A UAV bidirectional collaborative recovery method and recovery system based on multi-source positioning

Abstract

The present invention discloses a bidirectional collaborative recovery method and recovery system for unmanned aerial vehicles (UAVs) based on multi-source positioning, and relates to the field of UAV positioning and control technology. Multi-source perception systems such as laser radars and visual sensors are integrated into the recovery system, thereby constructing a recovery system with external perception capabilities and bidirectional interaction capabilities. It includes: the UAV sends a recovery request to the ground end; the ground end sends the ground end position and recovery channel parameters to the UAV; plans the expected recovery track, and flies to the recovery channel based on the planned track; the ground end locates the UAV in real time through the multi-source positioning system, and sends a track correction instruction to the UAV based on the positioning result, so that the UAV flies according to the track correction instruction. At the same time, the recovery system adjusts its attitude based on the real-time positioning result. The UAV is not only able to plan its track autonomously, but the recovery system can also guide the UAV to achieve precise docking, thereby improving the recovery success rate and attitude matching accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) positioning and control, and in particular to a UAV bidirectional collaborative recovery system and recovery method based on multi-source positioning. Background Art

[0002] In recent years, small and medium-sized drones (UAVs) have been widely used in military reconnaissance, logistics and transportation, emergency rescue, environmental monitoring, and other fields due to their low cost, high maneuverability, flexible deployment, and strong mission adaptability. To improve their operational efficiency and reusability, efficient, safe, and controllable UAV recovery technology after mission completion has gradually become a research hotspot.

[0003] Currently, mainstream methods for recovering fixed-wing drones include: taxiing landing, parachute drop, net capture, horizontal line interception, and skyhook recovery. These methods typically employ a strategy where the drone is the active control unit and the recovery system is the passive control unit. Specifically, the recovery system passively receives the drone, while the drone acts as the sole active controller and assumes all navigation, attitude adjustment, and docking tasks. While this one-way control model is simple, it also presents several challenges.

[0004] First, the drone's own positioning system mostly relies on positioning devices such as GPS and RTK. Due to positioning errors, drift accumulation and environmental interference, it is difficult to maintain high-precision trajectory control in actual missions; and traditional recovery methods have high requirements for the recovery path and attitude accuracy, which often leads to docking failures or structural damage. Secondly, the existing recovery system lacks perception and active adjustment capabilities, and cannot quickly adapt to the current state of the drone and environmental changes, resulting in poor dynamic responsiveness and insufficient robustness of the system. In addition, the closed-loop feedback mechanism of the one-way control system is not sound. If the drone deviates from the predetermined recovery channel, it often needs to retake off and restart the process, resulting in energy waste and mission delays. In severe cases, there is even a risk of recovery failure or collision. Summary of the Invention

[0005] In response to the above problems, the present invention proposes a bidirectional collaborative recovery method and recovery system for UAVs based on multi-source positioning, integrating multi-source perception systems such as lidar and visual sensors into the recovery system, thereby constructing a recovery system with external perception capabilities and bidirectional interaction capabilities. Through the multi-source fusion positioning method, the recovery system can obtain high-precision real-time posture of the UAV; at the same time, the recovery system can send track correction instructions to the UAV flight control system based on the positioning error, thereby realizing a bidirectional collaborative control mechanism in which the external system assists the flight control in trajectory correction, effectively alleviating the accuracy pressure of the UAV navigation system and improving the recovery success rate. In general, the present invention can solve the problems of one-way control, inaccurate positioning, weak attitude matching capability and poor environmental adaptability existing in the prior art through multi-sensor data fusion, multi-source perception positioning and a UAV collaborative recovery solution that supports bidirectional interaction.

