A method for autonomous precision landing of unmanned aerial vehicle
Through visual positioning and real-time tracking technology, combined with RTK and GPS feedback, hardware delay is eliminated, and high-altitude PID and low-altitude PID + ping-pong control are adopted, solving the complexity of autonomous and precise landing of drones and achieving high-level under complex conditions Accuracy and high stability landing.
Patent Information
- Application Number
- CN202210543572.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-19
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2042-05-19
AI Technical Summary
The autonomous precise landing technology of drones poses great risks and technical difficulties. Due to the area of the automatic airport apron and meteorological conditions, the landing accuracy and stability requirements are extremely high. It is difficult for the existing technology to achieve high timeliness and timeliness under complex weather conditions. High-precision control.
Adopt visual positioning and visual tracking technology, combining apriltag positioning marking and KCF real-time tracking technology, combining RTK and GPS speed feedback to eliminate lens hardware delays, using a combination of high-altitude PID tracking and low-altitude PID + ping-pong control, through multi-data source Kalman filtering and Speed integral compensation improves the accuracy of drone position feedback and the robustness of control system.
It realizes millisecond-level drone position feedback, improves landing accuracy and stability of the control system, meets the high-timedness needs under complex weather conditions, and significantly improves the landing accuracy of the drone.
Smart Images

Figure CN114779798B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an autonomous precision landing method for an unmanned aerial vehicle, belonging to the technical field of unmanned aerial vehicle control methods. Background Art
[0002] With the continuous development of advanced technologies such as information technology, communication technology, and computer technology, drones are increasingly used in modern warfare and have become an important development target for all countries at present and in the future. Not only in the military field, but also in agricultural production, disaster warning, remote sensing, telemetry, and aerial photography, drones are playing an increasingly important role.
[0003] The control of the UAV system is a human-in-the-loop control, and the ground control system is equivalent to the cockpit of a manned aircraft. Ground operators monitor the flight status of the UAV through the ground control system and make corresponding adjustments and interventions. The ground control system is the command center of the entire UAV system and the "brain" that controls the UAV to complete related flight missions. As the terminal for UAV control and flight mission formulation, the ground control system is also called the "mission planning and control system." The ground control system is an indispensable component of the UAV system and an important guarantee for the flight safety of the UAV and the completion of the flight mission. It has great practical significance for the research and development of the UAV ground control system.
[0004] The research on autonomous precision landing technology for drones is a key link with great risks and technical difficulties in the process of fully autonomous operation. It has a great impact on the stability and reliability of the entire system. Limited by the area of the automatic airport apron and the influence of meteorological conditions during landing, the landing accuracy and stability of drones are extremely high. Summary of the invention
[0005] The purpose of the present invention is to provide a method for autonomous and precise landing of unmanned aerial vehicles. The method is based on visual positioning and visual tracking technology, adopts the internationally common Apriltag positioning marker, KCF real-time tracking technology, and combines RTK speed feedback and GPS speed feedback to achieve the performance indicators of eliminating lens hardware delay and millisecond-level unmanned aerial vehicle position feedback; the high-altitude PID tracking program can track the movement of the aircraft well, and the combination of low-altitude PID control + ping-pong control effectively improves the robustness of the control program and effectively reduces the influence of delay on the stability of the control system, meets the high timeliness of the feedback path required for unmanned aerial vehicle control under complex weather conditions, provides the necessary conditions for precise landing control, and significantly improves the landing accuracy of the unmanned aerial vehicle, effectively solving the above-mentioned problems existing in the background technology.
[0006] The technical solution of the present invention is: a method for autonomous precision landing of a drone, comprising the following steps:
[0007] Step 1: Based on the UAV precise positioning technology of visual positioning and visual tracking, the positioning theory and tracking of UAVs are carried out. In view of the complex background, high security requirements and the misrecognition rate required to be less than 5% in the application scenario, a suitable positioning form is selected. According to the characteristics of image changes during the scene movement, a video tracking algorithm that meets the requirements is selected, and the above algorithms and models are verified to determine the solution with the best overall performance;
[0008] Step 2: Based on the control algorithm of PID control and ping-pong control, the delay time uncertainty system is optimized, the robustness of the algorithm is determined, and a set of aircraft dynamic models are identified according to the aircraft data as the basis for theoretical simulation. According to the characteristics of the aircraft landing, the landing phase is divided into high-altitude and low-altitude segments, and different control combinations are used to greatly improve the safety and real-time performance of the system.
