A mobile platform-based autonomous landing system and method for a rotary-wing UAV

Abstract

The present invention discloses a system and method for autonomous landing of a rotary-wing UAV based on a mobile platform. By designing a multi-scale nested combination mark to meet the recognition requirements of a monocular camera at different distance gradients, the drone's monocular camera maintains a recognizable mark at different altitudes during landing. Simultaneously, an auxiliary landing system calculates and estimates the mobile platform's speed and position, and uses the acquired drone's speed and position to assist in landing. This system and method achieves autonomous landing of the drone on the mobile platform by switching between three basic states: GPS navigation, visual cruise, and visual navigation. It features high landing efficiency and strong stability.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) control, and in particular relates to a system and method for autonomously landing a rotary-wing UAV based on a mobile platform. Background Art

[0002] In recent years, drone intelligent technology has developed rapidly, and drones are often used as auxiliary equipment in applications such as environmental monitoring, natural exploration, and disaster relief. However, the accident rate during the landing and recovery phase of a drone's mission far exceeds that of other stages. Therefore, to maximize drones' ability to take off, achieving precise autonomous landing on moving targets has become a research hotspot in the field of drone technology.

[0003] Autonomous drone landing relies on navigation technology, and traditional navigation techniques mostly rely on a combination of GPS and inertial navigation. GPS navigation suffers from large errors and lacks complete attitude information, making precise landing impossible. Inertial measurement units, on the other hand, suffer from accumulated errors, leading to reduced accuracy. Traditional navigation techniques are unable to meet the requirements for precise positioning. With technological advancements, visual navigation, recognized by institutions both domestically and internationally for its strong anti-interference capabilities, low cost, and high precision, has become widely used in drone landing applications.

[0004] Patent application number 201810735398.1 provides a mobile platform-based autonomous drone landing control system and method. The system includes a vision module, a landing target detection module, a drone pose estimation module, and a parameter comparison module. During landing, the vision module captures real-time images, performs threshold segmentation on the images, and processes them. The rotation and translation matrices of the drone relative to the landing target are calculated based on the processing results. Ultimately, the attitude angle is obtained, and the drone's pose information is estimated in real time to determine and adjust its flight. However, since the camera is fixed to the drone, drone vibration may affect the pose estimation data.

[0005] Patent application number 201911396466.7 provides a GPS- and vision-based autonomous drone landing system and method. Upon receiving a one-button return-to-home command, the drone maintains a normal flight altitude and uses GPS positioning to begin flying directly above the landing point. It then descends at a vertical speed of 5 m / s to a height of 5 meters directly above the target. The drone then descends to the ground at a vertical speed of 1 m / s, using the OpenMV visual positioning system for navigation. However, this method relies solely on simple color recognition and uses only the difference between the color target center and the drone's coordinates as a control parameter, resulting in significant control errors. The entire method also fails to address the transition between GPS and vision-based navigation.

[0006] The autonomous landing process of a drone needs to be constantly adjusted along with the landing process judgment criteria. In the existing technology, the stability of the drone landing process is poor. Summary of the Invention

[0007] Purpose of the invention: The purpose of the present invention is to provide a rotor UAV autonomous landing system based on a mobile platform. The purpose of the present invention is to provide a rotor UAV autonomous landing method based on a mobile platform.

[0008] Technical solution: The present invention relates to a mobile platform-based autonomous landing system for a rotary-wing UAV, wherein a combination mark for being recognized by the UAV's visual system includes a black square pattern, wherein the black square pattern is embedded with a white H-shaped pattern, wherein the white H-shaped pattern is embedded with a small-scale QR code and two groups of large-scale QR codes, wherein the two groups of large-scale QR codes are located on both sides of the small-scale QR code.

[0009] Preferably, the visual system includes a monocular camera and a pan / tilt platform capable of controlling the pitch angle of the monocular camera.

[0010] Preferably, the combined marker is connected to a mobile platform, and the mobile platform is provided with an auxiliary landing system for collecting speed differences and position differences between the UAV and the mobile platform.

[0011] Furthermore, the combined logo uses a diffuse reflective white soft rubber material as the background of the nested pattern, which can meet the recognition effect of the monocular camera at different distances and is affixed to the plane of the mobile platform.