[0006] The technical solution of the present invention is: comprising the following steps:

[0007] S1: The drone sends a recovery request and the current drone's flight parameters to the ground terminal; the flight parameters include the drone's longitude, latitude, altitude, speed, heading, remaining endurance, and flight attitude;

[0008] S2: After receiving the recovery request from the drone, the ground terminal sends the ground terminal position and recovery channel parameters to the drone;

[0009] S3: The UAV plans the desired recovery trajectory based on the ground terminal's position and recovery channel parameters as well as the UAV's current position and heading, and flies toward the recovery channel based on the planned trajectory;

[0010] S4: When the UAV enters the recovery channel, it sends a docking request to the ground system. The ground system uses a multi-source positioning system to locate the UAV in real time and sends a track correction command to the UAV based on the positioning results. The UAV then flies according to the track correction command. At the same time, the recovery system adjusts its attitude based on the real-time positioning results to achieve docking with the UAV.

[0011] S5: When the drone reaches the recovery distance threshold, the ground terminal determines the recovery conditions. If the recovery conditions are not met, the drone performs a go-around and restarts the recovery process. If the recovery conditions are met, the drone continues to maintain its heading until the recovery is complete.

[0012] The multi-source positioning system includes a plurality of sensing sources arranged around the recovery system, the sensing sources including visual cameras and lidar sensors;

[0013] After the drone enters the recovery channel, the ground terminal uses a multi-source positioning system to locate the drone in real time. The specific steps include:

[0014] Time synchronization of point cloud data and image data acquired by each perception source;

[0015] Perform coordinate transformation on the point cloud data to obtain the projected point cloud in the image coordinate system. The coordinates in the laser radar coordinate system are , its coordinates in the visual camera coordinate system are , and there are ,in and are the rotation matrix and translation matrix from the lidar coordinate system to the visual camera coordinate system, respectively. The coordinates of the projected point cloud in the image coordinate system can be expressed as:

[0016] , where Represents the horizontal direction in the image, represents the vertical direction in the image, and K represents the transformation matrix;

[0017] Based on the deep learning model, the target drone in the image data is identified and the pixel range of the image occupied by the target drone is screened out, which is recorded as , where 、 is the horizontal and vertical coordinate position of the target in the image, 、 The horizontal and vertical coordinate areas occupied by the target are respectively, and the projected Projected point cloud coordinates Perform screening, the screening conditions are as follows:

[0018] ;

[0019] The processed data from different perception sources are fused and calculated, and recorded For LiDAR The generated target point cloud data is represented by a four-dimensional coordinate vector, as follows:

[0020] , which is the laser radar Coordinate system to lidar The transformation relationship between coordinate systems, where and They are rotation and translation relationships respectively. Each lidar point cloud is in the lidar The coordinate vector in the coordinate system can be expressed as: , is the number of point clouds produced by the Xth radar;

[0021] Get LiDAR The stitched point cloud data in the coordinate system , for each point after completing the point cloud stitching , define its neighborhood:

[0022] ,in is the set of spliced ​​point clouds, is the neighborhood radius;

[0023] like , is the quantity threshold, then the point As the core point, all density-reachable points are expanded into a cluster to obtain several clusters , let the cluster with the largest number of points be the target drone cluster, that is , then the average position of the cluster is the target position: .

[0024] The recovery system includes an arresting device 100, a robot system 200, a ground rail 300, a multi-source positioning system 400 and a drone 500; the arresting device 100 is installed on the power execution end of the robot system 200 to achieve arresting and decelerating the drone; the robot system 200 has a spatial multi-degree-of-freedom attitude adjustment capability to support the arresting device to actively adjust according to the drone's posture state; the ground rail 300 is installed under the robot system to support the robot system to move along the track to match the dynamic trajectory of the drone; the multi-source positioning system 400 is arranged on both sides of the robot system 200 and the ground rail 300 to achieve high-precision real-time positioning of the drone within a limited range.

