Four-rotor unmanned aerial vehicle seamless positioning method based on multi-source fusion

Abstract

The invention discloses a four-rotor unmanned aerial vehicle seamless positioning method based on multi-source fusion, and belongs to the technical field of unmanned aerial vehicle positioning. According to the method, a positioning area is divided into an indoor area and an outdoor area according to a GNSS position precision factor, and a VIO positioning mode is adopted indoors; a GNSS / VIO loose coupling positioning mode based on federated Kalman filtering is adopted outdoors, firstly, external parameters of a GNSS subsystem and a VIO subsystem are calibrated online in a sliding window in combination with a time sequence difference thought, real-time space alignment is achieved, and then fusion of positioning information is conducted through a federated Kalman filter. And meanwhile, a smooth switching scheme of the positioning system is designed, so that the system can accurately identify and smoothly transition when the positioning area is changed. According to the invention, the high-precision positioning capability of the quad-rotor unmanned aerial vehicle in a complex environment is enhanced, and a three-dimensional perception solution with high universality is provided for vertical application of the unmanned aerial vehicle in low-altitude economy.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle (UAV) positioning, and in particular relates to a seamless positioning method for a quad-rotor UAV based on multi-source fusion. Background Art

[0002] The booming development of the low-altitude economy has placed higher demands on the intelligence level of equipment. Quadrotors, with their vertical take-off and landing, low-altitude flight, and rapid response capabilities, have become the most flexible means of transport in the low-altitude economy, capable of covering scenarios that are difficult for traditional aircraft to reach. High-precision positioning, as a key basic capability for autonomous operation of drones, can provide them with accurate environmental information, helping them to achieve autonomous navigation, target identification, and mission execution in dynamic and complex environments. However, as the scope of application expands, the mission scenarios of drones are becoming increasingly complex, gradually transforming from a single indoor or outdoor environment to mission scenarios covering complex indoor and outdoor environments. Therefore, ensuring high-reliability, high-precision, real-time seamless indoor and outdoor positioning is of certain practical significance for the expansion of drones in the low-altitude economy.

[0003] Currently, multi-sensor fusion positioning technology is a hot topic of research and attention for many scholars. For GNSS / VIO fusion positioning, the system initialization process requires establishing an accurate conversion relationship between the local VIO frame and the global GNSS frame. This conversion will be different each time the system starts, making offline calibration impossible. In addition, when traveling indoors and outdoors, satellite signals may be suddenly lost or gradually recaptured. However, current research has not provided reliable solutions to the above problems. Therefore, it is urgent to overcome the difficulties of accurate initialization and smooth and seamless positioning technology of GNSS / VIO fusion positioning systems, enhance the adaptability of drones in complex environments, and provide a highly versatile three-dimensional perception solution for the application of drones in the low-altitude economy. Summary of the Invention

[0004] The present invention aims to address the shortcomings of the existing technology by providing a seamless positioning method for quadrotor drones based on multi-source fusion. This method accurately divides the positioning area according to the GNSS position precision factor and uses the corresponding positioning method. For outdoor positioning, an online extrinsic parameter calibration method based on sliding windows and temporal difference is designed to determine the convergence of the extrinsic parameters. Historical GNSS information from outdoor positioning is used to maintain accuracy during indoor positioning. This ensures smooth and seamless positioning of the drone in both indoor and outdoor environments.

[0005] In order to achieve the purpose of the present invention, the technical solution adopted by the present invention is as follows:

[0006] A seamless positioning method for quadrotor drones based on multi-source fusion, including indoor and outdoor positioning, uses the position precision factor of the GNSS receiver to determine whether the quadrotor drone is indoors or outdoors. When the quadrotor is indoors, it uses Vision Induction (VIO) for positioning, while when it is outdoors, it uses a GNSS / VIO loosely coupled method for positioning.

