Figure 1 This invention provides a method for cooperative search and localization of unmanned aerial vehicle (UAV) swarms in complex terrain, comprising the following steps:

[0069] Step 1: Data Acquisition and Related Storage

[0070] S1: Composition and Acquisition of 3D Terrain Data

[0071] Data sources: LiDAR point cloud data (acquisition accuracy: point cloud density ≥ 50 points / ㎡, elevation error ≤ 0.5m) pre-stored in the park's geographic information database, multispectral remote sensing images (resolution ≥ 1m × 1m, including near-infrared (NIR) and red (RED) bands), and coordinate system reference (using WGS84 geodetic coordinate system).

[0072] Data preprocessing: The raw data in the database needs to be denoised and aligned to ensure that the coordinate systems of data from different sources (LiDAR, remote sensing) are consistent.

[0073] S2: Transmission and processing of initial information on the missing target

[0074] Information composition: including the geodetic coordinates of the target's last known location (obtained through GPS / BeiDou positioning of the reporting terminal, with an accuracy of ≤10m), physical characteristics (quantifiable features such as the child's height, clothing color, and items carried, stored in structured data form), and time of loss of contact (accurate to the minute).

[0075] Transmission mechanism: The reporting terminal sends information to the ground control station through an encrypted communication link (such as HTTPS protocol of 4G / 5G, AES encryption of dedicated wireless data transmission) to ensure the security and integrity of data transmission.

[0076] S3: Data Association and Storage

[0077] Association logic: The 3D terrain data is bound to the target information through "time stamp + spatial coordinates" (such as the raster index in the terrain data corresponding to the last location of the target).

[0078] Storage method: Distributed database (such as PostgreSQL+PostGIS) is used for storage, supporting geospatial indexing to ensure that the surrounding terrain data can be quickly retrieved when dividing the target area.

[0079] Step Two: Topographic Analysis and Regional Division

[0080] S1: 3D Terrain Data Preprocessing

[0081] Point cloud filtering: A statistical filtering algorithm is used to remove noise from the lidar point cloud. For each point, the average distance between it and its 50 neighboring points is calculated. If the deviation exceeds 1.5 times the standard deviation, it is judged as a noise point (such as birds or temporary obstructions), and more than 95% of the effective points are retained.

[0082] Rasterization: The target area is divided into raster units with a resolution of 1m × 1m. The grid size can be dynamically adjusted according to the complexity of the terrain (e.g., the grid size can be refined to 0.5m × 0.5m in dense forest areas) to ensure the accuracy of feature calculation.

[0083] S2: Slope Calculation :

[0084] For each grid Extract the three-dimensional coordinates (x, y, Z) of all terrain points within it, and fit the plane equation using the least squares method: Z = ax + by + c (where a, b, and c are plane parameters). The slope value is calculated using the plane inclination angle formula.

[0085]

[0086] (Unit: degrees, range 0°~90°, reflecting the steepness of the terrain)

[0087] Vegetation coverage calculation :

[0088] The Normalized Difference Vegetation Index (NDVI) is calculated based on multispectral remote sensing data using the following formula:

[0089]

[0090] in For near-infrared reflectivity, Reflectivity in the red light band (range: -1 to 1)

[0091] Then, the NDVI is converted into vegetation cover (reflecting the density of vegetation) using a piecewise function:

[0092] when hour, (Non-vegetated areas);

[0093] when hour, (Linear mapping);

[0094] when hour, (Dense vegetation area);

[0095] S3: Multi-threshold region classification and boundary determination

[0096] Classification logic: Regions are divided based on raster feature values ​​and preset thresholds.

[0097] Regular road sections and areas: and The terrain is flat and has few obstructions;

[0098] Unconventional road sections or Further subdivided into:

[0099] Dense vegetation sub-region The vegetation is severely obstructing the view;

[0100] Steep slope area The terrain has a large slope.

[0101] Boundaries and coordinate references: Region boundaries are pieced together using raster indexes (e.g., the boundary of a densely vegetated area is...). and (a set of raster cells), and record the origin geodetic coordinates of each raster cell. (The elevation benchmark from the lidar point cloud) provides a coordinate reference for subsequent positioning.

[0102] Step 3: Drone Type Configuration and Deployment

[0103] S1: Drone Type and Sensor Compatibility

[0104] Three types of drones are configured based on regional characteristics to ensure that sensor performance matches terrain requirements:

[0105] The first type of drone (densely vegetated sub-region): equipped with a thermal imaging sensor (640×512 resolution, temperature measurement range -20℃~150℃, frame rate 30fps, penetrating vegetation to detect heat sources) and a high-definition camera (20 million pixels, 30x optical zoom, assisted visual recognition).