[0025] The robot system 200 is a six-axis robot, which is divided into axis, axis, axis, axis, axis, The above-mentioned rotating shaft enables the robot flange position to have the ability to adjust the posture in six degrees of freedom in space. By arranging the arresting device at the execution end of the six-axis robot, the active capture of the drone can be achieved; the default The axis angle is zero;

[0026] During the drone recovery process, the robot rotates Axis makes recovery plane Parallel to the track plane By rotating The axis enables the arresting device to adapt to the roll angle of the UAV; the translation of the robot on the ground track can make the recovery plane coincide with the track plane;

[0027] On the recycling plane In the process, the robot system adjusts axis, axis, The axis makes the arresting gear pitch Arrive at the predetermined recycling point coordinates ,set up axis, axis, The axis rotation angles are 、 、 ;

[0028] Establish a coordinate system at the robot base , axis, axis, Establish coordinate systems at the axes 、 、 , based on the adjacent link coordinate system transformation matrix , calculate the coordinate system Relative to the coordinate system The homogeneous transformation matrix :

[0029] ;

[0030] A5 axis coordinate system Relative to the base coordinate system The transformation matrix is ​​expressed as:

[0031] ;

[0032] in and Represents the coordinate system Position on the recycling plane:

[0033] ;

[0034] ;

[0035] Thus we get axis, axis, Axis angle 、 、 They are:

[0036] ;

[0037] ;

[0038] ;

[0039] Finally, get axis, axis, Axis angle 、 、 , and enter Axis angle and Axis angle, to control the robot system.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] First, by introducing a two-way collaborative control mechanism, the drone can not only autonomously plan its trajectory based on the location information of the ground recovery system, but the recovery system also guides the drone to achieve precise docking in real time through sensory feedback, thereby improving the recovery success rate and attitude matching accuracy.

[0042] 2. The multi-source positioning system uses lidar and visual sensors for fusion positioning, which greatly improves positioning accuracy and anti-interference capabilities compared to relying solely on GPS.

[0043] 3. The two-way collaborative recovery method based on multi-source positioning can realize positioning error correction and dynamic adjustment of recovery attitude, so that the recovery system has the function of correction feedback during the UAV recovery process, which can effectively solve problems such as UAV trajectory deviation and attitude disturbance, and improve the adaptability of the recovery system in complex scenarios; at the same time, this method takes into account the UAV's ability to take off again after a recovery failure. By setting the recovery fault-tolerant judgment logic, the UAV can take off again and retry the recovery after docking failure, thereby ensuring system stability and mission safety.

[0044] In summary, the present invention proposes a bidirectional collaborative recovery method for UAVs and recovery systems that integrates perception fusion, track correction, and attitude adjustment. It is suitable for various UAV autonomous recovery scenarios such as vehicle-mounted, ship-mounted, and fixed platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the UAV recovery system based on multi-source positioning.

[0046] Figure 2a Schematic diagram of the arresting device.

[0047] Figure 2b This is a schematic diagram of the blocking block and electromagnetic lock structure.

[0048] Figure 2c This is a schematic diagram of the front limit and guide device structure.

[0049] Figure 2d This is a schematic diagram of the arrangement of buffers and arresting cables.

[0050] Figure 2eSchematic diagram of the drone interception process.

[0051] Figure 3 A diagram illustrating the definition of each axis of the robot system.

[0052] Figure 4 Schematic diagram of the coordinate system in the robot control method.

[0053] Figure 5 This is a schematic diagram of the recovery position in the robot control method.

[0054] Figure 6 This is a flow chart of the UAV bidirectional collaborative recovery method based on multi-source positioning.

[0055] Figure 7 Schematic diagram of the multi-source positioning method. DETAILED DESCRIPTION

[0056] In order to clearly illustrate the technical features of the present invention, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings.

[0057] Figure 1 The figure shows the status of a UAV bidirectional collaborative recovery system based on multi-source positioning proposed in the present invention during a UAV recovery operation. The recovery system includes an arresting device 100, a robot system 200, a ground rail 300, a multi-source positioning system 400 and a UAV 500.

[0058] The arresting device 100 is installed on a robot actuator with six-degree-of-freedom posture adjustment capability and can be actively adjusted according to the position and posture information of the drone 500. The ground rail 300 is installed below the robot system 200 to support the robot system 200 to move along the rail, thereby matching the dynamic trajectory of the drone 500.