[0009] Step 3: Based on RTK data and GPS data, combined with the hardware delay elimination algorithm of integral compensation, and with sufficient knowledge of the camera hardware characteristics, multi-data source Kalman filtering and velocity integral compensation are used to effectively improve the accuracy of the aircraft position, provide the control algorithm with high-reliability position feedback, and improve the control quality of the system;
[0010] Step 4, based on the control algorithm in step 2 and the data feedback in step 3, when the feedback amount is accurate enough and the feedback speed is fast enough, combine effective control theory and control parameters to ensure the high landing accuracy of the UAV.
[0011] In step 1, the video tracking algorithm is used to accurately locate the spatial position and attitude of the camera mounted on the drone. The KCF tracking algorithm mainly solves the problem of long positioning recognition time and poor real-time performance. The AprilTag position estimation and positioning system is applied, and the key parameters such as pixel distance ratio are updated faster. Step 1 includes the following steps:
[0012] Step 1.1, study the visual positioning algorithm, start from the principle of coordinate transformation, and derive the spatial posture formula;
[0013] Step 1.2, introduce the kernel function into the KCF visual tracking algorithm, and perform nonlinear mapping on the sample input of the linear regression to regress the nonlinear model, so as to better fit the tracking model and improve the tracking accuracy.
[0014] The step 1.1 comprises the following steps:
[0015] Step 1.1.1, during the landing process of the aircraft, estimate its position and attitude and perform coordinate space conversion; the coordinate systems involved include the world coordinate system, the body coordinate system, the camera coordinate system and the image coordinate system;
[0016] Step 1.1.2, constructing a camera imaging model, including a pinhole model and a CCD camera imaging model;
[0017] Step 1.1.3, based on the coordinate transformation in step 1.1.1 and the camera imaging model in step 1.1.2, establish a mapping relationship from the coordinates of the object space point in the camera coordinate system to the coordinates of the corresponding image point in the image pixel coordinate system;
[0018] Step 1.1.4, establish the transformation relationship from the body coordinate system to the camera coordinate system, and obtain the position and attitude of the aircraft in the world coordinate system based on the image information;
[0019] Step 1.1.5, for the AprilTag type pose calibration code, after detecting the square edge of the code, decode it according to the custom rules; during the decoding process, it is allowed to block part of the code content, and codes of different sizes can be nested in the blank area;
[0020] Step 1.1.6, for the SIFT feature matching class code, filter the feature points after matching and calculate the affine transformation matrix;
[0021] Step 1.1.7, derive the rotation angle of each axis of XYZ according to the affine transformation matrix, and correct the transformed image to obtain the pixel distance ratio;
[0022] Step 1.1.8, the rotation angle can be solved according to the homography transformation matrix to obtain the aircraft camera attitude.
[0023] The 1.2 includes the following steps:
[0024] Step 1.2.1, collect the image blocks in the rectangular frame, and then construct a training sample set with the target object located at different positions in the rectangular frame through circular shift;
[0025] Step 1.2.2, assign different labels according to the distance of the target object from the center of the image block; since the training sample set is obtained by cyclic shifting the image block in the first frame, the entire sample set is a circulant matrix. The elements of each column of the circulant matrix are obtained by cyclically shifting the elements of the previous column. The circulant matrix can be diagonalized by the Fourier matrix, so the linear regression training can be accelerated by calculation;
[0026] Step 1.2.3, train a linear regression model based on the training sample set and its labels;
[0027] Step 1.2.4, introduce the kernel function to nonlinearly map the sample input of linear regression to regress the nonlinear model;.
[0028] The step 2 is based on the control algorithm of PID control and ping-pong control, and adopts different control combinations to greatly improve the security and real-time performance of the system; the step 2 includes the following steps:
[0029] Step 2.1, automatic disturbance rejection optimization of PID control algorithm;
[0030] Step 2.2, ping-pong control system construction;
[0031] Step 2.3, based on the anti-disturbance optimization of the PID control algorithm in step 2.1 and the construction of the ping-pong control system in step 2.2, build the overall framework of the PID-ping-pong control system.