[0012] Furthermore, the assisted landing system integrates a GPS navigation module, an attitude sensor, a remote communication module, and a speaker. The GPS navigation module and attitude sensor data are integrated to provide real-time estimates of the current speed and position of the mobile platform. The remote communication module receives the drone's position and speed data and transmits the speed and position differences between the drone and the mobile platform to the drone, guiding it toward the mobile platform.

[0013] Furthermore, the gimbal in the visual system is a single-axis gimbal, the monocular camera is a wide-angle camera, the gimbal is fixed directly below the drone, and the monocular camera is fixed to the gimbal, keeping the initial field of view of the monocular camera always facing downward. The single-axis gimbal can perform pitch control for searching for the combination mark and visually guiding landing.

[0014] Furthermore, the center of the white H-shaped pattern embedded in the black square pattern coincides with the center of the black square pattern, the small-scale QR code is located in the center of the black square pattern, the small-scale QR code and two groups of large-scale QR codes are arranged horizontally, the large-scale QR code is located on both sides of the small-scale QR code, the side length of the large-scale QR code is greater than that of the small-scale QR code, and the QR code is an aruco QR code.

[0015] Furthermore, the autonomous landing system also includes a remote controller, which is used to send a one-button landing command or cancel the landing command to manually control the drone.

[0016] The present invention provides a method for autonomously landing a rotary-wing UAV based on a mobile platform, the method comprising the following steps:

[0017] (1) After receiving the return command during flight, the UAV is controlled to fly toward the mobile platform at the maximum speed based on the position and speed difference between the UAV and the mobile platform;

[0018] (2) After the UAV flies to the vicinity of the mobile platform according to the position data of the mobile platform, it uses the visual system to detect the combination mark on the surface of the mobile platform until the target combination mark is found;

[0019] (3) The UAV recognizes the combined mark through the visual system to obtain the relative position distance between the UAV and the mobile platform, and the UAV obtains the relative height position with the mobile platform through the laser sensor; during the descent process of the UAV, the cascade PID method of position and speed is used in the horizontal direction to fly to the mobile platform, and the position is the relative position obtained by processing the combined mark image collected by the UAV visual system; in the height direction, the speed segmentation method is used to gradually descend to the mobile platform.

[0020] Preferably, the pitch angle difference and roll angle difference of the drone are judged in real time at the end of the drone's descent process. When the pitch angle difference or roll angle difference exceeds a preset value, the drone flies to the starting height of the end descent stage and re-executes the landing process of the end descent stage.

[0021] Preferably, the combination mark includes a black square pattern, the black square pattern is embedded with a white H-shaped pattern, the white H-shaped pattern is embedded with a small-scale QR code and two groups of large-scale QR codes, and the two groups of large-scale QR codes are located on both sides of the small-scale QR code. The specific steps of the method for identifying the combination mark are as follows:

[0022] (a) Obtain the relative height between the UAV and the ground or combined landmarks through a laser sensor;

[0023] (b) Image preprocessing: using the local threshold method to perform image binarization on the images collected by the UAV vision system;

[0024] (c) Extract candidate contours, screen out the quadrilateral contour and its four vertices, and when the drone altitude is within a first altitude range, determine whether there is an H-shaped pattern contour within the quadrilateral contour. If so, retain the quadrilateral contour and the H-shaped pattern contour; if not, discard the candidate contour. When the drone altitude is within a second altitude range, add screening for QR codes within the H-shaped pattern contour. When the drone altitude is within a third altitude range, only perform QR code recognition.

[0025] (d) Obtain the perspective matrix, set the minimum diagonal length of the quadrilateral as the side length of the new image after perspective transformation, and calculate the perspective matrix;

[0026] (e) Target recognition: Based on the relative height of the UAV, the H-shaped pattern outline and QR code in the candidate outline are identified.