[0059] The UAV 500 and the recovery system are both equipped with independent positioning and communication devices, and the two can establish a communication connection to achieve two-way positioning and data exchange; in addition to its own positioning system, the recovery system is also equipped with a multi-source positioning system 400 for more accurately positioning the UAV 500. The multi-source positioning system 400 is arranged on both sides of the robot 200 system, and can perform more accurate real-time positioning of the UAV 500 within a limited range.

[0060] The above generally describes the recovery system proposed in the present invention, elaborating on the overall layout and functions of the arresting device 100, robotic system 200, ground rail 300, multi-source positioning system 400, and drone 500. The structure and principles of each of the above components will be described in detail below.

[0061] See also Figure 2a 、 2b, 2c, 2d, 2e, the arresting device 100 includes an arresting frame 101, an arresting block 102, a slide rail 103, a front limiter 104, a rear limiter 105, a buffer 108, and an arresting cable 108c;

[0062] The arresting frame 101 is fixedly mounted on the output end of the multi-axis robot, and a pair of slide rails 103 are arranged in parallel, both of which are fixedly mounted on the arresting frame 101. The arresting block 102 can be slidably mounted on each slide rail 103, and the front limit 104 and the rear limit 105 are fixedly mounted at the head and tail ends of the slide rail 103. A number of pulleys are installed at the front and bottom of the slide rail 103. The cavity 108a of the buffer 108 is fixedly mounted below the slide rail 103, and a pulley is also installed on the pull rod 108b of the buffer 108. One end of the arresting cable 108c is fixedly connected to the arresting frame 101, and the other end is fixedly connected to the arresting block 102 after passing through each pulley in turn. One side of the arresting block 102 is provided with a slot adapted to the wing of the fixed-wing UAV 500.

[0063] About the specific structure of the blocking block:

[0064] The blocking block 102 includes a horseshoe plate 102a, an impact sponge 102b, a pulley 102c and a limiting pulley 102d. The limiting pulley 102d is rotatably connected to the bottom of the pulley 102c and accommodated in the slide rail 103. The horseshoe plate 102a is fixedly installed on the pulley 102c. The impact sponge 102b is fixedly installed in the horseshoe plate 102a, and the impact sponge 102b is provided with the card slot.

[0065] Furthermore, the blocking block 102 further includes a brake lug 102e;

[0066] As a supporting device, a number of electromagnetic locks 107 are fixedly installed on the slide rail 103 at equal intervals. The brake ear 102e is located in the area directly above the electromagnetic lock 107 and is used to lock the position of the arresting block 102 when the drone is arrested.

[0067] Regarding the specific structure of the arresting frame:

[0068] The main body of the arresting frame 101 adopts a frame structure, with an arc-shaped guide device 106 connected to the front side, and two horizontal beams extending backward, on which the slide rails 103 are installed;

[0069] The guiding device 106 includes a roller seat 106a, a roller 106b, and an I-beam 106c. The I-beam 106c is an arc structure and is fixedly connected to the arresting frame 101. The roller seats 106a are arranged in series on the I-beam 106c. Each roller seat 106a has the same structure and is equipped with a flexible roller 106b to guide the drone smoothly into the arresting frame during recovery.

[0070] There are four guiding devices 106, which are arranged symmetrically in pairs. The I-beams 106c of the same pair of guiding devices 106 are between the wing receiving spaces, which gradually narrow and connect to the area where the arresting block 102 is located. The guiding devices 106 are used to guide the drone 500 during the recovery process, so that the drone can be smoothly docked with the arresting frame 101 and the wing can be introduced into the arresting block 102.

[0071] The arresting frame 101 has a frame-like structure with a hollowed-out center for passing drones. A trapezoidal steel plate is installed underneath to enhance the frame's strength. Ten mounting holes are punched on one side of the plate to connect it to the robot system's power output. Two rearward-extending crossbeams are arranged on the left and right sides of the arresting frame 101, on which slide rails 103 are mounted. Each slide rail 103 is slidably mounted with an arresting block 102, forming a deceleration channel during drone recovery.