[0032] The step 2.1 includes the following steps:
[0033] Step 2.1.1, pre-process the given signal and arrange the transition process according to the tolerance of the controlled object; the transition process makes the error change asymptotically rather than suddenly, achieving the control goal of no overshoot and fast tracking, and solving the contradiction between the rapidity and overshoot of PID control; using the characteristic of tracking differentiator that can track the signal, a nonlinear tracking differentiator is used in the ADRC to arrange the transition process;
[0034] Step 2.1.2, using a nonlinear differential tracker with very low noise amplification effect to extract the differential of the given signal, and using an extended observer to extract the differential of the feedback signal;
[0035] In step 2.1.3, the extended state observer is used to estimate the total disturbance acting on the system in real time and compensate for it; zero static error can be achieved without introducing integral feedback, thus avoiding the side effects of integral feedback.
[0036] In step 3, based on RTK data and GPS data, combined with the hardware delay elimination algorithm of integral compensation, and with sufficient knowledge of the camera hardware characteristics, a multi-data source Kalman filter and velocity integral compensation method are used; specifically, the following steps are included:
[0037] Step 3.1, using the stored speed history, after integration, it becomes the position deviation, and the position deviation compensation and the visual positioning position are summed; the stored speed value at each sampling moment is multiplied by the sampling duration, and after the effective window length is determined by timestamp comparison, the product is summed to obtain the actual estimated value of the position deviation;
[0038] Step 3.2: synchronize timestamps using C ++ The mutex lock in the structure realizes data sharing between the integral estimation thread and the feedback result calculation thread. The timestamp variable in the structure can be used to easily obtain and compare the order of sampling moments.
[0039] Step 3.3, overall design of UAV precision autonomous landing middleware; UAV precision autonomous landing middleware is a C++ class implementation of autonomous landing, which adopts a modular bottom-up design approach and implements code recognition thread, KCF tracking thread, integral estimation thread, position acquisition function and landing control function.
[0040] The step 3.3 includes the following steps:
[0041] Step 3.3.1, initialize the class and image cache, and start the integral estimation thread, code recognition thread and KCF tracking thread;
[0042] Step 3.3.2, obtain the drone speed according to the set sampling time and store it in the circular queue;
[0043] Step 3.3.3, identify the AprilTag positioning code and store the identification result in the member variable of the class;
[0044] Step 3.3.4, divide the ROI area using the code information stored in the member variable as the center point, and regularly update the ROI area range;
[0045] Step 3.3.5, position correction, after correction, return to the PID-ping-pong control function to calculate the control quantity;
[0046] Step 3.3.6, segment according to the set value during class initialization, use PID control at high altitude and PID-ping-pong control with dead zone at low altitude.
[0047] The step 4 includes the following steps:
[0048] Step 4.1: Apply the control algorithm in step 2;
[0049] Step 4.2: Obtain the feedback data of the drone at high altitude and low altitude in step 3;
[0050] Step 4.3: Combine effective control theory and control parameters to control the UAV to land accurately.
[0051] The beneficial effects of the present invention are as follows: based on visual positioning and visual tracking technology, the internationally used Apriltag positioning marker and KCF real-time tracking technology are adopted, and RTK speed feedback and GPS speed feedback are combined to achieve the performance indicators of eliminating lens hardware delay and millisecond-level UAV position feedback; the high-altitude PID tracking program can track the aircraft movement well, and the combination of low-altitude PID control + ping-pong control effectively improves the robustness of the control program and effectively reduces the influence of delay on the stability of the control system, meets the high timeliness of the feedback path required for UAV control under complex weather conditions, provides the necessary conditions for precise landing control, and significantly improves the landing accuracy of the UAV. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] Figure 1 is a flow chart of the method of the present invention;
[0053] Figure 2 The corresponding relationship between the coordinate systems of the aircraft visual landing in the method of the present invention;
[0054] Figure 3 Schematic diagram of ping-pong control of the method of the present invention
[0055] Figure 4 It is a block diagram of the PID-ping-pong control system of the method of the present invention;
[0056] Figure 5 It is the basic principle diagram of integral compensation of the method of the present invention;
[0057] Figure 6 A timestamp synchronization flow chart of the method of the present invention;
[0058] Figure 7 The overall block diagram of the UAV precise autonomous landing middleware of the method of the present invention;
[0059] Figure 8 This is an internal flow chart of the initialization module of the present invention. DETAILED DESCRIPTION
[0060] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments, which are preferred embodiments of the present invention. It should be understood that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments; it should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0061] A method for autonomous precision landing of a UAV comprises the following steps:
[0062] Step 1: Based on the UAV precise positioning technology of visual positioning and visual tracking, the positioning theory and tracking of UAVs are carried out. In view of the complex background, high security requirements and the misrecognition rate required to be less than 5% in the application scenario, a suitable positioning form is selected. According to the characteristics of image changes during the scene movement, a video tracking algorithm that meets the requirements is selected, and the above algorithms and models are verified to determine the solution with the best overall performance;
[0063] Step 2: Based on the control algorithm of PID control and ping-pong control, the delay time uncertainty system is optimized, the robustness of the algorithm is determined, and a set of aircraft dynamic models are identified according to the aircraft data as the basis for theoretical simulation. According to the characteristics of the aircraft landing, the landing phase is divided into high-altitude and low-altitude segments, and different control combinations are used to greatly improve the safety and real-time performance of the system.
