A method and system for precise point delivery of a UAV material delivery cabinet

Abstract

The application discloses a kind of unmanned aerial vehicle material delivery cabinet precision fixed-point delivery method and system, it is related to unmanned aerial vehicle positioning technical field;The application generates obstacle avoidance track by obtaining initial position of unmanned aerial vehicle, in combination with digital elevation model;When it is less than first distance threshold from target material delivery cabinet, use memory alloy anti-deformation target and CNN neural network to solve pose error;When it is less than second distance threshold from target material delivery cabinet, fusion laser slope detection, RTK residual and wind disturbance offset, generate four-dimensional calibration matrix by Kalman filtering;When it is less than third distance threshold from target material delivery cabinet, drive three-axis holder to execute dynamic compensation, synchronous verification positioning error, cabinet inclination, wind disturbance trigger release;After landing, realize closed-loop verification by pressure sensing;The application breaks through delivery bottleneck under complex environment, significantly improves delivery accuracy and reliability, with terrain self-adaptive leveling, dual-mode all-weather identification and multi-level security protection capability.

Description

Technical Field

[0001] This invention belongs to the field of drone positioning technology, and specifically relates to a method and system for precise and targeted delivery of supplies by drone delivery cabinets. Background Technology

[0002] Currently, drone delivery technology is being used more and more widely in disaster relief, emergency medical delivery, and military supply.

[0003] Existing technologies suffer from significant single-GPS positioning errors, with accuracy dropping sharply when obstructed by buildings or subjected to electromagnetic interference. Traditional airdrops rely on the hovering accuracy of drones, and the landing point of supplies deviates severely in strong winds. Furthermore, the delivery process lacks real-time feedback control on the cabinet's position and orientation. These technologies suffer from insufficient positioning accuracy, limitations in static delivery, and a lack of dynamic calibration capabilities. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the problems in related technologies, this invention provides a method for precise and targeted delivery of supplies using drone delivery cabinets, thereby overcoming the aforementioned technical issues in existing related technologies.

[0006] (II) Technical Solution

[0007] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0008] S1. Generate a three-dimensional waypoint sequence based on the initial position of the UAV and the position of the target material delivery container;

[0009] S2. The UAV flies along a three-dimensional waypoint sequence; when the distance between the UAV and the target material delivery cabinet is less than the first distance threshold, the target image features on the top of the target material delivery cabinet are extracted through a CNN neural network to obtain the target features.

[0010] S3. Process the target features using the perspective n-point algorithm to obtain the actual spatial pose of the target; correct the UAV trajectory based on the actual spatial pose of the target to obtain the trajectory correction command packet.

[0011] S4. The UAV flies according to the trajectory correction instruction package. When the distance between the UAV and the target material delivery cabinet is less than the second distance threshold, the positioning deviation in the plane direction of the target material delivery cabinet, the position offset component caused by the slope, and the position offset caused by the wind are corrected to obtain a four-dimensional calibration matrix.

[0012] S5. When the drone descends to a height less than the third distance threshold from the target material delivery cabinet, adjust the spatial pose of the target material delivery cabinet based on the four-dimensional calibration matrix. After adjustment, determine whether the three material release thresholds are met simultaneously. If they are met, trigger the material release command. Otherwise, return to S4 and adjust the four-dimensional calibration matrix until the material release command is triggered.

[0013] S6. After triggering the resource release command, release the resources and determine whether the resource release was successful.

[0014] This invention achieves precise UAV delivery in complex environments through multi-level positioning correction; it generates a three-dimensional waypoint sequence based on RTK positioning and digital elevation model to avoid no-fly zones; it extracts anti-deformation target features through adaptive visual recognition, and calculates spatial pose using a perspective n-point algorithm to dynamically correct the trajectory; it integrates planar positioning deviation, slope offset, and wind disturbance drift to output a four-dimensional calibration matrix; it drives a three-axis gimbal to level the cabinet, unlocking in milliseconds after meeting three thresholds, and the pressure sensor verifies successful landing; it overcomes the meter-level error problem of traditional delivery in scenarios such as strong winds, rugged terrain, and low visibility, achieving both accuracy and reliability while improving emergency response efficiency.

[0015] Preferably, step S1 includes the following steps:

[0016] S11. After the UAV takes off, the RTK differential positioning module is activated to receive BeiDou / GPS satellite signals and ground-based augmentation station correction data to obtain the initial UAV positioning data.

[0017] The drone's location is obtained by calculating the real-time latitude, longitude, and altitude in the drone's positioning data using carrier phase differential technology.

[0018] By comparing satellite visibility and position accuracy factors, if both requirements are met, the UAV's position is taken as its initial position; otherwise, the system switches to GLONASS / Galileo multi-system fusion positioning to obtain the UAV's initial position.

[0019] S12. Collect target delivery point data to obtain the location of the target material delivery cabinet; the cloud dispatch system sends the target material delivery cabinet location and safe flight altitude instructions to the drone via the 5G network;

[0020] The flight control system, based on the UAV's position coordinates and the target material delivery container's location, combined with a digital elevation model, automatically avoids preset no-fly zones, generates a Bézier curve trajectory with optimal energy consumption, marks meteorological monitoring points, and obtains UAV trajectory data.