[0072] Connected to the front side of the arresting frame 101 are four curved guides 106, arranged symmetrically in pairs. A wing receiving space is located between each pair of guides 106, which gradually narrows and connects to the area where the arresting block 102 is located. The guides 106 guide the drone 500 during recovery, allowing it to dock smoothly with the arresting frame 101 and guide its wings into the arresting block 102.

[0073] There are two buffers 108, which are symmetrically installed below the arresting frame 101; a number of pulleys are installed at the front and bottom of the arresting frame 101, and a pulley is also installed on the pull rod of the buffer 108; the two ends of the arresting cable are respectively fixed to the middle steel plate of the arresting frame 101 and the front end of the arresting block 102, and the middle part of the arresting cable is supported by multiple fixed pulleys below the arresting frame body, and bypasses the movable pulley at the end of the buffer pull rod, thereby increasing the deceleration stroke during the recovery of the drone, thereby reducing the overload of the drone recovery.

[0074] Regarding the installation location of the robot system:

[0075] The bottom of the robot system 200 is mounted on a ground rail in a movable manner, and a walking mechanism is provided at the bottom of the robot system 200 .

[0076] See also Figure 3The robot system has six independent rotating shafts, starting from the root. axis, axis, axis, axis, axis, Axis, the above axis enables the robot flange position to have six degrees of freedom posture adjustment capability, among which the default The axis rotation angle is zero. In addition, the arresting device is placed at the execution end of the robot to achieve coordinated recovery of the drone. The recovery window is a rectangular opening at the front end of the arresting device for achieving drone interception.

[0077] For the robot system in this embodiment, the present invention provides a control method for adjusting the posture of the drone according to its posture state:

[0078] See also Figure 4 During the drone recovery process, the robot rotates Axis makes recovery plane Parallel to the track plane By rotating The axis enables the arresting device to adapt to the roll angle of the UAV; the recovery plane can be made to coincide with the track plane by the translation of the robot on the ground track.

[0079] See also Figure 5 , on the recycling plane In the process, the robot system adjusts axis, axis, The axis makes the arresting gear pitch Arrive at the predetermined recycling point coordinates ,set up axis, axis, The axis rotation angles are 、 、 .

[0080] Establish a coordinate system at the robot base , axis, axis, Establish coordinate systems at the axes 、 、 , based on the adjacent link coordinate system transformation matrix , calculate the coordinate system Relative to the coordinate system The homogeneous transformation matrix :

[0081] ;

[0082] A5 axis coordinate system Relative to the base coordinate system The transformation matrix is ​​expressed as:

[0083] ;

[0084] in and Represents the coordinate system Position on the recycling plane:

[0085] ;

[0086] ;

[0087] Thus we get axis, axis, Axis angle 、 、 They are:

[0088] ;

[0089] ;

[0090] ;

[0091] Finally, get axis, axis, Axis angle 、 、 , and enter Axis angle, Axis angle, to control the robot system.

[0092] Figure 6 This embodiment provides a method for bidirectional collaborative recovery of drones based on multi-source positioning. This recovery method is applicable to both the recovery system and the drone, and the two achieve bidirectional collaborative recovery operations through information exchange, feedback correction, and dynamic adjustment. When the drone needs to be recovered by the recovery system, the recovery method provided by the present invention is triggered, which specifically includes the following steps:

[0093] S1: The drone sends a recovery request and the current drone flight parameters to the ground terminal.

[0094] In some specific embodiments, this step can be performed as follows: after the UAV sends a recovery request to the ground end, if the ground end has the recovery conditions, it will return a confirmation recovery instruction to the UAV. The UAV will then continuously send its own flight parameters to the ground end and update them in real time to ensure that the ground end can grasp the latest flight data in real time.

[0095] It should be noted that the recovery request also includes the flight parameters at the current moment; the flight parameters include data such as the longitude, latitude, altitude, speed, heading, remaining endurance, and flight attitude of the drone.

[0096] S2: After receiving the recovery request from the drone, the ground terminal sends the ground terminal position and recovery channel parameters to the drone.

[0097] After receiving a recovery request from a drone, the ground system determines recovery based on the drone's flight parameters and the ground system's location and status. If the recovery conditions are met, the system sends a confirmation of the recovery command to the drone, along with the ground system's specific location and recovery channel parameters. The recovery channel is a rectangular airspace with a certain width, height, and length, and is only accessible to the target drone.