[0106] The second type of drone (steep slope area): equipped with lidar (point cloud density 200 points / ㎡, ranging accuracy ±5cm, anti-terrain tilt interference) and inertial navigation module (UTC M8T, positioning accuracy 1cm+1ppm, to compensate for GPS occlusion error).

[0107] The third type of drone (for regular road sections): Equipped with a high-definition camera (12 megapixels, wide-angle lens, suitable for visual search on flat terrain) and a GPS module (positioning accuracy ±1m). All three types of drones are equipped with a unique device ID (such as "T1-001" or "T2-001") and a dedicated communication frequency band (2.4GHz or 5.8GHz, to avoid co-channel interference), and are pre-installed with an encrypted communication protocol (MQTT-SN, supporting low-bandwidth data transmission).

[0108] S2: Deployment Instructions and Confirmations

[0109] Command generation: Based on the area division results, the ground control station generates deployment commands that include the target area boundary (grid index range), rendezvous point coordinates, and cruising altitude.

[0110] Transmission and confirmation: The command is sent to the UAV via a wireless data transmission link (transmission rate ≥1Mbps, delay ≤200ms). After receiving the command, the UAV sends back a confirmation signal containing the device ID and command verification code. After the ground control station verifies the confirmation, the deployment is completed (if no confirmation is received within the timeout period, the command is resent, and the number of retries is ≤3).

[0111] Step 4: Global Path Planning and Data Backhaul

[0112] S1: Path Planning Algorithm and Parameters

[0113] The ground control station generates a global search path based on the following parameters:

[0114] Region boundaries: Ensure the path does not exceed the target region's grid boundaries;

[0115] Drone performance: flight time (total path length ≤ 80% of flight range), maximum climb angle (≤ 30°, suitable for steep slopes);

[0116] Sensor field of view: path overlap rate ≥30% (thermal imaging / LiDAR), ≥20% (visual) to avoid missed detection.

[0117] An improved A* algorithm is used for path planning, incorporating a terrain cost factor. and The higher the value, the greater the cost weight, making the paths denser in complex areas (steep slopes, dense forests) and improving coverage accuracy.

[0118] S2: Data Acquisition and Feedback Mechanism

[0119] Data collected: When the drone cruises along the path, it collects sensor data (thermal imaging video, laser point cloud, visual image) and status data (location, battery level, equipment temperature) in real time.

[0120] Backhaul method: Data is backhauled to the ground control station via a Mesh self-organizing network. A dynamic compression algorithm (50% point cloud compression rate, H.265 image compression format) is used to reduce bandwidth requirements. Data storage adopts a multi-copy strategy (local cache + ground control station backup) to prevent data loss.

[0121] Step 5: Target Detection and Construction of Multi-Angle Detection Array

[0122] S1: Suspected Target Identification

[0123] S2: Probe array assembly strategy

[0124] After receiving a suspected signal, the ground control station quickly locates the UAVs within 500m of the target using spatial indexing and generates a rendezvous command: Rendezvous range: Ensure that at least 3 different types of UAVs (such as 1 thermal imaging, 1 lidar, and 1 vision) approach from different directions (30° depression, 45° side, and 15° elevation) to form a detection array without blind spots; Cooperative control: Ensure that multiple UAVs collect data simultaneously through time synchronization (error ≤ 100ms) to avoid phase differences caused by target movement.

[0125] Step Six: Target Verification and Absolute Coordinate Calculation

[0126] S1: Target Validation and Multi-Source Data Fusion

[0127] The detection array UAV transmits multi-source data back to the ground control station to verify the target's authenticity:

[0128] Thermal imaging data verification: whether the target heat source is continuously moving (excluding static heat sources such as rocks);

[0129] LiDAR data verification: Does the target's three-dimensional contour match human body dimensions (height 1.0~1.8m, width 0.4~0.6m)?

[0130] Visual data verification: whether the clothing color and items carried are consistent with the initial information.

[0131] When the fusion confidence level is ≥90%, the target is confirmed as missing.