[0021] Based on UAV flight path data, the geodetic coordinates are converted to the local ENU coordinate system to obtain a three-dimensional waypoint sequence;

[0022] This invention ensures the reliability of the initial position by combining RTK differential positioning with multi-system redundancy. After the target coordinates are sent from the cloud, the flight control system integrates a digital elevation model to generate obstacle avoidance trajectory in real time, optimizes energy consumption based on Bézier curves, and finally converts the geodetic coordinates into three-dimensional waypoints in the ENU coordinate system. This solves the problem of avoiding no-fly zones in complex terrain, improves positioning robustness and trajectory planning efficiency, and reduces flight energy consumption.

[0023] Preferably, step S2 includes the following steps:

[0024] S21. The UAV flies along a three-dimensional waypoint sequence; set the first distance threshold and visibility threshold;

[0025] Collect real-time environmental data to obtain real-time visibility; when the distance between the drone and the target material delivery cabinet is less than the first distance threshold, select the acquisition mode based on the real-time visibility to identify and acquire the dedicated target data on the top of the target material delivery cabinet to obtain the target image.

[0026] S22. Use background suppression algorithm to eliminate vegetation and shadow interference in target image, enhance the contrast of target edge, and obtain effective target image area;

[0027] S23. By using a CNN neural network to identify the QR code positioning pattern in the effective target image region, analyze the data module arrangement pattern, and track the displacement of the deformation vertex, the target features are obtained.

[0028] This invention acquires target images by selecting the acquisition mode based on real-time visibility, eliminates interference through background suppression, and uses a CNN network to analyze the features of deformed targets; thus achieving highly robust recognition in complex backgrounds and laying the foundation for accurate pose calculation.

[0029] In step S21, when the distance between the drone and the target material delivery cabinet is less than a distance threshold, the process of selecting a data acquisition mode based on real-time visibility to identify and acquire dedicated target data on the top of the delivery cabinet, and obtaining target image data, includes the following steps:

[0030] When the distance between the drone and the target material delivery cabinet is less than the first distance threshold and the real-time visibility is greater than or equal to the visibility threshold, the camera on the underside of the drone is automatically turned on, the focus is adjusted to aim at the delivery area, the special target on the top of the delivery cabinet is identified, and the target image is obtained.

[0031] When the distance between the drone and the target material delivery cabinet is less than the first distance threshold and the real-time visibility is less than the visibility threshold, switch to infrared laser to illuminate the target, and use the night vision camera to identify reflective markings to obtain the target image;

[0032] This invention intelligently switches between daytime high-definition video recording and nighttime infrared laser recognition modes based on environmental visibility to accurately capture images of deformation-resistant targets; it enables reliable target recognition under day and night conditions and in adverse weather, providing stable input for subsequent pose calculation.

[0033] Preferably, step S3 includes the following steps:

[0034] S31. Based on the perspective n-point algorithm, establish the target world coordinate system, match the target features with the target features in the preset three-dimensional model, and solve the rotation matrix and translation vector to obtain the actual spatial pose of the target.

[0035] The four-dimensional error vector of the target is obtained by comparing the horizontal error vector and the attitude angle deviation between the actual spatial pose and the theoretical spatial pose of the target.

[0036] S32. Collect real-time coordinates and speed data of the UAV to obtain the real-time RTK positioning data of the first UAV; perform a timestamp matching operation between the real-time RTK positioning data of the first UAV and the four-dimensional error vector of the target to obtain the timestamp matched data.

[0037] Transform the coordinate system to the ENU reference system, unify the timestamp matching data to the ENU reference system, and obtain data in a unified coordinate system;

[0038] An extended Kalman filter is used to fuse the RTK positioning data in the unified coordinate system with the target's four-dimensional error vector to obtain the first fused positioning error;

[0039] Based on the first fusion positioning error, the eastward speed, northward speed, and altitude maintenance command of the UAV are corrected to obtain the trajectory correction command package;

[0040] This invention uses a perspective n-point algorithm to calculate the target's spatial pose and generate a four-dimensional error vector. This vector is then timestamped with real-time RTK data and converted to the ENU coordinate system. After fusion with an extended Kalman filter, a trajectory correction command is output. This achieves dynamic compensation for visual-satellite positioning errors, improving trajectory correction accuracy to the centimeter level and effectively overcoming positioning drift caused by strong wind disturbances.

[0041] Preferably, step S4 includes the following steps:

[0042] S41. The UAV flies according to the trajectory correction instruction package; a second distance threshold is set; when the UAV descends to a height less than the second distance threshold from the target material delivery cabinet, the positioning deviation in the collection surface direction, the position offset component caused by the slope, and the position offset component caused by the wind are calculated.

[0043] S42. Based on the optimal estimation algorithm of Kalman filtering, a state vector containing position deviation and velocity is established; the positioning deviation in the plane direction, the position offset component caused by slope, and the position offset caused by wind are integrated to construct the observation vector.

[0044] Based on the state vector and observation vector, the optimal estimate is output through a prediction-update loop to obtain the fused compensation level offset and slope angle;

[0045] S43. Based on the principle of spatial coordinate transformation, the fused compensation horizontal offset and slope angle are converted into directly executable command parameters to obtain a four-dimensional calibration matrix.

[0046] This invention integrates positioning deviation, slope offset, and wind disturbance drift, outputs the optimal compensation value through Kalman filtering, and converts it into a four-dimensional calibration matrix instruction; it overcomes the problem of combined interference from terrain undulations and strong winds, achieving centimeter-level hovering accuracy and laying the foundation for final precise deployment.

[0047] Preferably, step S41 includes the following steps:

[0048] S411. Compare the height differences of the four corners of the target material delivery cabinet to obtain the ground slope angle data; convert the ground slope angle data into a horizontal position offset through trigonometric function spatial projection to obtain the position offset component caused by the slope.