[0098] S3: The UAV plans the desired recovery trajectory based on the ground terminal position and recovery channel parameters as well as the UAV's current position and heading, and performs tracking control based on the planned trajectory.

[0099] The drone plans a desired recovery trajectory based on the ground terminal's position, recovery path parameters, and the drone's current position and heading. This includes: if the drone's current altitude is too far from the recovery path and the descent path exceeds the drone's maximum maneuverability, it will execute a slow descent in a circling pattern. Once the drone's altitude drops to an acceptable level, recovery trajectory planning resumes, adjusting the drone's flight direction to align with the recovery path.

[0100] Tracking control based on the planned trajectory also includes: longitudinal altitude and speed control, lateral roll and lateral deviation control, and longitudinal and lateral stability must be maintained before the drone enters the recovery channel. If a gust of wind or other environmental influences cause the drone to experience significant attitude and heading fluctuations, a go-around should be performed immediately. After recovery, the route should be planned and the recovery process should be repeated at an appropriate time.

[0101] S4: When the UAV enters the recovery channel, it sends a docking request to the ground end; the ground end uses a multi-source positioning system to locate the UAV in real time, and sends a track correction instruction to the UAV based on the positioning results, so that the UAV flies according to the track correction instruction. At the same time, the recovery system adjusts its attitude based on the real-time positioning results to meet the docking requirements with the UAV.

[0102] After the drone enters the recovery channel, the ground terminal uses a multi-source positioning system to locate the drone in real time, including the following solutions:

[0103] See also Figure 7 The multi-source positioning system in this embodiment has four perception sources, each of which is equipped with a visual camera and a lidar. By fusing the image data and point cloud data generated by each positioning source, the target drone point cloud under a single positioning source can be accurately screened. The target point clouds generated by all positioning sources are then spliced ​​and clustered to finally determine the positioning data of the target drone.

[0104] Because visual sensors and lidar sensors acquire data at different frequencies, image and point cloud data must be timestamped to ensure effective time synchronization of data near the same time point. This allows for coordinate transformation to determine the position of each point cloud within the image, known as the projected point cloud. Furthermore, deep learning-based image object detection algorithms can effectively identify and filter target drones within image data, thereby determining the pixel range of the target drone within the image. By fusing the projected point cloud with the target drone's pixel range, the target point cloud data generated by the target drone can be accurately obtained.

[0105] S5: When the drone reaches the recovery distance threshold, the ground terminal determines the recovery conditions. If the recovery conditions are not met, the drone performs a go-around and restarts the recovery process. If the recovery conditions are met, the drone continues to maintain its heading until the recovery is complete.

[0106] The recovery distance threshold represents the necessary distance that the drone must maintain from the recovery system when performing a go-around. If the distance between the drone and the recovery system is lower than this threshold, dangerous situations such as scratches or collisions may occur during the go-around.

[0107] The recovery condition judgment may include the following aspects: whether the UAV is in a stable attitude state, whether the communication and positioning signals are good, whether the docking error between the recovery system and the UAV is within an acceptable range, and whether there are only obstacles in the recovery channel.

[0108] About the method of calculating the position of the projected point cloud in the image:

[0109] Assume the visual camera coordinate system is , the laser radar coordinate system is , the image coordinate system is .

[0110] The laser radar detects a point in the environment The coordinates in the laser radar coordinate system are , its coordinates in the visual camera coordinate system are , and there are ,in and They are the rotation matrix and translation matrix from the laser radar coordinate system to the visual camera coordinate system, and The specific value of depends on the relative installation position of the vision camera and the lidar.

[0111] Points The position in the image is , where Represents the horizontal direction in the image, Represents the vertical direction in the image, is the intrinsic parameter matrix of the visual camera and is determined by its own structure.