[0064] Step 3: Based on RTK data and GPS data, combined with the hardware delay elimination algorithm of integral compensation, and with sufficient knowledge of the camera hardware characteristics, multi-data source Kalman filtering and velocity integral compensation are used to effectively improve the accuracy of the aircraft position, provide the control algorithm with high-reliability position feedback, and improve the control quality of the system;
[0065] Step 4, based on the control algorithm in step 2 and the data feedback in step 3, when the feedback amount is accurate enough and the feedback speed is fast enough, combine effective control theory and control parameters to ensure the high landing accuracy of the UAV.
[0066] In step 1, the video tracking algorithm is used to accurately locate the spatial position and attitude of the camera mounted on the drone. The KCF tracking algorithm mainly solves the problem of long positioning recognition time and poor real-time performance. The AprilTag position estimation and positioning system is applied, and the key parameters such as pixel distance ratio are updated faster. Step 1 includes the following steps:
[0067] Step 1.1, study the visual positioning algorithm, start from the principle of coordinate transformation, and derive the spatial posture formula;
[0068] Step 1.2, introduce the kernel function into the KCF visual tracking algorithm, and perform nonlinear mapping on the sample input of the linear regression to regress the nonlinear model, so as to better fit the tracking model and improve the tracking accuracy.
[0069] The step 1.1 comprises the following steps:
[0070] Step 1.1.1, during the landing process of the aircraft, estimate its position and attitude and perform coordinate space conversion; the coordinate systems involved include the world coordinate system, the body coordinate system, the camera coordinate system and the image coordinate system;
[0071] Step 1.1.2, constructing a camera imaging model, including a pinhole model and a CCD camera imaging model;
[0072] Step 1.1.3, based on the coordinate transformation in step 1.1.1 and the camera imaging model in step 1.1.2, establish a mapping relationship from the coordinates of the object space point in the camera coordinate system to the coordinates of the corresponding image point in the image pixel coordinate system;
[0073] Step 1.1.4, establish the transformation relationship from the body coordinate system to the camera coordinate system, and obtain the position and attitude of the aircraft in the world coordinate system based on the image information;
[0074] Step 1.1.5, for the AprilTag type pose calibration code, after detecting the square edge of the code, decode it according to the custom rules; during the decoding process, it is allowed to block part of the code content, and codes of different sizes can be nested in the blank area;
[0075] Step 1.1.6, for the SIFT feature matching class code, filter the feature points after matching and calculate the affine transformation matrix;
[0076] Step 1.1.7, derive the rotation angle of each axis of XYZ according to the affine transformation matrix, and correct the transformed image to obtain the pixel distance ratio;
[0077] Step 1.1.8, the rotation angle can be solved according to the homography transformation matrix to obtain the aircraft camera attitude.
[0078] The 1.2 includes the following steps:
[0079] Step 1.2.1, collect the image blocks in the rectangular frame, and then construct a training sample set with the target object located at different positions in the rectangular frame through circular shift;
[0080] Step 1.2.2, assign different labels according to the distance of the target object from the center of the image block; since the training sample set is obtained by cyclic shifting the image block in the first frame, the entire sample set is a circulant matrix. The elements of each column of the circulant matrix are obtained by cyclically shifting the elements of the previous column. The circulant matrix can be diagonalized by the Fourier matrix, so the linear regression training can be accelerated by calculation;
[0081] Step 1.2.3, train a linear regression model based on the training sample set and its labels;
[0082] Step 1.2.4, introduce the kernel function to nonlinearly map the sample input of linear regression to regress the nonlinear model;.