[0027] Preferably, the specific steps of the method for identifying the H-shaped pattern outline in step (e) are as follows:

[0028] (S1.1) Performing perspective transformation on the dodecagonal contour points retained in the candidate contour and projecting them onto the new image;

[0029] (S1.2) H-shaped pattern contour recognition, using the scale-invariant moment to match the template contour through the Euclidean distance method, and performing two judgments, the first one is the original image, and the second one is the image rotated 90°;

[0030] (S1.3) Using the identified contour points as a reference, perform minimum rectangle fitting and extract the four vertices after fitting;

[0031] (S1.4) Using the inverse perspective matrix transformation, the positions of the four vertices in the original image are calculated as the four interior corner points.

[0032] (S1.5) Using the four vertices of the original quadrilateral as exterior corner points, we obtain eight ordered corner points.

[0033] Preferably, the specific steps of the method for identifying the QR code in step (e) are as follows:

[0034] (S2.1) Projecting the quadrilateral grayscale image onto the new image;

[0035] (S2.2) extracting and correcting image bit processing;

[0036] (S2.3) Complete the QR code decoding and recognition;

[0037] (S2.4) The identified quadrilateral contour points are used as its four corner points.

[0038] Preferably, if the target is lost during the landing of the UAV in step (3), the UAV rises at a set speed and executes a field of view enlargement mechanism to increase the longitudinal recognition range and the lateral recognition range of the visual system.

[0039] Furthermore, after receiving the return command during flight, the drone maintains the current flight altitude, locates itself through the GPS navigation module, and records the current speed V of the drone. 机 , mobile platform speed V 动, decompose the UAV speed along the direction of the mobile platform speed and the direction perpendicular to the mobile platform speed, and obtain V 机1 and V 机2 When the drone is behind the mobile platform, the speed is positive, otherwise it is negative. 动 >V 机1 , the drone cannot successfully move to the mobile platform, and the speaker prompts the mobile platform to slow down. 机1 With V 动 The difference is δ v , the difference between the current GPS position of the drone and the mobile platform is δ pos , in meters, δ v , δ pos As a frame of data, it is sent to the drone via remote communication.

[0040] Furthermore, due to the errors in the GPS positioning of the UAV and the mobile platform, the process of the UAV searching for a certain platform through vision is as follows: the UAV keeps the camera field of view vertically downward to acquire images for target detection, and determines whether the combination mark is within the field of view. If the target is not identified, the gimbal controls the camera 30° forward and downward, and the UAV descends to a height of 8.5 meters from the ground and deflects the fuselage for cruising. This process is still navigated according to the GPS position information until the target combination mark is found.

[0041] Furthermore, the drone identifies and tracks the target in real time. During the landing process, when the drone is between 2 and 10 meters, the drone descends at a speed of 3 m / s; when the drone is between 1 and 2 meters, the drone descends at a speed of 0.5 m / s; when the drone is between 0.3 and 1 meter, the drone descends at a speed of 0.3 m / s; when the drone is between 0.1 and 0.3 meters, the drone descends at a speed of 0.1 m / s. This process determines the angle difference δ between pitch and roll in real time. pitch and δ roll If the angle difference is greater than 5°, the drone will fly to 0.3 meters and execute the process again, otherwise it will continue to land. pitch and δ roll Through image processing, we can see that 0.1 meters is the height of the drone from the ground. When the height from the ground is 0.1 meters, it means that the drone has completed landing.

[0042] Furthermore, the drone identifies the nested patterns of the combined logo separately according to the distance gradient. When the drone is between 2 and 10 meters, it performs long-range recognition. Due to the poor accuracy of long-range QR codes, it only recognizes the "H in a square" logo, first identifying the black square background and then the white H logo. When the drone is between 1 and 2 meters, both the "H in a square" and the large-scale QR code can be clearly identified. This is the second stage of recognition, during which both are recognized simultaneously. Due to the lack of accuracy of large-scale QR codes in this altitude range and the fact that the "H in a square" logo easily fills the camera screen, and given the increased distortion closer to the camera edge, the long-range recognition data is primarily used, supplemented by the close-range recognition data, and the two data are then fused. When the drone is between 0.3 and 1 meter, it performs close-range recognition, recognizing only the QR code. When the drone is between 0.3 and 1 meter, data from the large-scale QR code is prioritized. When the drone is between 0.1 and 0.3 meters, only the small-scale QR code can be recognized due to the limited camera field of view. During landing, the drone moves in the direction of shortest distance from the mobile platform.