[0049] S412. Collect the coordinate data of the current UAV to obtain the RTK coordinate data of the second real-time UAV; calculate the deviation between the RTK coordinate data of the second real-time UAV and the position of the target material delivery cabinet, and eliminate the multipath error caused by building reflection to obtain the positioning deviation in the plane direction.

[0050] S413. Based on the aerodynamics and motion prediction model, the positional shift caused by the wind is obtained by estimating the drift caused by the wind in combination with the fall time using the three-dimensional wind vector obtained by the wind speed sensor.

[0051] This invention analyzes the ground slope by measuring the height difference at the four corners of the cabinet and converts it into a horizontal offset. Combined with RTK positioning deviation, it suppresses multipath error and wind disturbance drift prediction model, and accurately quantifies the three major constraint offset components. It achieves centimeter-level error modeling in complex terrain and strong wind environments, providing accurate input for dynamic compensation.

[0052] Preferably, step S5 includes the following steps:

[0053] S51. When the drone descends to a height less than the third distance threshold from the target material delivery cabinet, collect the cabinet-to-ground height value.

[0054] S52. Set the mechanical transmission ratio relationship; based on spatial coordinate transformation, mechanical kinematics principle and cabinet-to-ground height value, convert the four-dimensional calibration matrix into specific motion commands for the servo motor through the mechanical transmission ratio relationship, and obtain three sets of motor control commands; the three sets of motor control commands include roll motor stroke, pitch motor stroke and rotation platform angle.

[0055] The spatial pose of the delivery cabinet is corrected by driving a three-axis gimbal with three sets of motor control commands to perform physical compensation actions. During the correction process, the cabinet's posture is monitored in real time by a high-precision gyroscope, and the motor output is dynamically adjusted by a PID control algorithm to eliminate execution errors and obtain a delivery cabinet that is stable in the target posture.

[0056] S53. Set the second fusion positioning error threshold, cabinet tilt angle threshold, and wind disturbance threshold.

[0057] The current positioning error is collected and fused by the RTK differential positioning module and the lidar to obtain the second fused positioning error; the cabinet tilt angle is directly measured by the gyroscope to obtain the real-time tilt angle of the cabinet; and the real-time wind disturbance is obtained by high-frequency sampling by the accelerometer.

[0058] If the following conditions are met simultaneously: second fusion positioning error < second fusion positioning error threshold, cabinet real-time tilt angle < cabinet tilt angle threshold, and real-time wind disturbance < real-time wind disturbance, then the triple material release threshold is met, and the material release command is triggered; otherwise, return to S4 and adjust the compensation parameters until the material release command is triggered.

[0059] This invention converts a four-dimensional calibration matrix into three-axis gimbal servo commands, driving the roll, pitch, and yaw mechanisms to dynamically level the cabinet's attitude. It also uses a triple threshold-based judgment system—LiDAR-RTK fusion positioning, gyroscope tilt monitoring, and accelerometer wind disturbance detection—to trigger release upon reaching the threshold. This achieves millimeter-level execution accuracy and multi-dimensional safety verification, ensuring successful deployment in complex environments.

[0060] Preferably, step S6 includes the following steps:

[0061] S61. After triggering the resource release command, the resources will be released.

[0062] S62. Set pressure threshold; the distributed pressure sensor array at the bottom of the delivery cabinet monitors the impact force in real time the moment the material touches the ground. When the sensor detects a continuous pressure ≥ the pressure threshold, it is determined to be a valid landing and the material delivery success notification is sent.

[0063] If the pressure sensor does not detect a valid impact signal, the deployment is deemed abnormal, and the abnormal response plan is triggered.

[0064] This invention achieves rapid release of materials through millisecond-level unlocking of electromagnetic locks and spring-assisted push, and utilizes a distributed pressure sensor array to monitor impact force in real time, triggering an emergency response plan in case of anomalies; ensuring that materials land accurately with zero damage and improving the success rate of delivery.

[0065] A system for precise delivery of supplies from a drone delivery cabinet, and a method for precise delivery of supplies from a drone delivery cabinet, comprising an initial positioning and trajectory planning module, a target recognition and feature extraction module, a pose calculation and trajectory correction module, a compensation matrix generation module, a pose fine-tuning and release determination module, and a status feedback module.

[0066] The initial positioning and trajectory planning module is used for accurate positioning of the UAV after takeoff and safe and efficient flight path generation; using the target cabinet position, digital elevation model and safe altitude command issued by the cloud, it automatically avoids no-fly zones, generates a Bézier curve trajectory based on optimal energy consumption, and converts the geodetic coordinates into a three-dimensional waypoint sequence in the local ENU coordinate system to guide the UAV flight;

[0067] The target recognition and feature extraction module is used to adaptively select a daytime high-definition camera or a nighttime infrared laser + night vision mode based on the real-time environmental visibility when the distance of the drone is less than the first threshold, to identify and collect the image of the specially made deformable reflective QR code target on the top of the material delivery cabinet; after eliminating interference through the background suppression algorithm, the module uses a CNN neural network to accurately extract key visual features such as the QR code positioning pattern, data module arrangement, and the deformation vertices of the shape memory alloy frame.