[0083] The step 2 is based on the control algorithm of PID control and ping-pong control, and adopts different control combinations to greatly improve the security and real-time performance of the system; the step 2 includes the following steps:
[0084] Step 2.1, automatic disturbance rejection optimization of PID control algorithm;
[0085] Step 2.2, ping-pong control system construction;
[0086] Step 2.3, based on the anti-disturbance optimization of the PID control algorithm in step 2.1 and the construction of the ping-pong control system in step 2.2, build the overall framework of the PID-ping-pong control system.
[0087] The step 2.1 includes the following steps:
[0088] Step 2.1.1, pre-process the given signal and arrange the transition process according to the tolerance of the controlled object; the transition process makes the error change asymptotically rather than suddenly, achieving the control goal of no overshoot and fast tracking, and solving the contradiction between the rapidity and overshoot of PID control; using the characteristic of tracking differentiator that can track the signal, a nonlinear tracking differentiator is used in the ADRC to arrange the transition process;
[0089] Step 2.1.2, using a nonlinear differential tracker with very low noise amplification effect to extract the differential of the given signal, and using an extended observer to extract the differential of the feedback signal;
[0090] In step 2.1.3, the extended state observer is used to estimate the total disturbance acting on the system in real time and compensate for it; zero static error can be achieved without introducing integral feedback, thus avoiding the side effects of integral feedback.
[0091] In step 3, based on RTK data and GPS data, combined with the hardware delay elimination algorithm of integral compensation, and with sufficient knowledge of the camera hardware characteristics, a multi-data source Kalman filter and velocity integral compensation method are used; specifically, the following steps are included:
[0092] Step 3.1, use the stored speed history to convert it into position deviation after integration, sum the position deviation compensation and the position of visual positioning; multiply the stored speed value at each sampling moment by the sampling duration, determine the effective window length by timestamp comparison, and sum the products to obtain the actual estimated value of the position deviation;
[0093] Step 3.2: synchronize timestamps using C ++ The mutex lock in the structure realizes data sharing between the integral estimation thread and the feedback result calculation thread. The timestamp variable in the structure can be used to easily obtain and compare the order of sampling moments.
[0094] Step 3.3, overall design of UAV precision autonomous landing middleware; UAV precision autonomous landing middleware is a C++ class implementation of autonomous landing, which adopts a modular bottom-up design approach and implements code recognition thread, KCF tracking thread, integral estimation thread, position acquisition function and landing control function.
[0095] The step 3.3 includes the following steps:
[0096] Step 3.3.1, initialize the class and image cache, and start the integral estimation thread, code recognition thread and KCF tracking thread;
[0097] Step 3.3.2, obtain the drone speed according to the set sampling time and store it in the circular queue;
[0098] Step 3.3.3, identify the AprilTag positioning code and store the identification result in the member variable of the class;
[0099] Step 3.3.4, divide the ROI area using the code information stored in the member variable as the center point, and regularly update the ROI area range;
[0100] Step 3.3.5, position correction, after correction, return to the PID-ping-pong control function to calculate the control quantity;
[0101] Step 3.3.6, segment according to the set value during class initialization, use PID control at high altitude and PID-ping-pong control with dead zone at low altitude.
[0102] The step 4 includes the following steps:
[0103] Step 4.1: Apply the control algorithm in step 2;
[0104] Step 4.2: Obtain the feedback data of the drone at high altitude and low altitude in step 3;
[0105] Step 4.3: Combine effective control theory and control parameters to control the UAV to land accurately.