[0043] Furthermore, if the target is lost during landing, the drone ascends at 4 m / s and introduces a field-of-view expansion mechanism. This mechanism uses a monocular wide-angle camera to identify the nested symbols and employs a PD control method to control the gimbal's pitch angle, ensuring that the camera's optical axis is always aligned with the identified nested symbols, thus expanding the vertical recognition range. The horizontal arrangement of the QR codes in the designed nested symbols further expands the wide-angle camera's horizontal recognition range.

[0044] Furthermore, in step (a) of the combination mark recognition method, when the UAV is moving, the laser beam of the laser sensor may illuminate the combination mark on the ground or the mobile platform, and the real-time data only corresponds to one of the situations. Therefore, the relative height of the UAV and the ground or the combination mark can be obtained through the laser sensor.

[0045] Furthermore, in step (b), during image preprocessing, a local threshold method is used to perform image binarization in order to reduce the influence of uneven illumination, and a median method can be used to determine an adaptive threshold.

[0046] Furthermore, in step (c), the first altitude range is 2 to 10 meters, the second altitude range is 1 to 2 meters, and the third altitude range is 0.1 to 1 meter. Correspondingly, in step (d), different recognition methods are selected based on the relative altitude of the drone, including the "H in a square" mark and the Aruco QR code recognition method. When the relative altitude is 2 to 10 meters, the "H in a square" mark recognition algorithm is used. When the relative altitude is 1 to 2 meters, both methods are used simultaneously. When the relative altitude is 0.1 to 1 meter, the Aruco QR code recognition algorithm is used.

[0047] Furthermore, the pixel coordinates of the corner points can be obtained after the target recognition algorithm, and the relative position and posture information of the target in the camera coordinate system can be obtained after the coordinate system transformation. m ,Y w ,0) is mapped to the image pixel coordinate system coordinate point (u,v).

[0048] Solve the translation matrix T that maps the world coordinate system to the camera coordinate system = (T x T y T z ) T ; Among them, the internal parameter matrix of the monocular camera: can be obtained by performing distortion correction through the camera correction module of Matlab.

[0049]

[0050] It is the rotation matrix that maps the world coordinate system to the camera coordinate system, T=(T x T y T z ) T It is the translation matrix that maps the world coordinate system to the camera coordinate system;

[0051] The gimbal is fixed to the bottom of the aircraft, and the camera is fixed to the gimbal. The camera coordinate system coincides with the gimbal coordinate system. Since the gimbal can be pitched, the transformation relationship from the gimbal coordinate system to the aircraft coordinate system must also be calculated. The drone reads the gimbal attitude and its own attitude information and makes the difference to obtain the roll deviation α', pitch deviation β', and yaw deviation θ'. Therefore

[0052]

[0053] is the rotation matrix mapping the gimbal coordinate system to the body coordinate system; R b =R*R' is the rotation matrix that maps the world coordinate system to the body coordinate system.

[0054] From the above rotation matrix R b The relative position information between the UAV and the moving target is calculated using the translation vector T.

[0055] Beneficial effects: The combination mark designed by the present invention can meet the recognition effect of the monocular camera of the drone within a vertical height range of 10 meters, and can use the change in the field of view caused by the change in the relative distance of the monocular camera. The combination mark can be placed on any mobile platform as a mobile landing point. The landing conditions are prompted by the speaker through system calculation, which can avoid the failure of the drone to land due to the unknown speed of the mobile platform and increase the stability of autonomous landing. During the visual landing process, the longitudinal field of view is increased by pitching the gimbal, and the lateral field of view is increased by using a wide-angle camera in conjunction with the horizontally arranged QR code mark in the combination mark, which reduces the probability of losing the target to a certain extent. The mobile landing process is improved through the three steps of GPS navigation, visual cruising and visual landing. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a schematic plan view of the combined logo of the present invention;

[0057] Figure 2 This is a flow chart of the autonomous landing method of a rotary-wing UAV based on a mobile platform of the present invention;

[0058] Figure 3 For the present invention gps Definition graph;

[0059] Figure 4 This is a flow chart of the combined mark recognition method of the present invention. DETAILED DESCRIPTION

[0060] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0061] A mobile platform-based autonomous landing system for a rotary-wing UAV comprises a combination marker, an auxiliary landing system, a visual system, and a remote controller.