[0112] That is, the coordinates of the projected point cloud in the image coordinate system are:

[0113] ;

[0114] Calculation method for fusion of projected point cloud and pixel range of target drone:

[0115] After target detection based on the image information of the visual camera, the pixel range of the target drone in the image is obtained, which is recorded as , where 、 is the horizontal and vertical coordinate position of the target in the image, 、 They are the horizontal and vertical areas occupied by the target respectively.

[0116] The projected Projected point cloud coordinates Perform screening, the screening conditions are as follows:

[0117] ;

[0118] That is, only the point cloud within the target pixel range is retained. On the one hand, the number of point clouds is reduced, thereby improving the calculation speed. On the other hand, the screening of the target point cloud helps to improve the positioning accuracy.

[0119] About the calculation method of point cloud fusion and positioning:

[0120] For ease of description, it is arranged at the location source The laser radar is recorded as laser radar .

[0121] remember For LiDAR The generated target point cloud data is represented by a four-dimensional coordinate vector, where For LiDAR No. The coordinate vector of the point cloud in the radar coordinate system.

[0122] set up:

[0123] ;

[0124] For LiDAR Coordinate system to lidar The transformation relationship between coordinate systems, where and They are rotation and translation relationships respectively. Each lidar point cloud is in the lidar The coordinate vector in the coordinate system can be expressed as:

[0125] ;

[0126] Thus, we get the laser radar The stitched point cloud data in the coordinate system .

[0127] For each point after point cloud stitching is completed , define its neighborhood:

[0128] ;

[0129] like , then point As the core point, all density-reachable points are expanded into a cluster to obtain several clusters , let the cluster with the largest number of points be the target drone cluster, that is , then the average position of the cluster is the target position:

[0130] .

[0131] There are many specific implementation ways of the present invention. The above is only the preferred implementation method of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principles of the present invention. These improvements should also be considered as the scope of protection of the present invention.

Claims

1. A method for bidirectional collaborative recovery of UAVs based on multi-source positioning, characterized in that: The following steps are involved: S1: The drone sends a recovery request and the current drone's flight parameters to the ground terminal; the flight parameters include the drone's longitude, latitude, altitude, speed, heading, remaining endurance, and flight attitude; S2: After receiving the recovery request from the drone, the ground terminal sends the ground terminal position and recovery channel parameters to the drone; S3: The UAV plans the desired recovery trajectory based on the ground terminal's position and recovery channel parameters as well as the UAV's current position and heading, and flies toward the recovery channel based on the planned trajectory; S4: When the UAV enters the recovery channel, it sends a docking request to the ground system. The ground system uses a multi-source positioning system to locate the UAV in real time and sends a track correction command to the UAV based on the positioning results. The UAV then flies according to the track correction command. At the same time, the recovery system adjusts its attitude based on the real-time positioning results to achieve docking with the UAV. S5: When the drone reaches the recovery distance threshold, the ground terminal determines the recovery conditions. If the recovery conditions are not met, the drone performs a go-around and restarts the recovery process. If the recovery conditions are met, the drone continues to maintain its heading until recovery is complete. The recovery system comprises an arresting device (100), a robot system (200), a ground rail (300), a multi-source positioning system (400) and a drone (500); the arresting device (100) is installed at the power execution end of the robot system (200) to achieve arresting and decelerating the drone; the robot system (200) has a spatial multi-degree-of-freedom attitude adjustment capability to support the arresting device to actively adjust according to the posture state of the drone; the ground rail (300) is installed below the robot system to support the robot system to move along the rail, thereby matching the dynamic trajectory of the drone; the multi-source positioning system (400) is arranged on both sides of the robot system (200) and the ground rail (300) to achieve high-precision real-time positioning of the drone within a limited range; The robot system (200) is a six-axis robot, which has A1 axis, A2 axis, A3 axis, A4 axis, A5 axis, and A6 axis from its root. The above-mentioned rotating axes enable the robot flange position to have a six-degree-of-freedom attitude adjustment capability in space. By arranging the arresting device at the execution end of the six-axis robot, the active capture of the drone is achieved. The default rotation angle of the A4 axis is zero. During the UAV recovery process, the robot rotates the A1 axis to make the recovery plane Y1O1Z1 parallel to the track plane Y2O2Z2; the arresting device adapts to the UAV's roll angle by rotating the A6 axis; the robot's translation on the ground track can make the recovery plane coincide with the track plane; In the recovery plane Y1O1Z1, the robot system adjusts the A2 axis, A3 axis, and A5 axis to make the arresting device reach the predetermined recovery point coordinate position (L capture ,H capture ), let the rotation angles of A2 axis, A3 axis and A5 axis be θ1, θ2 and θ3 respectively; The coordinate system {0} is established at the robot base, and the coordinate systems {1}, {2}, and {3} are established at the A2 axis, A3 axis, and A5 axis respectively. Based on the transformation matrix of the adjacent link coordinate system Calculate the homogeneous transformation matrix of coordinate system {3} relative to coordinate system {0} The transformation matrix of the coordinate system {3} at the A5 axis relative to the base coordinate system {0} is expressed as: where y A5 With z A5 Indicates the position of coordinate system {3} on the recycling plane: y A5 =L capture -(L3+L 4_1 )cosθ+L 4_2 sinθ; z A5 =H capture -(L3+L 4_1 )sinθ-L 4_2 cosθ; The rotation angles θ1, θ2, and θ3 of the A2, A3, and A5 axes are obtained as follows: θ3 = θ - θ1 - θ2; Finally, the rotation angles θ1, θ2, and θ3 of the A2, A3, and A5 axes are obtained, and the rotation angles of the A1 and A6 axes are input to control the robot system.