[0007] The indoor positioning methods are:

[0008] S11, moving the quadrotor drone to be positioned outdoors, obtaining the initial position P0 of the quadrotor drone to be positioned through a GNSS receiver, capturing an environmental image in real time through a binocular camera on the quadrotor drone to be positioned, obtaining acceleration and angular velocity information of the quadrotor drone to be positioned in real time through an IMU on the quadrotor drone to be positioned, and representing the acceleration and angular velocity information with IMU information; obtaining the longitude, latitude, and altitude information of the quadrotor drone to be positioned in real time through the GNSS receiver on the quadrotor drone to be positioned;

[0009] S12, obtaining the position P11 of the quadrotor drone to be positioned in the VIO world coordinate system in real time based on the environmental image information and IMU information obtained in real time in step S11;

[0010] S13, converting the longitude information, latitude information, and altitude information obtained in real time in step S11 into a position P12 in the ENU coordinate system with the initial position P0 obtained in step S11 as the origin;

[0011] S14, calculating the rotation matrix and translation vector from the VIO world coordinate system to the ENU coordinate system using the position P11 obtained in step S12 and the position P12 obtained in step S13, where the rotation matrix and translation vector are represented by extrinsic parameters;

[0012] S15, moving the quadrotor drone to be positioned back into the room, capturing an image of the environment using a binocular camera on the quadrotor drone to be positioned, and obtaining acceleration and angular velocity information of the quadrotor drone to be positioned using an IMU on the quadrotor drone to be positioned, where the acceleration and angular velocity information are represented by IMU information;

[0013] S16, according to the environmental image information and IMU information obtained in step S15, the position P13 of the quadrotor drone to be positioned in the VIO world coordinate system is obtained, and the position P13 of the quadrotor drone to be positioned in the VIO world coordinate system is converted to the position P14 in the ENU coordinate system through the external parameters obtained in step S14 to complete the indoor positioning.

[0014] The steps for outdoor positioning are:

[0015] S21, obtaining an initial position P0 of the quadrotor drone to be positioned through a GNSS receiver, capturing an environmental image in real time through a binocular camera on the quadrotor drone to be positioned, obtaining acceleration and angular velocity information of the quadrotor drone to be positioned in real time through an IMU on the quadrotor drone to be positioned, wherein the acceleration and angular velocity information are represented by IMU information; and obtaining longitude, latitude, and altitude information of the quadrotor drone to be positioned in real time through a GNSS receiver on the quadrotor drone to be positioned;

[0016] S22, obtaining the position P21 of the quadrotor drone to be positioned in the VIO world coordinate system based on the environmental image information and IMU information obtained in real time in step S21;

[0017] S23, converting the longitude information, latitude information, and altitude information obtained in step S21 into a position P22 in the ENU coordinate system with the initial position P0 obtained in step S21 as the origin;

[0018] S24, calculating the rotation matrix and translation vector from the VIO world coordinate system to the ENU coordinate system using the position P21 obtained in step S22 and the position P22 obtained in step S23, where the rotation matrix and translation vector are represented by extrinsic parameters;

[0019] S25, converting the position P21 of the quadrotor drone to be positioned in the VIO world coordinate system obtained in step S22 into the ENU coordinate system using the external reference obtained in S24 to obtain the position P23;

[0020] S26, using a federated Kalman filter to fuse the position P22 obtained in step S23 with the position P23 obtained in step S25 to obtain the position of the quadrotor drone to be located.

[0021] In order to make the external parameters have better convergence, the quadrotor drone should be given motion excitation in all directions when placed outdoors.

[0022] The method for determining whether the quadrotor drone to be located is indoors or outdoors by using the position precision factor of the GNSS receiver is as follows:

[0023] When the GNSS receiver's position precision factor is greater than the position precision factor critical value for T consecutive times, the quadcopter to be located is judged to be indoors, and T is not less than 10. Otherwise, the quadcopter to be located is judged to be outdoors. The precision factor critical value is the set value. The larger the precision factor, the lower the accuracy of GNSS positioning.