[0062] The combination sign uses a diffuse reflective white soft rubber material as the background of the nested pattern. The nested pattern is a planar multi-scale pattern that can meet the recognition effect of the monocular camera at different distances and is attached to the plane of the mobile platform. In order to ensure that the combination sign can be used to guide landing at different heights, the combination sign is designed in the form of an H-shaped and QR code nested. Figure 1 As shown, the combined logo has a white background with a black square pattern on it. The black square pattern has a white H-shaped pattern embedded in it. The white H-shaped pattern has a small-scale QR code and two groups of large-scale QR codes embedded in it. The center of the white H-shaped pattern embedded in the black square pattern coincides with the center of the black square pattern. The small-scale QR code is located in the center of the black square pattern. The small-scale QR code and the two groups of large-scale QR codes are arranged horizontally. The large-scale QR code is located on both sides of the small-scale QR code. The side length of the large-scale QR code is greater than that of the small-scale QR code. The QR code is an Aruco QR code.

[0063] In this embodiment, the specific design dimensions of the combined logo are as follows: the side length of the white square substrate is 500 mm; the side length of the black square pattern is 440 mm; the white H-shaped logo consists of two rectangles with a length of 380 mm and a width of 120 mm, and one rectangle with a length of 220 mm and a width of 140 mm; the side length of the large-scale QR code is 105 mm; and the side length of the small-scale QR code is 35 mm.

[0064] In this embodiment, the auxiliary landing system is fixed inside the black cube shell and placed in the black gaps above and below the white H-shaped sign in the combination sign. It integrates a GPS navigation module, an attitude sensor, a remote communication module, and a speaker. The GPS navigation module and the attitude sensor are integrated to perform data fusion to estimate the current speed and posture information of the mobile platform in real time, and obtain the speed and posture information of the drone through remote communication. The current speed of the drone is recorded as V 机 , mobile platform speed V 动 , decompose the UAV speed along the direction of the mobile platform speed and the direction perpendicular to the mobile platform speed, and obtain V 机1 and V 机2 When the drone is behind the mobile platform, the speed is positive, otherwise it is negative. 动 >V 机1 , the drone cannot successfully move to the mobile platform, and the speaker prompts the mobile platform to slow down. 机1 With V 动 The difference is δ v , the difference between the current GPS position of the drone and the mobile platform is δ pos , in meters, δ v , δ pos As a frame of data, it is sent to the drone via remote communication.

[0065] In this embodiment, the vision system includes a gimbal and a monocular camera. The gimbal is a single-axis gimbal for pitch control, and the monocular camera is a wide-angle camera. The gimbal is fixed directly below the drone, and the monocular camera is fixed to the gimbal, with the monocular camera's initial field of view always pointing downward. This allows the drone to search for the combined landmarks and visually guide the landing.

[0066] In this embodiment, the remote controller is connected to the drone signal and is used to send a one-key landing command or cancel the landing command to manually control the drone.

[0067] The landing method of the autonomous landing system of the rotor UAV based on the mobile platform is as follows: Figure 2 The specific steps are as follows:

[0068] Step 1: After receiving the return command during normal flight, the drone maintains the current flight altitude and locates itself through the GPS navigation module. If the GPS position difference is δpos Less than δ gps If the long-distance GPS navigation is completed, the process goes to step 2. Otherwise, the UAV flies to the mobile platform at the maximum speed. gps It is the distance difference between the UAV and the mobile platform during the tracking response time period. When using GPS navigation, the mobile platform moves first and then the UAV moves and tracks. The mobile platform and the UAV are abstracted as particles. Initially, the mobile platform and the UAV are at the same position, such as Figure 3 As shown in the figure, the position of the mobile platform is the position after a period of time, and the UAV is just about to start moving tracking. The distance between the two is δ gps .