[0106] In summary, the present invention firstly adopts AprilTag positioning marker and KCF real-time tracking technology, combined with RTK speed feedback and GPS speed feedback, to achieve the performance index of eliminating lens hardware delay and millisecond-level UAV position feedback; then adopts high-altitude PID control, low-altitude PID control + ping-pong control combined scheme, the high-altitude PID tracking program can track the aircraft movement well, the combination of low-altitude PID control + ping-pong control effectively improves the robustness of the control program, and effectively reduces the influence of delay on the stability of the control system; secondly, the speed information provided by the UAV is fully utilized, the integral interpolation method is adopted, and the Kalman filtering technology is combined to effectively eliminate the lens hardware delay of 280ms, which is of great significance to improving the stability of the control system; finally, the precise landing of the multi-rotor UAV in various situations is realized, which can be conveniently used for fixed machine nests and mobile machine nests, realizes the fixed deployment and mobile deployment of the machine nest, and effectively expands the practical field of multi-rotor UAVs.
[0107] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, and it is not a limitation on the protection scope of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the protection scope of the present invention.
Claims
1. A method for autonomous precision landing of an unmanned aerial vehicle, characterized in that The following steps are involved: Step 1: Based on the UAV precise positioning technology of visual positioning and visual tracking, the positioning theory and tracking of UAVs are carried out. In view of the complex background, high security requirements and the misrecognition rate required to be less than 5% in the application scenario, a suitable positioning form is selected. According to the characteristics of image changes during the scene movement, a video tracking algorithm that meets the requirements is selected, and the above algorithms and models are verified to determine the solution with the best overall performance; Step 2: Based on the control algorithm of PID control and ping-pong control, the delay time uncertainty system is optimized, the robustness of the algorithm is determined, and a set of aircraft dynamic models are identified according to the aircraft data as the basis for theoretical simulation. According to the characteristics of the aircraft landing, the landing phase is divided into high-altitude and low-altitude segments, and different control combinations are used to greatly improve the safety and real-time performance of the system. Step 3: Based on RTK data and GPS data, combined with the hardware delay elimination algorithm of integral compensation, and with sufficient knowledge of the camera hardware characteristics, multi-data source Kalman filtering and velocity integral compensation are used to effectively improve the accuracy of the aircraft position, provide the control algorithm with high-reliability position feedback, and improve the control quality of the system; Step 4: Based on the control algorithm in step 2 and the data feedback in step 3, and provided that the feedback is accurate enough and the feedback speed is fast enough, effective control theory and control parameters are combined to ensure the high landing accuracy of the UAV. In step 1, the video tracking algorithm is used to accurately locate the spatial position and posture of the camera mounted on the drone. The KCF tracking algorithm solves the problem of long positioning recognition time and poor real-time performance. The AprilTag position estimation and positioning system is applied, and the pixel distance ratio is updated faster. Step 1 includes the following steps: Step 1.1, study the visual positioning algorithm, start from the principle of coordinate transformation, and derive the spatial posture formula; Step 1.2, introduce the kernel function into the KCF visual tracking algorithm, and perform nonlinear mapping on the sample input of the linear regression to regress the nonlinear model, so as to better fit the tracking model and improve the tracking accuracy; The step 1.1 comprises the following steps: Step 1.1.1, during the landing process of the aircraft, estimate its position and attitude and perform coordinate space conversion; the coordinate systems involved include the world coordinate system, the body coordinate system, the camera coordinate system and the image coordinate system; Step 1.1.2, constructing a camera imaging model, including a pinhole model and a CCD camera imaging model; Step 1.1.3, based on the conversion of the coordinate system space in step 1.1.1 and the camera imaging model in step 1.1.2, establish a mapping relationship from the coordinates of the object space point in the camera coordinate system to the coordinates of the corresponding image point in the image pixel coordinate system; Step 1.1.4, establish the transformation relationship from the body coordinate system to the camera coordinate system, and obtain the position and attitude of the aircraft in the world coordinate system based on the image information; Step 1.1.5, for the AprilTag type pose calibration code, after detecting the square edge of the code, decode it according to the custom rules; during the decoding process, it is allowed to block part of the code content, and codes of different sizes can be nested in the blank area; Step 1.1.6, for the SIFT feature matching class code, filter the feature points after matching and calculate the affine transformation matrix; Step 1.1.7, derive the rotation angle of each axis of XYZ according to the affine transformation matrix, and correct the transformed image to obtain the pixel distance ratio; Step 1.1.8, the rotation angle can be solved according to the homography transformation matrix to obtain the aircraft camera attitude.