[0132] S2: Absolute coordinate calculation

[0133] S2.1: Determine the grid cell where the missing target is located: Based on the location information transmitted back by the UAV from the detection array, match it to the grid cell divided in S2. Get the slope of the grid. Vegetation coverage and the geodetic coordinates of the grid origin;

[0134] S2.2: Extract the initial coordinates for region adaptation:

[0135] If the target is located in a steep slope area , (The second slope threshold in S2), using the lidar point cloud data of the second type of UAV as a reference, extracts the three-dimensional coordinates of the target relative to the lidar. ;

[0136] If the target is located in a densely vegetated sub-region ( , (The second vegetation coverage threshold in S2), fused with the thermal imaging plane coordinates of the first type of UAV. LiDAR elevation data of the second type of UAV , forming initial three-dimensional coordinates ;

[0137] If the target is located in a regular road section area ( and , , (As the first threshold in S2), the initial coordinates are obtained by fusing visual positioning data from a third type of UAV with data from the Chinese BeiDou satellite system. ;

[0138] S2.3: Coordinate correction based on regional features:

[0139] Steep slope areas: Utilize the slope of this grid in S2 ,right Terrain tilt correction is performed using the following formula: Planar coordinates corrected to ;

[0140] Densely vegetated sub-regions: based on thermal imaging plane coordinates Perform filtering. The higher the value, the larger the filter window;

[0141] S2.4: Convert to absolute coordinates: using the formula , , Calculate the absolute coordinates of the target;

[0142] in These are the corrected relative coordinates. For grid The coordinates of the origin.

[0143] S3: Terrain Obstruction Error Elimination

[0144] Eliminating positioning errors caused by terrain occlusion involves the following logic, which is tied to the region segmentation results:

[0145] For steep slope areas Calling the plane equation used in S2 for slope calculation of this area To compensate for the drift error of the inertial navigation module of the second type of UAV, the compensation amount is related to the plane parameters. , The tilt is positively correlated;

[0146] Targeting densely vegetated sub-regions Based on S2 used for calculation NDVI value The reflection intensity in the filtered lidar point cloud is lower than The points retain valid data on penetration through vegetation;

[0147] For the boundary area between conventional and unconventional road sections: using the area boundary defined in S2 as a benchmark, the positioning data of UAVs on both sides are weighted and fused, and the grid data on both sides of the boundary are... , The smaller the difference, the more balanced the weights.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for complex terrain oriented UAV swarm cooperative search and location, characterized in that, Includes the following steps: S1: The ground control station retrieves the three-dimensional terrain data of the target city mountain park from the pre-stored park geographic information database, and at the same time receives the initial information of the missing target sent by the reporting terminal through the mobile communication network, and stores the two types of data together in the local task database; S2: The ground control station calls the terrain analysis module to process the three-dimensional terrain data in S1, calculates the real-time slope value and vegetation coverage of each area, and divides the areas that simultaneously meet the conditions of slope <15° and vegetation coverage <30% into regular road section areas, and divides the areas that meet the conditions of slope ≥15° or vegetation coverage ≥30% into non-regular road section areas. In the non-regular road section areas, dense vegetation sub-areas with coverage ≥50% and steep slope sub-areas with slope ≥25° are further marked, thereby obtaining the area division results. Based on the area division results, the overall area is generated, and the area boundary and grid origin geodetic coordinates are determined. S3: Based on the area division results in S2, the ground control station configures the types of drones in the drone swarm into three categories and sends task deployment instructions to the drone swarm: instructing the first type of drones to go to the dense vegetation sub-area, instructing the second type of drones to go to the steep slope sub-area, and instructing the third type of drones to cover the regular road section area. After receiving the instructions, each drone sends a confirmation signal back to the ground control station through the preset frequency band. S4: Based on the area, area boundary, and drone type allocated to each drone in S3, the ground control station generates a global search path for each drone and sends it out separately. Each drone in the drone swarm performs a search task within its allocated area, collecting environmental data and avoiding obstacles while transmitting the data back to the ground control station. S5: During the search process based on S4, when any UAV detects a suspected target, it extracts the features of the suspected target and matches them with the initial information of the missing target in S1. If the match is successful, a suspected signal is triggered, and the ground control station instructs the UAVs in the mission cluster to gather towards the missing target to form a multi-angle detection array. S6: Based on the multi-angle detection array formed by S5, each UAV in the detection array transmits detection data to the ground control station from different angles. After verifying that it is a lost target, the sensor data from different positions and perspectives are fused to calculate the absolute coordinates of the target and eliminate the positioning error caused by terrain obstruction.

2. The method of claim 1, wherein, The three-dimensional terrain data includes historical slope and vegetation distribution data, and the initial information of the missing target includes the last known location and physical characteristics.