[0069] In step 2, there is a discrepancy between the GPS positioning of the drone and the mobile platform. A visual cruise is required to search for the mobile platform. The process is as follows: the drone maintains its camera's vertical downward field of view to acquire images for target detection and determine whether the combined logo is within its field of view. If no target is identified, the gimbal controls the camera to point forward and downward 30°. The drone descends to a height of 8.5 meters above the ground and rotates to cruise. This process still uses GPS position information for navigation. If the target is found, the process proceeds to step 3.

[0070] Step 3: The UAV identifies and tracks the target in real time, obtains the relative position distance between the UAV and the mobile platform through the monocular camera recognition combination mark, and obtains the height position through the laser sensor. In the horizontal direction, the cascade PID method of position and speed is used to fly to the mobile platform, and the speed is δ v , the relative position is obtained by image processing; in the height direction, the drone gradually descends to the mobile platform using a speed segmentation method. In this embodiment, when the drone is between 2 and 10 meters, the drone descends at a speed of 3m / s; when the drone is between 1 and 2 meters, the drone descends at a speed of 0.5m / s; when the drone is between 0.3 and 1 meter, the drone descends at a speed of 0.3m / s; when the drone is between 0.1 and 0.3 meters, the drone descends at a speed of 0.1m / s; this process determines the angle difference δ between pitch and roll in real time pitch and δ roll If the angle difference is greater than 5°, the drone will fly to 0.3 meters and execute this process again, otherwise it will continue to land. pitch and δ roll Through image processing, we determine that 0.1 meters is the drone's height above the ground. When the height reaches 0.1 meters, the drone has landed. If the target is lost during this process, the drone ascends at 4 meters per second and returns to step 2. A field of view expansion mechanism is introduced during this process.

[0071] In this embodiment, the strategy for using the visual system to recognize the combined symbol in steps 2 and 3 is to first read the drone's current distance from the combined symbol or the ground, and then identify the nested patterns within the combined symbol based on the distance gradient. When the drone is between 2 and 10 meters, it performs long-range recognition. Due to the poor accuracy of QR codes at long distances, it only recognizes the "H in a square" symbol. It first recognizes the black square background, then the white H symbol. When the drone is between 1 and 2 meters, both the "H in a square" and the large-scale QR code are clearly recognized. This is the second stage of recognition, in which both are recognized simultaneously, with long-range recognition data as the primary focus and close-range recognition data as the supplementary data. A weighted average method is used for data fusion. When the drone is between 0.3 and 1 meter, it performs close-range recognition, recognizing only the QR code. When the drone is between 0.3 and 2 meters, it prioritizes data from the large-scale QR code. When the drone is between 0.1 and 0.3 meters, it can only recognize the small-scale QR code due to the limited camera field of view.

[0072] Furthermore, the field of view expansion mechanism introduced in step 3 utilizes a monocular wide-angle camera to identify the combined logo. The PD control method is used to control the pan / tilt angle, ensuring that the camera's optical axis is always aligned with the identified nested logo, thereby expanding the vertical recognition range. The pixel difference between the nested logo and the image pixel center in the pixel coordinate system is used as the input parameter. Because the QR codes in the designed combined logo are arranged horizontally, the wide-angle camera's horizontal recognition range can be further expanded.

[0073] In this embodiment, during the landing process of the drone, the drone flies to the mobile platform in accordance with the principle that the drone moves in the direction with the shortest distance from the mobile platform.

[0074] In this embodiment, the combined sign recognition algorithm is used during the landing process of the drone. Figure 4 The specific steps are as follows:

[0075] Step S1: Obtain the relative height between the drone and the ground or the combined mark through the laser sensor.

[0076] Step S2: Image preprocessing: To reduce the influence of uneven illumination, a local threshold method is used to perform image binarization. In this embodiment, a median method is used to determine an adaptive threshold.

[0077] Step S3: Extract candidate contours. Use a polygonal approximation algorithm to filter out the quadrilateral contour and its four vertices, and sort them clockwise. At the same time, determine whether there are sub-contours of the quadrilateral contour based on the relative height. A sub-contour refers to a contour with twelve sides. When the height is between 1 and 10 meters, the "H in a square" mark can be identified, and a judgment is required. If it exists, the quadrilateral contour and its sub-contours are retained. If not, the candidate contour is discarded. When the height is between 0.1 and 1 meter, only the Aruco QR code is recognized, and no judgment is required.