2. The method for bidirectional collaborative recovery of UAVs based on multi-source positioning according to claim 1, characterized in that: The multi-source positioning system includes a plurality of sensing sources arranged around the recovery system, the sensing sources including visual cameras and lidar sensors; After the drone enters the recovery channel, the ground terminal uses a multi-source positioning system to locate the drone in real time. The specific steps include: Time synchronization of point cloud data and image data acquired by each perception source; Perform coordinate transformation on the point cloud data to obtain the projected point cloud in the image coordinate system. The coordinate of a point P detected by the laser radar in the environment in the laser radar coordinate system is P c , its coordinate in the visual camera coordinate system is P c , and there are in and are the rotation matrix and translation matrix from the lidar coordinate system to the visual camera coordinate system, respectively. The coordinates of the projected point cloud in the image coordinate system can be expressed as: Where u represents the horizontal direction in the image, v represents the vertical direction in the image, and K represents the transformation matrix; Based on the deep learning model, the target drone in the image data is identified and the pixel range of the image occupied by the target drone is screened out, which is recorded as R = (u, v, Δu, Δv). In the formula, u and v are the horizontal and vertical coordinate positions of the target in the image, Δu and Δv are the horizontal and vertical coordinate areas occupied by the target respectively. The n projected point cloud coordinates P fi =[u i v i 1] T (i=1,2,3,…,n) for screening. The screening conditions are as follows: The processed data of different perception sources are fused and calculated, denoted by P i X=[x i X y i X z i X 1] T The target point cloud data generated by the laser radar X∈{A,B,C,D} is represented by a four-dimensional coordinate vector, and the following formula is set: That is the conversion relationship between the laser radar X coordinate system and the laser radar A coordinate system, where and are rotation and translation relationships respectively. The coordinate vectors of each lidar point cloud in the lidar A coordinate system can be expressed as: Nx is the number of point clouds generated by the xth radar; Get the spliced ​​point cloud data in the laser radar A coordinate system For each point P after point cloud stitching is completed i , define its neighborhood: N ε (P i )={P j ∈P total |‖P j -P i ‖ 2 ≤ε}, where P total is the set of spliced ​​point clouds, ε is the neighborhood radius; If |N ε (P i )|≥MinPts, MinPts is the quantity threshold, then point P i As the core point, all density-reachable points are expanded into a cluster, and several clusters c1, c2, ...c N c i ∈P total , let the cluster with the largest number of points be the target drone cluster, i.e. c UAV =arg max|c i |, then the average position of the cluster is the target position:

Images