[0068] The pose calculation and trajectory correction module uses the extracted target features and the perspective n-point algorithm to calculate the actual spatial pose of the target relative to the UAV, and calculates a four-dimensional error vector including horizontal position error and attitude angle deviation. After the error vector is matched with the UAV's real-time RTK positioning data through timestamp matching and ENU coordinate system, it is fused through an extended Kalman filter to generate correction command packets for the UAV's eastward and northward speed and altitude, and dynamically adjust the flight trajectory.

[0069] The compensation matrix generation module is used to collect and fuse planar positioning deviation, position offset component caused by ground slope, and estimated drift caused by wind when the UAV descends to a distance less than the second threshold. It integrates these error sources using the Kalman filter optimal estimation algorithm, outputs the fused horizontal offset and slope angle information, and finally constructs a spatial four-dimensional calibration matrix that can be directly used for the actuator.

[0070] The pose fine-tuning and release determination module is used to activate the lidar array to accurately measure the cabinet-to-ground height when the UAV descends to a distance less than the second threshold. Using a four-dimensional calibration matrix, through spatial coordinate transformation and mechanical transmission ratio, it generates servo motor control commands to drive the three-axis gimbal, and executes physical compensation actions to accurately correct the spatial pose of the delivery cabinet. It also integrates RTK and lidar positioning errors in real time, monitors cabinet tilt angle and wind disturbance, and strictly determines whether the preset triple release threshold is met simultaneously to decide whether to trigger release.

[0071] The status feedback module is used to execute the material release action when the release conditions are met; at the moment the material touches the ground, the impact force is detected by the distributed pressure sensor array at the bottom of the cabinet to determine effective landing and report successful delivery; if no effective impact is detected, an anomaly is determined.

[0072] (III) Beneficial Effects

[0073] The present invention has the following beneficial effects:

[0074] This invention enables precise delivery in complex environments. Through a three-level positioning technology loop of "satellite positioning - visual target recognition - lidar near-end ranging", it dynamically corrects wind disturbance, slope and multipath errors, reducing the meter-level delivery error of traditional UAVs to the centimeter level, significantly improving the accuracy of material landing in disaster-prone mountainous areas and strong wind environments.

[0075] This invention uses a dual-mode recognition system (high-definition daytime video recording / infrared laser nighttime imaging) to ensure stable feature extraction even in rainy, foggy, dark, or target-deformed scenarios. By combining background suppression algorithms with CNN neural networks, it effectively overcomes recognition failures caused by vegetation obscuring and shadow interference.

[0076] This invention features terrain-adaptive intelligent leveling. It utilizes a real-time slope-position offset conversion model and dynamically analyzes the ground tilt angle through the height difference between the four corners of the cabinet. This drives a three-axis gimbal to perform roll, pitch, and yaw compensation, enabling the delivery cabinet to automatically level itself on non-planar terrains such as steep slopes and gravel, eliminating the risk of materials slipping on the ground.

[0077] This invention features a multi-source proactive risk defense mechanism; by constructing a triple release threshold judgment system based on positioning error, cabinet tilt angle, and wind disturbance, release is triggered only when all three conditions are met simultaneously; in case of an anomaly, an emergency system is activated, and impact force feedback verification is provided to prevent material fall or jamming accidents.

[0078] This invention generates Bezier curve tracks based on digital elevation models and updates waypoints in real time with wind speed to reduce energy consumption during headwind flight. During the deployment phase, it employs electromagnetic locks for millisecond-level unlocking and spring-assisted propulsion to shorten the loiter time and improve emergency mission response efficiency.

[0079] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0080] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0081] Figure 1 This is a flowchart illustrating a method for precise and targeted delivery of supplies using a drone delivery cabinet according to the present invention.

[0082] Figure 2 This is a schematic diagram of a system for precise and targeted delivery of supplies by drones, as described in this invention. Detailed Implementation

[0083] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0084] In the description of this invention, it should be understood that the terms "opening", "upper", "lower", "top", "middle", "inner", etc., which indicate orientation or positional relationship, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the invention.

[0085] Example 1:

[0086] Please see Figure 1 This invention discloses a method for precise and targeted delivery of supplies using a drone delivery cabinet, comprising the following steps:

[0087] S1. Generate a three-dimensional waypoint sequence based on the initial position of the UAV and the position of the target material delivery container;

[0088] S1 includes the following steps:

[0089] S11. After the UAV takes off, the RTK differential positioning module is activated to receive BeiDou / GPS satellite signals and ground-based augmentation station correction data to obtain the initial UAV positioning data.

[0090] The drone's location is obtained by calculating the real-time latitude, longitude, and altitude in the drone's positioning data using carrier phase differential technology.

[0091] By comparing satellite visibility (≥8 satellites) and position accuracy factor (PDOP < 1.5), if both satellite visibility and position accuracy factor meet the requirements, the UAV's position is taken as the UAV's initial position; otherwise, the system switches to GLONASS / Galileo multi-system fusion positioning to obtain the UAV's initial position.

[0092] S12. Collect target delivery point data to obtain the location of the target material delivery cabinet; the cloud dispatch system sends the target material delivery cabinet location and safe flight altitude instructions to the drone via 4G / 5G network;

[0093] The flight control system, based on the UAV's position coordinates and the target material delivery container's location, combined with a digital elevation model (DEM), automatically avoids preset no-fly zones, generates a Bézier curve trajectory with optimal energy consumption, marks meteorological monitoring points (wind speed needs to be updated in real time), and obtains UAV trajectory data.