[0024] In step S13, the method of converting the longitude information λ, latitude information φ, and altitude information h obtained in step S11 into the position P12 in the ENU coordinate system is:

[0025] Assume the initial position is: longitude (λ0), latitude (φ0), altitude (h0);

[0026] E=(λ-λ0)×cosφ0×R e

[0027] N=(φ-φ0)×R n

[0028] U=h-h0

[0029] Among them, R e is the mean radius of the Earth, R n is the radius of curvature of the meridian at the reference latitude, which can be approximated as the mean radius of the Earth;

[0030] In S14, the method for calculating the external parameters from the VIO world coordinate system to the ENU coordinate system through the position P11 and the position P12 is:

[0031] Using the lower-frequency GNSS timestamp as a benchmark, linear interpolation is performed on the VIO timestamp to achieve time alignment.

[0032] Construct a sliding window, establish a least squares model with coordinate system alignment in the sliding window, and solve the least squares model to obtain the initial extrinsic parameters;

[0033] The improved temporal difference method is used to transform the optimization of the external parameters into the following state estimation problem:

[0034]

[0035] Among them, α is the learning rate, x n+1 is the current initial external parameter, and are the estimated values ​​of the external parameters at the current moment and the estimated values ​​of the external parameters at the previous moment respectively;

[0036] The least squares model is:

[0037]

[0038] Among them, {R EV ,T EV} is an external reference;

[0039] The method of converting the position P13 to the position P14 in the ENU coordinate system by using the external reference in S16 is:

[0040] First, determine whether the external parameter converges. If it converges successfully, record the convergence value R of the external parameter. in and T in ;

[0041] P14=Rin P13+T in

[0042] If it does not converge, then the average value is obtained by removing the outliers of the historical external parameters and obtaining R ave and T ave ;

[0043]

[0044]

[0045] Among them, t s is the time when the quadrotor drone to be positioned is moved back to the room, t k is the current moment, Δt is the time it takes to linearly transition the quadrotor drone to be positioned from outdoor to indoor; P10 is the position of the quadrotor drone to be positioned when it moves back to the indoor position;

[0046] In step S26, the method of fusing the position P22 and the position P23 by the sampled federated Kalman filter is:

[0047] The federated Kalman filter includes two sub-filters and a main filter; the two sub-filters are sub-filter a and sub-filter b;

[0048] Use sub-filter a to filter the noise at position P22;

[0049] Use sub-filter b to filter the noise at position P23;

[0050] The main filter is used to fuse the noise-filtered position P22 and the noise-filtered position P23;

[0051] Distribute the global optimization information of the main filter to the sub-filters;

[0052] The mathematical models of the two sub-filters are the same, and the state space equations of the sub-filters are:

[0053]

[0054] in, is the state vector at time k, is the state vector at time k-1, Z k is the observation vector, A is the state transfer matrix, H is the observation matrix, w k and v k All are Gaussian white noise, w k is called process noise, v k It is called measurement noise;

[0055]

[0056] in, is the position in the ENU coordinate system, Velocity in the ENU coordinate system;

[0057] The observation vector of sub-filter a is:

[0058]

[0059] Among them, V 22 Speed ​​at position P22;

[0060] The observation vector of sub-filter b is:

[0061]

[0062] Among them, V 23 Speed ​​at position P23;

[0063] The state transfer matrix A is:

[0064]

[0065] Where δt is the time interval between the outputs of sub-filter a and sub-filter b; I is the identity matrix;

[0066] Observation matrix H = [I 6×6 ]

[0067] The formula for fusion in the main filter is:

[0068]

[0069] in, is the error covariance matrix of sub-filter a at time k, is the error covariance matrix of sub-filter b at time k, is the state estimate of sub-filter a at time k, is the state estimate of sub-filter b at time k; k is the fusion result at moment k.

[0070] The allocation principle of the fused global information to the sub-filters is:

[0071]

[0072] in, is the error covariance matrix of the k+1 sub-filter a, is the error covariance matrix of the sub-filter b at time k+1, is the state estimate of the k+1 sub-filter a, is the state estimate of sub-filter b at time k+1, and are the information allocation coefficients of sub-filter a and sub-filter b, respectively, satisfying the following principles:

[0073]

[0074] Define the performance indicators of sub-filter a and sub-filter b and They are:

[0075]

[0076] The distribution coefficient is calculated as:

[0077]

[0078] It can be seen that when a certain sub-filter has a higher precision performance, the corresponding performance index shows a smaller value, and the information distribution is also smaller, which also means that the error covariance matrix of the sub-filter at the next moment is smaller, and its estimation result occupies a higher weight in the main filter.