2. The method for autonomous precision landing of an unmanned aerial vehicle according to claim 1, characterized in that: The 1.2 includes the following steps: Step 1.2.1, collect the image blocks in the rectangular frame, and then construct a training sample set with the target object located at different positions in the rectangular frame through circular shift; Step 1.2.2, assign different labels according to the distance of the target object from the center of the image block; since the training sample set is obtained by cyclic shifting the image block in the first frame, the entire sample set is a circulant matrix. The elements of each column of the circulant matrix are obtained by cyclically shifting the elements of the previous column. The circulant matrix can be diagonalized by the Fourier matrix, so the linear regression training can be accelerated by calculation; Step 1.2.3, train a linear regression model based on the training sample set and its labels; Step 1.2.4, introduce the kernel function to nonlinearly map the sample input of linear regression to regress the nonlinear model.
3. The method for autonomous precision landing of an unmanned aerial vehicle according to claim 1, characterized in that: The step 2 is based on the control algorithm of PID control and ping-pong control, and adopts different control combinations to greatly improve the security and real-time performance of the system; the step 2 includes the following steps: Step 2.1, automatic disturbance rejection optimization of PID control algorithm; Step 2.2, ping-pong control system construction; Step 2.3, based on the anti-disturbance optimization of the PID control algorithm in step 2.1 and the construction of the ping-pong control system in step 2.2, build the overall framework of the PID-ping-pong control system.
4. The method for autonomous precision landing of a drone according to claim 3, characterized in that: The step 2.1 includes the following steps: Step 2.1.1, pre-process the given signal and arrange the transition process according to the tolerance of the controlled object; the transition process makes the error change asymptotically rather than suddenly, achieving the control goal of no overshoot and fast tracking, and solving the contradiction between the rapidity and overshoot of PID control; using the characteristic of tracking differentiator that can track the signal, a nonlinear tracking differentiator is used in the ADRC to arrange the transition process; Step 2.1.2, using a nonlinear differential tracker with very low noise amplification effect to extract the differential of the given signal, and using an extended observer to extract the differential of the feedback signal; Step 2.1.3, using the extended state observer to estimate the total disturbance acting on the system in real time and compensate for it; Zero static error can be achieved without introducing integral feedback, thus avoiding the side effects of integral feedback.
5. The method for autonomous precision landing of an unmanned aerial vehicle according to claim 1, characterized in that: In step 3, based on RTK data and GPS data, combined with the hardware delay elimination algorithm of integral compensation, and with sufficient knowledge of the camera hardware characteristics, a multi-data source Kalman filter and velocity integral compensation method are used; specifically, the following steps are included: Step 3.1, using the stored speed history, after integration, it becomes the position deviation, and the position deviation compensation and the visual positioning position are summed; the stored speed value at each sampling moment is multiplied by the sampling duration, and after the effective window length is determined by timestamp comparison, the product is summed to obtain the actual estimated value of the position deviation; Step 3.2: synchronize timestamps using C ++ The mutex in realizes data sharing between the integral estimation thread and the feedback result calculation thread. The timestamp variable in the structure can be used to easily obtain and compare the sampling moments. Step 3.3, overall design of UAV precision autonomous landing middleware; UAV precision autonomous landing middleware is a C++ class implementation of autonomous landing, which adopts a modular bottom-up design approach and implements code recognition thread, KCF tracking thread, integral estimation thread, position acquisition function and landing control function.
6. The method for autonomous precision landing of a drone according to claim 5, characterized in that: The step 3.3 includes the following steps: Step 3.3.1, initialize the class and image cache, and start the integral estimation thread, code recognition thread and KCF tracking thread; Step 3.3.2, obtain the drone speed according to the set sampling time and store it in the circular queue; Step 3.3.3, identify the AprilTag positioning code and store the identification result in the member variable of the class; Step 3.3.4, divide the ROI area using the code information stored in the member variable as the center point, and regularly update the ROI area range; Step 3.3.5, position correction, after correction, return to the PID-ping-pong control algorithm to calculate the control quantity; Step 3.3.6, segment according to the set value during class initialization, use PID control at high altitude and PID-ping-pong control with dead zone at low altitude.
7. The method for autonomous precision landing of an unmanned aerial vehicle according to claim 1, characterized in that: The step 4 includes the following steps: Step 4.1: Apply the control algorithm in step 2; Step 4.2: Obtain the feedback data of the drone at high altitude and low altitude in step 3; Step 4.3: Combine effective control theory and control parameters to control the UAV to land accurately.