[0094] Based on UAV flight path data, the geodetic coordinates (Lon / Lat) are converted to the local ENU coordinate system to obtain a three-dimensional waypoint sequence;

[0095] S2. The UAV flies along a three-dimensional waypoint sequence; when the distance between the UAV and the target material delivery cabinet is less than the first distance threshold, the target image features on the top of the target material delivery cabinet are extracted through a CNN neural network to obtain the target features.

[0096] S2 includes the following steps:

[0097] S21. The UAV flies along a three-dimensional waypoint sequence; set a first distance threshold (e.g., 50m) and a visibility threshold;

[0098] Real-time environmental data is collected to obtain real-time visibility. When the distance between the drone and the target material delivery cabinet is less than the first distance threshold, the acquisition mode is selected based on the real-time visibility to identify and collect the data of the special target on the top of the target material delivery cabinet, and the target image is obtained. The target is a deformable QR code (60×60cm) made of shape memory alloy frame, a black module inlaid with fluorescent material (daytime reflectivity >80%), and the frame hinge point allows ±15° deformation (wind pressure resistant design).

[0099] In step S21, when the distance between the drone and the target material delivery cabinet is less than a distance threshold, the process of selecting a data acquisition mode based on real-time visibility to identify and acquire dedicated target data on the top of the delivery cabinet, and obtaining target image data, includes the following steps:

[0100] When the distance between the drone and the target material delivery cabinet is less than the first distance threshold and the real-time visibility is greater than or equal to the visibility threshold, the 2-megapixel high-definition camera (frame rate ≥ 30fps) on the underside of the drone is automatically turned on, the focus is adjusted to aim at the delivery area, the special target on the top of the delivery cabinet is identified, and the target image is obtained.

[0101] When the distance between the drone and the target material delivery cabinet is less than the first distance threshold and the real-time visibility is less than the visibility threshold, switch to infrared laser to illuminate the target, and use the night vision camera to identify reflective markings to obtain the target image;

[0102] S22. Use background suppression algorithm to eliminate vegetation and shadow interference in target image, enhance the contrast of target edge, and obtain effective target image area;

[0103] S23. By recognizing the QR code positioning pattern (three corner marks) in the effective target image area through the CNN neural network, analyzing the data module arrangement pattern, and tracking the displacement of the deformation vertex (memory alloy frame feature points), the target features are obtained.

[0104] S3. Process the target features using the perspective n-point algorithm to obtain the actual spatial pose of the target; correct the UAV trajectory based on the actual spatial pose of the target to obtain the trajectory correction command packet.

[0105] S3 includes the following steps:

[0106] S31. Based on the perspective n-point algorithm, establish the target world coordinate system (with the center of the QR code as the origin), match the target features with the target features in the preset three-dimensional model, and solve the rotation matrix R and translation vector T to obtain the actual spatial pose of the target.

[0107] The four-dimensional error vector of the target is obtained by comparing the horizontal error vector and the attitude angle deviation between the actual spatial pose and the theoretical spatial pose of the target; the horizontal error vector includes the x-coordinate error and the y-coordinate error, and the attitude angle deviation includes the pitch angle and the roll angle (unit: degrees);

[0108] S32. Collect real-time coordinates and speed data of the UAV to obtain the real-time RTK positioning data of the first UAV; perform timestamp matching (error < 10ms) operation between the real-time RTK positioning data of the first UAV and the four-dimensional error vector of the target to obtain the timestamp matched data.

[0109] Transform the coordinate system to the ENU (East-North-Sky) reference system, unify the timestamp matching data to the ENU reference system, and obtain data in a unified coordinate system;

[0110] An extended Kalman filter is used to fuse the RTK positioning data in the unified coordinate system with the target's four-dimensional error vector to obtain the first fused positioning error;

[0111] Based on the first fusion positioning error, the eastward speed, northward speed, and altitude maintenance command of the UAV are corrected to obtain the trajectory correction command package;

[0112] S4. The UAV flies according to the trajectory correction instruction package. When the distance between the UAV and the target material delivery cabinet is less than the second distance threshold, the positioning deviation in the plane direction of the target material delivery cabinet, the position offset component caused by the slope, and the position offset caused by the wind are corrected to obtain a four-dimensional calibration matrix.

[0113] S4 includes the following steps:

[0114] S41. The UAV flies according to the trajectory correction instruction package; a second distance threshold is set (e.g., 10m); when the UAV descends to a height less than the second distance threshold from the target material delivery cabinet, the positioning deviation in the collection surface direction, the position offset component caused by the slope, and the position offset component caused by the wind are collected.

[0115] S41 includes the following steps:

[0116] S411. Compare the height differences of the four corners of the target material delivery cabinet to obtain the ground slope angle data; convert the ground slope angle data into a horizontal position offset through trigonometric function spatial projection to obtain the position offset component caused by the slope.

[0117] S412. Collect the coordinate data of the current UAV to obtain the RTK coordinate data of the second real-time UAV; calculate the deviation between the RTK coordinate data of the second real-time UAV and the position of the target material delivery cabinet, and eliminate the multipath error caused by building reflection to obtain the positioning deviation in the plane direction.

[0118] S413. Based on the aerodynamics and motion prediction model, the positional shift caused by the wind is obtained by estimating the drift caused by the wind in combination with the fall time using the three-dimensional wind vector obtained by the wind speed sensor.

[0119] S42. Based on the optimal estimation algorithm of Kalman filtering, a state vector containing position deviation and velocity is established; the positioning deviation in the plane direction, the position offset component caused by slope, and the position offset caused by wind are integrated to construct the observation vector.