[0079] Beneficial effects

[0080] In response to the limitations of single-sensor positioning in complex indoor and outdoor environments, as well as the complementary characteristics of vision, inertial, and GNSS, this paper designs a seamless indoor and outdoor positioning solution. By designing a switching solution, the smoothness of the transition between indoor and outdoor environments is increased. A quadrotor drone hardware platform is built, and the performance of the solution is verified through public datasets and actual flight tests. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is the test result of the GNSS / VIO outdoor fusion positioning solution on the KITTI_00 dataset;

[0082] Figure 2 (a) is a test trajectory diagram of simulated indoor and outdoor shuttle positioning on the handheld dataset;

[0083] Figure 2 (b) is the test error diagram of simulated indoor and outdoor shuttle positioning on the handheld dataset. DETAILED DESCRIPTION

[0084] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0085] Example 1

[0086] Figure 1The figure below shows the test results of an outdoor positioning solution on the 00 sequence of the KITTI dataset. The gray dashed line in the figure represents the ground truth trajectory, and the colored solid line represents the algorithm's estimated trajectory. The color of the solid line increases from cold to warm, indicating an increasing error between the estimated and true trajectory. The KITTI dataset was collected on urban and rural roads by a vehicle-mounted platform equipped with multiple sensors: two Point Flea2 global shutter grayscale cameras and two color cameras with a resolution of 1.4 megapixels; an OXTS RT3003 inertial navigation system, including a GNSS receiver and IMU; and a 64-line Velodyne HDL-64E lidar. The sequences selected from the KITTI dataset vary in complexity, and the degree of GNSS signal obstruction in each dataset also varies. The 00 sequence has a high scene complexity, including complex routes such as straight lines, turns, and roundabouts, making it more challenging during testing. Therefore, the 00 sequence in the KITTI dataset was selected for testing. The testing process was as follows:

[0087] (1) First, the GNSS position precision factor in the dataset is read. If the precision factor is less than the set critical value for 10 consecutive times, it is determined to be an outdoor scene.

[0088] (2) Read the first longitude, latitude, and altitude values ​​in the data set: 48.9827°, 8.39045°, and 116.396m, respectively. This is the origin, P0. Then, read the real-time longitude, latitude, and altitude values ​​and convert them to coordinate P22 in the ENU coordinate system.

[0089] (3) The P21 method for calculating the current position of the quadrotor drone in the VIO world coordinate system is as follows:

[0090] Step 1. Data preprocessing: For binocular images, the Shi-Tomasi corner detection algorithm is used to extract feature points, and LK sparse pyramid optical flow is used for feature tracking. For IMU information, a pre-integration operation is performed.

[0091] Step 2, initialization: Mainly performs the initial pose solution of pure vision and correction of gyroscope bias. The pure vision pose solution mainly includes two steps: PnP and triangulation. The gyroscope bias is corrected and then pre-integrated.

[0092] Step 3, Sliding Window Optimization: Create a sliding window of size 10. At time i, the system state quantity to be optimized within the sliding window is defined as follows:

[0093]

[0094]

[0095] where n represents the number of key frames in the sliding window, m is the number of feature points in the sliding window. x0, x1, x n represents the state information of the IMU at k = 0, k = 1, k = n. The state information of the IMU at k is denoted as x k , x k contains the position of the IMU at k in the VIO world coordinate system velocity attitude and accelerometer bias b a and gyroscope bias b g . λ0, λ1, λ m represent the inverse depth of the corresponding feature points. Then a least squares objective function of the back-end optimization is established, which is solved iteratively by the Levenberg-Marquardt (LM) algorithm to obtain the position P21 in the VIO world coordinate system.