3. The method of claim 1, wherein, The terrain analysis module in S2 specifically includes: S2.1: 3D Terrain Data Preprocessing The 3D terrain data retrieved from S1 is preprocessed as follows: Point cloud filtering: A statistical filtering algorithm is used to remove noise points and retain effective shape points; Grid division: the target area is divided into grid units with a size of 1m x 1m, denoted as ; S2.2: Raster cell feature extraction; S2.3: Multi-threshold region classification.

4. The method of claim 1, wherein, S2.2 specifically includes: S2.21: Calculate the slope of each grid cell :​ The plane equation Z = ax + by + c is fitted using the least squares method. Where: Z is the elevation value, x, y are the plane coordinates, and abc are the plane parameters; Then through Convert plane inclination to slope value; S2.22: Calculate the vegetation cover for each grid cell :​ The Normalized Difference Vegetation Index (NDVI) is calculated using remote sensing spectral information from three-dimensional terrain data. wherein is the reflectivity in the near infrared band, is the reflectivity in the red light band (value range - 1~1); Vegetation coverage Converted from NDVI: When time, ; When Time, ; When time, .

5. The method of claim 4, wherein, The multi-threshold region classification specifically includes: The conventional road section region is a set of grid cells satisfying and wherein is a first slope threshold (value 15°), is a first vegetation coverage threshold (value 30%). The irregular road section area is a set of grid cells that satisfy all of or ​ The vegetation dense sub-area is a set of grid cells in the non-conventional road section area satisfying , wherein is a second vegetation coverage threshold (value 50%). The steep slope sub-region is a grid cell set satisfying in the unconventional road section region, wherein is a second slope threshold (25°). wherein : a slope threshold value to distinguish between regular and irregular road segments; : a vegetation coverage threshold value that distinguishes between regular and irregular road segments; : slope threshold of steep slope sub-area in irregular road section; : Vegetation coverage threshold for dense vegetation sub-area in irregular road segment.

6. The method of claim 1, wherein, The specific configuration types include: the first type of drone is equipped with a thermal imaging sensor and a high-definition camera; the second type of drone is equipped with a lidar and an inertial navigation module; the third type of drone is equipped with only a high-definition camera; and each of the three types of drones is assigned a unique device identifier and communication frequency band.

7. The method of claim 1, wherein, The process of calculating the absolute coordinates of the target in S6 is implemented in association with the region division algorithm in S2 through grid cells, and specifically includes the following steps: A1: Determine the grid unit where the missing target is located: according to the position information returned by the detection array unmanned aerial vehicle, match to the grid unit divided in S2 , call the slope of the grid , vegetation coverage and the geodetic coordinates of the grid origin; A2: Extract the initial coordinates for region adaptation: If the target is located in the steep slope sub-region , is the second slope threshold in S2), based on the laser radar point cloud data of the second type of unmanned aerial vehicle, the three-dimensional coordinates of the target relative to the laser radar are extracted ; If the target is located in a sub-region with dense vegetation , , the thermal imaging plane coordinates of the first type of unmanned aerial vehicle are fused with the laser radar elevation data of the second type of unmanned aerial vehicle , to form initial three-dimensional coordinates ; If the target is located in the conventional road segment area and , 、 , the initial coordinates are obtained by fusing the visual positioning of the third type of unmanned aerial vehicle and the data of China's Beidou satellite (S2 is the first threshold value) ; A3: Coordinate correction based on regional features: Steep slope sub-region: using the slope of the grid in S2 , the terrain tilt correction is performed on , the correction formula is , the plane coordinate correction is ; Vegetation dense sub-region: according to to the thermal imaging plane coordinates filtering, the higher, the larger the filter window; A4: Convert to absolute coordinates: using the formula , , Calculate the absolute coordinates of the target; in These are the corrected relative coordinates. For grid The coordinates of the origin.

8. A method for cooperative search and localization of UAV swarms in complex terrain according to claim 1, characterized in that, The elimination of positioning errors caused by terrain occlusion in S6 specifically involves the following logic, which is tied to the region division results of S2: For steep slope areas Calling the plane equation used in S2 for slope calculation of this area To compensate for the drift error of the inertial navigation module of the second type of UAV, the compensation amount is related to the plane parameters. , The tilt is positively correlated; Targeting densely vegetated sub-regions Based on S2 used for calculation NDVI value The reflection intensity in the filtered lidar point cloud is lower than The points retain valid data on penetration through vegetation; For the boundary area between conventional and unconventional road sections: using the area boundary defined in S2 as a benchmark, the positioning data of UAVs on both sides are weighted and fused, and the grid data on both sides of the boundary are... , The smaller the difference, the more balanced the weights.

Images