[0120] Based on the state vector and observation vector, the optimal estimate is output through a prediction-update loop to obtain the fused compensation level offset and slope angle;

[0121] S43. Based on the principle of spatial coordinate transformation, the fused compensation horizontal offset and slope angle are converted into directly executable command parameters to obtain a four-dimensional calibration matrix (including horizontal offset + gimbal compensation angle).

[0122] S5. When the drone descends to a height less than the third distance threshold from the target material delivery cabinet, adjust the spatial pose of the target material delivery cabinet based on the four-dimensional calibration matrix. After adjustment, determine whether the three material release thresholds are met simultaneously. If they are met, trigger the material release command. Otherwise, return to S4 and adjust the four-dimensional calibration matrix until the material release command is triggered.

[0123] S5 includes the following steps:

[0124] S51. When the drone descends to a height ≤ the third distance threshold (e.g., 3M) from the target material delivery cabinet, the lidar array (wavelength 905nm, Class 1 safety level) installed at the four corners of the delivery cabinet is activated. The four sets of laser beams are emitted vertically downwards, and the distance is calculated using the time-of-flight (ToF) principle to obtain the height value of the collection cabinet from the ground.

[0125] S52. Set the mechanical transmission ratio relationship; based on spatial coordinate transformation, mechanical kinematics principle and cabinet-to-ground height value, convert the four-dimensional calibration matrix (including horizontal offset and compensation angle) into specific motion commands for the servo motor through the mechanical transmission ratio relationship, and obtain three sets of motor control commands; the three sets of motor control commands include roll motor stroke, pitch motor stroke and rotation platform angle;

[0126] The spatial pose of the delivery cabinet is corrected by a three-axis gimbal driven by three sets of motor control commands to perform physical compensation actions (based on precision servo control and mechanism dynamics, roll compensation: differential push and pull of left and right servo motors to adjust the left and right tilt of the cabinet; pitch compensation: synchronous extension and retraction of front and rear servo motors to control the front and rear tilt of the cabinet; yaw compensation: the bottom rotating platform drives the cabinet to rotate around the vertical axis). During the correction process, the cabinet attitude is monitored in real time by a high-precision gyroscope, and the motor output is dynamically adjusted by a PID control algorithm to eliminate execution errors and obtain a delivery cabinet that is stable in the target attitude.

[0127] S53. Set the second fusion positioning error threshold, cabinet tilt angle threshold, and wind disturbance threshold.

[0128] The current positioning error is collected and fused by the RTK differential positioning module and the lidar to obtain the second fused positioning error; the cabinet tilt angle is directly measured by the gyroscope to obtain the real-time tilt angle of the cabinet; and the real-time wind disturbance is obtained by high-frequency sampling by the accelerometer.

[0129] If the following conditions are met simultaneously: second fusion positioning error < second fusion positioning error threshold, cabinet real-time tilt angle < cabinet tilt angle threshold, and real-time wind disturbance < real-time wind disturbance, then the triple material release threshold is met, and the material release command is triggered; otherwise, return to S4 and adjust the compensation parameters until the material release command is triggered.

[0130] S6. After triggering the resource release command, release the resources and determine whether the resource release was successful.

[0131] S6 includes the following steps:

[0132] S61. After triggering the material release command, the material is released; (the electromagnetic lock current is cut off to achieve millisecond-level unlocking, the spring mechanism pushes open the hatch (opening degree > 30 cm), and the booster spring gives the material an initial velocity of 1.2 m / s)

[0133] The moment the supplies hit the ground, the distributed pressure sensor array at the bottom of the delivery cabinet monitors the impact force in real time. When any sensor detects a continuous pressure of ≥5kg (lasting for more than 10 milliseconds), it is determined to be a valid landing and the delivery of supplies is confirmed as successful.

[0134] If the pressure sensor fails to detect a valid impact signal for 2 consecutive seconds, the deployment is deemed abnormal, triggering the emergency response plan. Considering the risk of material jamming or crashing, the parachute module is automatically activated, the gunpowder propulsion ejects the deceleration parachute (opening time < 0.3 seconds), and the parachute ropes pull the cabinet down stably (descent speed ≤ 3m / s).

[0135] Example 2:

[0136] Please see Figure 2 A system for precise delivery of supplies from a drone delivery cabinet, used to realize the aforementioned method for precise delivery of supplies from a drone delivery cabinet, includes an initial positioning and trajectory planning module, a target recognition and feature extraction module, a pose calculation and trajectory correction module, a compensation matrix generation module, a pose fine-tuning and release determination module, and a status feedback module.

[0137] The initial positioning and trajectory planning module is used for accurate positioning of the UAV after takeoff and safe and efficient flight path generation; using the target cabinet position, digital elevation model and safe altitude command issued by the cloud, it automatically avoids no-fly zones, generates a Bézier curve trajectory based on optimal energy consumption, and converts the geodetic coordinates into a three-dimensional waypoint sequence in the local ENU coordinate system to guide the UAV flight;

[0138] The target recognition and feature extraction module is used to adaptively select a daytime high-definition camera or a nighttime infrared laser + night vision mode based on the real-time environmental visibility when the distance of the drone is less than the first threshold, to identify and collect the image of the specially made deformable reflective QR code target on the top of the material delivery cabinet; after eliminating interference through the background suppression algorithm, the module uses a CNN neural network to accurately extract key visual features such as the QR code positioning pattern, data module arrangement, and the deformation vertices of the shape memory alloy frame.