[0078] Step S4: Calculate the perspective matrix. Set the minimum diagonal length of the quadrilateral as the side length of the new image after perspective transformation to calculate the perspective matrix.

[0079] Step S5: Target identification. Different identification methods are selected based on the relative height, including the "H in a square" mark and the Aruco QR code. When the relative height is 2 to 10 meters, the "H in a square" mark recognition algorithm is used. When the relative height is 1 to 2 meters, both methods are used simultaneously. When the relative height is 0.1 to 1 meter, the Aruco QR code recognition algorithm is used.

[0080] After the target recognition algorithm, the pixel coordinates of the corner points can be obtained, and the relative position and posture information of the target in the camera coordinate system can be obtained through coordinate system transformation. m ,Y w ,0) is mapped to the image pixel coordinate system coordinate point (u,v).

[0081] Solve the translation matrix T that maps the world coordinate system to the camera coordinate system = (T x T y T z ) T Among them, the intrinsic parameter matrix of the monocular camera can be obtained by performing distortion correction through the camera correction module of Matlab.

[0082]

[0083] It is the rotation matrix that maps the world coordinate system to the camera coordinate system, T=(T x T y T z ) T It is the translation matrix that maps the world coordinate system to the camera coordinate system;

[0084] The gimbal is fixed to the bottom of the aircraft, and the camera is fixed to the gimbal. The camera coordinate system and the gimbal coordinate system coincide. Since the gimbal can be pitched, the transformation relationship from the gimbal coordinate system to the aircraft coordinate system must also be calculated. The drone reads the gimbal attitude and its own attitude information and makes the difference to obtain the roll deviation α', pitch deviation β', and yaw deviation θ'. Therefore

[0085]

[0086] is the rotation matrix mapping the gimbal coordinate system to the body coordinate system; R b =R*R' is the rotation matrix that maps the world coordinate system to the body coordinate system.

[0087] From the above rotation matrix R b The relative position information between the UAV and the moving target is calculated using the translation vector T.

[0088] Furthermore, the identification method for "H in the square" is as follows Figure 4 The specific steps are as follows:

[0089] (S1.1) Performing perspective transformation on the dodecagonal contour points retained in the candidate contour and projecting them onto the new image;

[0090] (S1.2) H-shaped pattern contour recognition, using the scale-invariant moment to match the template contour through the Euclidean distance method, and performing two judgments, the first one is the original image, and the second one is the image rotated 90°;

[0091] (S1.3) Using the identified contour points as a reference, perform minimum rectangle fitting and extract the four vertices after fitting;

[0092] (S1.4) Using the inverse perspective matrix transformation, the positions of the four vertices in the original image are calculated as the four interior corner points.

[0093] (S1.5) Using the four vertices of the original quadrilateral as exterior corner points, we obtain eight ordered corner points.

[0094] Furthermore, for the identification method of the QR code, Figure 4 The specific steps are as follows:

[0095] (S2.1) Projecting the quadrilateral grayscale image onto the new image;

[0096] (S2.2) extracting and correcting image bit processing;

[0097] (S2.3) Complete the QR code decoding and recognition;

[0098] (S2.4) The identified quadrilateral contour points are used as its four corner points.

[0099] In summary, the present invention designs a multi-scale nested composite marker that satisfies the monocular camera's recognition requirements at varying distance gradients, ensuring that the drone's monocular camera maintains a recognizable marker at all altitudes during landing. Simultaneously, the assisted landing system calculates and estimates the mobile platform's speed and position, along with the acquired drone's speed and position, to assist in landing. This autonomous rotorcraft landing system and method achieves autonomous landing on a mobile platform by switching between three basic states: GPS navigation, visual cruise control, and visual navigation. It boasts high landing efficiency and robust stability.