[0139] The pose calculation and trajectory correction module uses the extracted target features and the perspective n-point algorithm to calculate the actual spatial pose of the target relative to the UAV, and calculates a four-dimensional error vector including horizontal position error and attitude angle deviation. After the error vector is matched with the UAV's real-time RTK positioning data through timestamp matching and ENU coordinate system, it is fused through an extended Kalman filter to generate correction command packets for the UAV's eastward and northward speed and altitude, and dynamically adjust the flight trajectory.

[0140] The compensation matrix generation module is used to collect and fuse planar positioning deviation, position offset component caused by ground slope, and estimated drift caused by wind when the UAV descends to a distance less than the second threshold. It integrates these error sources using the Kalman filter optimal estimation algorithm, outputs the fused horizontal offset and slope angle information, and finally constructs a spatial four-dimensional calibration matrix that can be directly used for the actuator.

[0141] The pose fine-tuning and release determination module is used to activate the lidar array to accurately measure the cabinet-to-ground height when the UAV descends to a distance less than the second threshold. Using a four-dimensional calibration matrix, through spatial coordinate transformation and mechanical transmission ratio, it generates servo motor control commands to drive the three-axis gimbal, and executes physical compensation actions to accurately correct the spatial pose of the delivery cabinet. It also integrates RTK and lidar positioning errors in real time, monitors cabinet tilt angle and wind disturbance, and strictly determines whether the preset triple release threshold is met simultaneously to decide whether to trigger release.

[0142] The status feedback module is used to execute the material release action when the release conditions are met; at the moment the material touches the ground, the impact force is detected by the distributed pressure sensor array at the bottom of the cabinet to determine effective landing and report successful delivery; if no effective impact is detected, an anomaly is determined.

[0143] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0144] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for precise, targeted delivery of supplies using a drone delivery system, characterized in that, Includes the following steps: S1. Generate a three-dimensional waypoint sequence based on the initial position of the UAV and the position of the target material delivery container; S2. The UAV flies along a three-dimensional waypoint sequence; when the distance between the UAV and the target material delivery cabinet is less than the first distance threshold, the target image features on the top of the target material delivery cabinet are extracted through a CNN neural network to obtain the target features. S3. Process the target features using the perspective n-point algorithm to obtain the actual spatial pose of the target; The UAV trajectory is corrected based on the actual spatial pose of the target to obtain a trajectory correction command packet. S4. The UAV flies according to the trajectory correction instruction package. When the distance between the UAV and the target material delivery cabinet is less than the second distance threshold, the positioning deviation in the plane direction of the target material delivery cabinet, the position offset component caused by the slope, and the position offset caused by the wind are corrected to obtain a four-dimensional calibration matrix. The four-dimensional calibration matrix includes the horizontal offset and the gimbal compensation angle. S5. When the drone descends to a height less than the third distance threshold from the target material delivery cabinet, the spatial pose of the target material delivery cabinet is adjusted based on the four-dimensional calibration matrix. After adjustment, it is determined whether the three material release thresholds are met simultaneously. If they are met, the material release command is triggered. Otherwise, return to S4 and adjust the four-dimensional calibration matrix until the material release command is triggered; the triple material release threshold is specifically to simultaneously satisfy the second fusion positioning error < the second fusion positioning error threshold, the cabinet real-time tilt angle < the cabinet tilt angle threshold, and the real-time wind disturbance amount < the wind disturbance amount threshold; wherein, the second fusion positioning error is obtained by jointly collecting the current positioning error by the RTK differential positioning module and the lidar and fusing them; S6. After triggering the resource release command, release the resources and determine whether the resource release was successful.

2. The method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 1, characterized in that, S1 includes the following steps: S11. After the UAV takes off, the RTK differential positioning module is activated to receive BeiDou / GPS satellite signals and ground-based augmentation station correction data to obtain the initial UAV positioning data. The drone's position is obtained by calculating the real-time latitude, longitude, and altitude in the initial drone positioning data using carrier phase differential technology. By comparing satellite visibility and position accuracy factors, if both requirements are met, the UAV's position is taken as its initial position; otherwise, the system switches to GLONASS / Galileo multi-system fusion positioning to obtain the UAV's initial position. S12. Collect target delivery point data to obtain the location of the target material delivery cabinet; the cloud dispatch system sends the target material delivery cabinet location and safe flight altitude instructions to the drone via the 5G network; The flight control system generates drone trajectory data based on the drone's position coordinates and the location of the target material delivery container, combined with a digital elevation model; The geodetic coordinates were converted to the local ENU coordinate system, and a three-dimensional waypoint sequence was obtained based on the UAV flight path data.

3. The method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 1, characterized in that, S2 includes the following steps: S21. The UAV flies along a three-dimensional waypoint sequence; set the first distance threshold and visibility threshold; Real-time environmental data is collected to obtain real-time visibility. When the distance between the drone and the target material delivery cabinet is less than the first distance threshold, the acquisition mode is selected based on the real-time visibility to identify and collect the data of the special target on the top of the target material delivery cabinet, and the target image is obtained. The special target is a deformable QR code made of a shape memory alloy frame, with black modules inlaid with fluorescent materials and frame hinge points that allow ±15° deformation. S22. Use background suppression algorithm to eliminate vegetation and shadow interference in target image, enhance the contrast of target edge, and obtain effective target image area; S23. By using a CNN neural network to identify the QR code positioning pattern in the effective target image region, analyze the data module arrangement pattern, and track the displacement of the deformation vertices, the target features are obtained.