Claims

1. A method for autonomous landing of a rotary-wing UAV based on a mobile platform, characterized in that: The method is implemented using a mobile platform-based autonomous landing system for a rotary-wing UAV. The combined mark for recognition by the UAV's visual system includes a black square pattern embedded with a white H-shaped pattern, which in turn embeds a small-scale QR code and two sets of large-scale QR codes, with the two sets of large-scale QR codes located on either side of the small-scale QR code. The visual system includes a monocular camera and a gimbal capable of controlling the pitch angle of the monocular camera. The combined mark is connected to a mobile platform, which is equipped with an auxiliary landing system for collecting speed and position differences between the UAV and the mobile platform. The method comprises the following steps: (1) After receiving the return command during flight, the UAV is controlled to fly toward the mobile platform at the maximum speed based on the position and speed difference between the UAV and the mobile platform; (2) After the UAV flies to the vicinity of the mobile platform based on the mobile platform's position data, it uses the visual system to detect the combination mark on the surface of the mobile platform until it finds the target combination mark; (3) The UAV uses the visual system to identify the combined signs to obtain the relative position distance between the UAV and the mobile platform, and the UAV uses the laser sensor to obtain the relative height position with the mobile platform; During the descent process, the UAV uses a cascade PID method of position and speed in the horizontal direction to fly to the mobile platform. The position is the relative position obtained by processing the combined landmark image collected by the UAV vision system; in the height direction, a speed segmentation method is used to gradually descend to the mobile platform. The specific steps of the method for identifying the combination mark are as follows: (a) Obtain the relative height between the UAV and the ground or combined landmarks using a laser sensor; (b) Image preprocessing: using the local threshold method to perform image binarization on the images collected by the UAV vision system; (c) Extract candidate contours, screen out the quadrilateral contour and its four vertices. When the drone's altitude is within the first altitude range, determine whether there is an H-shaped pattern contour within the quadrilateral contour. If so, retain the quadrilateral contour and the H-shaped pattern contour; if not, discard the candidate contour. When the drone's altitude is within the second altitude range, add a QR code filter within the H-shaped pattern contour. When the drone's altitude is within the third altitude range, only perform QR code recognition. (d) Calculate the perspective matrix. Set the minimum diagonal length of the quadrilateral as the side length of the new image after perspective transformation to calculate the perspective matrix. (e) Target recognition: identifying the H-shaped pattern outline and QR code in the candidate outline based on the relative altitude of the UAV; The specific steps of the method for identifying the H-shaped pattern outline in step (e) are as follows: (S1.1) performing perspective transformation on the dodecagonal contour points retained in the candidate contour and projecting them onto the new image; (S1.2) H-shaped pattern contour recognition, using scale-invariant moments to match the template contour using the Euclidean distance method, with two judgments performed: the first is the original image, and the second is the image rotated 90°; (S1.3) Using the identified contour points as a reference, perform minimum rectangle fitting and extract the four vertices after fitting; (S1.4) Using the inverse perspective matrix transformation, determine the positions of the four vertices in the original image as the four interior corner points. (S1.5) Using the four vertices of the original quadrilateral as exterior corner points, we obtain eight ordered corner points.

2. The method for autonomous landing of a rotary-wing UAV based on a mobile platform according to claim 1, characterized in that: At the end of the drone's descent process, the pitch angle difference and roll angle difference of the drone are judged in real time. When the pitch angle difference or roll angle difference exceeds the preset value, the drone flies to the starting height of the final descent stage and re-executes the final descent landing process.

3. The method for autonomous landing of a rotary-wing UAV based on a mobile platform according to claim 1, characterized in that: The specific steps of the QR code recognition method in step (e) are as follows: (S2.1) Projecting the quadrilateral grayscale image onto the new image; (S2.2) extracting and correcting image bit processing; (S2.3) Complete the QR code decoding and recognition; (S2.4) The identified quadrilateral contour points are used as its four corner points.

4. The method for autonomous landing of a rotary-wing UAV based on a mobile platform according to claim 1, characterized in that: If the target is lost during the landing of the UAV in step (3), the UAV ascends at the set speed and executes the field of view enlargement mechanism to improve the longitudinal recognition range and lateral recognition range of the visual system.

Images