4. The method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 3, characterized in that, In step S21, when the distance between the drone and the target material delivery cabinet is less than a distance threshold, the process of selecting a data acquisition mode based on real-time visibility to identify and acquire dedicated target data on the top of the delivery cabinet, and obtaining target image data, includes the following steps: When the distance between the drone and the target material delivery cabinet is less than the first distance threshold and the real-time visibility is greater than or equal to the visibility threshold, the camera on the underside of the drone is automatically turned on, the focus is adjusted to aim at the delivery area, the special target on the top of the delivery cabinet is identified, and the target image is obtained. When the distance between the drone and the target material delivery cabinet is less than the first distance threshold and the real-time visibility is less than the visibility threshold, the target is illuminated by infrared laser, and the reflective mark is identified by the night vision camera to obtain the target image.

5. The method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 1, characterized in that, S3 includes the following steps: S31. Process the target features using the perspective n-point algorithm to obtain the actual spatial pose of the target; The four-dimensional error vector of the target is obtained by comparing the horizontal error vector and the attitude angle deviation between the actual spatial pose and the theoretical spatial pose of the target. S32. Collect real-time coordinates and speed data of the UAV to obtain the real-time RTK positioning data of the first UAV; perform a timestamp matching operation between the real-time RTK positioning data of the first UAV and the four-dimensional error vector of the target to obtain the timestamp matched data. Transform the coordinate system to the ENU reference system, unify the timestamp matching data to the ENU reference system, and obtain data in a unified coordinate system; An extended Kalman filter is used to fuse the RTK positioning data in the unified coordinate system with the target's four-dimensional error vector to obtain the first fused positioning error; The UAV is corrected based on the first fusion positioning error to obtain a trajectory correction command packet.

6. The method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 1, characterized in that, S4 includes the following steps: S41. The UAV flies according to the trajectory correction instruction package; a second distance threshold is set; when the UAV descends to a height less than the second distance threshold from the target material delivery cabinet, the positioning deviation in the collection surface direction, the position offset component caused by the slope, and the position offset component caused by the wind are calculated. S42. The optimal estimation algorithm based on Kalman filtering integrates the positioning deviation in the plane direction, the position offset component caused by the slope, and the position offset caused by the wind to obtain the fused compensated horizontal offset and slope angle. S43. The fused compensation horizontal offset and slope angle are converted into directly executable command parameters to obtain a four-dimensional calibration matrix.

7. A method for precise point-to-point delivery of supplies using a drone delivery cabinet according to claim 6, characterized in that, S41 includes the following steps: S411. Compare the height differences of the four corners of the target material delivery cabinet to obtain the ground slope angle data; convert the ground slope angle data into a horizontal position offset to obtain the position offset component caused by the slope. S412. Collect the coordinate data of the current UAV to obtain the RTK coordinate data of the second real-time UAV; calculate the deviation between the RTK coordinate data of the second real-time UAV and the position of the target material delivery cabinet, and eliminate the multipath error caused by building reflection to obtain the positioning deviation in the plane direction. S413. Based on the aerodynamics and motion prediction model, the positional shift caused by the wind is obtained by combining the three-dimensional wind vector obtained from the wind speed sensor with the fall time to estimate the drift caused by the wind.

8. The method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 1, characterized in that, S5 includes the following steps: S51. When the drone descends to a height less than the third distance threshold from the target material delivery cabinet, collect the cabinet-to-ground height value. S52. Set the mechanical transmission ratio relationship; based on the cabinet-to-ground height value, convert the mathematical parameters of the four-dimensional calibration matrix into specific motion commands for the servo motor through the mechanical transmission ratio relationship, and obtain three sets of motor control commands; The spatial pose of the delivery cabinet is corrected by driving a three-axis gimbal with three sets of motor control commands to perform physical compensation actions. During the correction process, the cabinet's posture is monitored in real time by a high-precision gyroscope, and the motor output is dynamically adjusted by a PID control algorithm to eliminate execution errors and obtain a delivery cabinet that is stable in the target posture. S53. Set the second fusion positioning error threshold, cabinet tilt angle threshold, and wind disturbance threshold. The cabinet tilt angle is directly measured by a gyroscope to obtain the real-time tilt angle; the real-time wind disturbance is obtained by high-frequency sampling using an accelerometer. If the following conditions are met simultaneously: second fusion positioning error < second fusion positioning error threshold, cabinet real-time tilt angle < cabinet tilt angle threshold, and real-time wind disturbance < wind disturbance threshold, then the triple material release threshold is met, and the material release command is triggered; otherwise, return to S4 and adjust the compensation parameters until the material release command is triggered.

9. A method for precise and targeted delivery of supplies using a drone delivery cabinet according to claim 1, characterized in that, S6 includes the following steps: S61. After triggering the resource release command, the resources will be released. S62. Set pressure threshold; the distributed pressure sensor array at the bottom of the delivery cabinet monitors the impact force in real time the moment the material touches the ground. When the sensor detects a continuous pressure ≥ the pressure threshold, it is determined to be a valid landing and the material delivery success notification is sent. If the pressure sensor does not detect a valid impact signal, the deployment is deemed abnormal, triggering the abnormal response plan.

10. A system for precise, targeted delivery of supplies using a drone delivery system, characterized in that: A method for accurately delivering supplies to a drone delivery cabinet as described in any one of claims 1-9, the system comprising an initial positioning and trajectory planning module, a target recognition and feature extraction module, a pose calculation and trajectory correction module, a compensation matrix generation module, a pose fine-tuning and release determination module, and a status feedback module.

Images