[0096] (4) A sliding window with an external parameter calculation size of 20 is constructed, and a least squares model of external parameter calculation is constructed by P21 and P22 in the sliding window:

[0097]

[0098] where {R EV ,T EV} is the external parameter. The initial external parameter is solved by singular value decomposition method, and then the external parameter value is optimized by improved time difference method, and the optimized external parameter value is denoted as the external parameter estimation value. The specific method is:

[0099] First, the initial external parameter estimation values at k and k+1 are calculated by averaging:

[0100]

[0101] where x k is the initial external parameter at k, is the external parameter estimation value at k, x k+1 is the initial external parameter at k+1, is the external parameter estimation value at k+1, and x1 and x2 are the first and second initial external parameter values after starting positioning, respectively.

[0102] In order to improve the real-time performance of state estimation, the above two equations are combined and simplified as:

[0103]

[0104] That is, the estimated value at the current moment is corrected by the error between the measured value at the current moment and the estimated value at the previous moment. However, as time goes by, the coefficient before the error becomes smaller and smaller, that is, the weight of the correction value becomes smaller and smaller, which will reduce the accuracy of subsequent estimated values. Therefore, a constant α is used to replace 1 / (n+1), that is:

[0105]

[0106] Where α is the learning rate, which is set to 0.4.

[0107] (5) The position P21 of the quadrotor drone to be positioned in the VIO world coordinate system is converted to the ENU through the calculated external parameters to obtain the position P23.

[0108] (6) Use the federated Kalman filter to fuse P22 and P23.

[0109] The final fusion trajectory is as follows Figure 1 The EVO tool was used to evaluate the absolute trajectory error. Compared with the true value, the maximum error was 2.893507m and the root mean square error was 1.743103m.

[0110] It has been verified that the root mean square error of the fusion positioning algorithm is less than 0.3% D (D is the movement distance)

[0111] Example 2

[0112] A self-built drone was used to collect environmental data sets in an outdoor parking lot. To simulate indoor and outdoor travel, the GNSS precision factor was artificially increased to simulate entering indoor areas. This method can test whether the positioning system can achieve smooth and seamless positioning in complex indoor and outdoor scenarios. Figure 2 (a) is a test trajectory diagram of indoor and outdoor shuttle positioning on a handheld dataset. Figure 2 (b) is the test error curve of indoor and outdoor shuttle positioning on the handheld dataset. The GNSS precision factor is artificially increased to simulate entering the room within 45s-100s.

[0113] First, the quadrotor drone is in an outdoor scene, and the initial longitude, latitude, and altitude information obtained by the GNSS receiver are: 39.9635°, 116.305°, and 46.4224m, respectively. P21 and P22 are obtained according to steps S21 to S23 of outdoor positioning.

[0114] The rotation matrix and translation vector from the VIO world coordinate system to the ENU coordinate system are calculated in real time in the sliding window through P21 and P22. The external parameters converge at t = 28s, and the convergence value is:

[0115]

[0116] T in =[0.31201, 0.951531, 0.13323]

[0117] At t=45s, the precision factor obtained by the GNSS receiver begins to exceed the critical value. At t=46s, the precision factor exceeds the critical value for 10 consecutive times, and it is determined that the current positioning area begins to enter the indoor area.

[0118] After entering the indoor area, the VIO positioning method is used to solve the current position P13 of the quadrotor drone in the VIO world coordinate system, and then P13 is calculated by R in and T in Converted to the ENU coordinate system, we get position P14, namely:

[0119] P14=R in P13+T in

[0120] At t = 100, the DOP obtained by the GNSS receiver begins to fall below the critical value. At t = 101s, the DOP falls below the critical value 10 times in a row, indicating that the current positioning area has entered the outdoor area. Outdoor positioning is continued to obtain the position of the quadrotor in the ENU coordinate system in the outdoor scene.

[0121] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A seamless positioning method for a quadrotor drone based on multi-source fusion, characterized by: This positioning method includes indoor positioning and outdoor positioning. The position precision factor of the GNSS receiver is used to determine whether the quadrotor drone to be positioned is indoors or outdoors. When it is indoors, VIO is used for positioning, and when it is outdoors, the GNSS / VIO loosely coupled method is used for positioning.

2. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 1, characterized in that: Indoor positioning methods are: S11, moving the quadrotor drone to be positioned outdoors, obtaining the initial position P0 of the quadrotor drone to be positioned through a GNSS receiver, capturing an environmental image in real time through a binocular camera on the quadrotor drone to be positioned, obtaining acceleration and angular velocity information of the quadrotor drone to be positioned in real time through an IMU on the quadrotor drone to be positioned, and representing the acceleration and angular velocity information with IMU information; obtaining the longitude, latitude, and altitude information of the quadrotor drone to be positioned in real time through the GNSS receiver on the quadrotor drone to be positioned; S12, obtaining the position P11 of the quadrotor drone to be positioned in the VIO world coordinate system in real time based on the environmental image information and IMU information obtained in real time in step S11; S13, converting the longitude information, latitude information, and altitude information obtained in real time in step S11 into a position P12 in the ENU coordinate system with the initial position P0 obtained in step S11 as the origin; S14, calculating the rotation matrix and translation vector from the VIO world coordinate system to the ENU coordinate system using the position P11 obtained in step S12 and the position P12 obtained in step S13, where the rotation matrix and translation vector are represented by extrinsic parameters; S15, moving the quadrotor drone to be positioned back into the room, capturing an image of the environment using a binocular camera on the quadrotor drone to be positioned, and obtaining acceleration and angular velocity information of the quadrotor drone to be positioned using an IMU on the quadrotor drone to be positioned, where the acceleration and angular velocity information are represented by IMU information; S16, according to the environmental image information and IMU information obtained in step S15, the position P13 of the quadrotor drone to be positioned in the VIO world coordinate system is obtained, and the position P13 of the quadrotor drone to be positioned in the VIO world coordinate system is converted to the position P14 in the ENU coordinate system through the external parameters obtained in step S14 to complete the indoor positioning.

3. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 1, characterized in that: The steps for outdoor positioning are: S21, obtaining an initial position P0 of the quadrotor drone to be positioned through a GNSS receiver, capturing an environmental image in real time through a binocular camera on the quadrotor drone to be positioned, obtaining acceleration and angular velocity information of the quadrotor drone to be positioned in real time through an IMU on the quadrotor drone to be positioned, wherein the acceleration and angular velocity information are represented by IMU information; and obtaining longitude, latitude, and altitude information of the quadrotor drone to be positioned in real time through a GNSS receiver on the quadrotor drone to be positioned; S22, obtaining the position P21 of the quadrotor drone to be positioned in the VIO world coordinate system based on the environmental image information and IMU information obtained in real time in step S21; S23, converting the longitude information, latitude information, and altitude information obtained in step S21 into a position P22 in the ENU coordinate system with the initial position P0 obtained in step S21 as the origin; S24, calculating the rotation matrix and translation vector from the VIO world coordinate system to the ENU coordinate system using the position P21 obtained in step S22 and the position P22 obtained in step S23, where the rotation matrix and translation vector are represented by extrinsic parameters; S25, converting the position P21 of the quadrotor drone to be positioned in the VIO world coordinate system obtained in step S22 into the ENU coordinate system using the external reference obtained in S24 to obtain the position P23; S26, using a federated Kalman filter to fuse the position P22 obtained in step S23 with the position P23 obtained in step S25 to obtain the position of the quadrotor drone to be located.

4. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 1, characterized in that: The method for determining whether the quadrotor drone to be located is indoors or outdoors by using the position precision factor of the GNSS receiver is as follows: When the position precision factor of the GNSS receiver is greater than the position precision factor critical value for T consecutive times, the quadrotor drone to be located is judged to be indoors, and T is not less than 10. Otherwise, the quadrotor drone to be located is judged to be outdoors. The precision factor critical value is a set value. The larger the precision factor, the lower the accuracy of GNSS positioning.

5. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 2, characterized in that: In step S13, the method of converting the longitude information λ, latitude information φ, and altitude information h obtained in step S11 into the position P12 in the ENU coordinate system is: Assume the initial position is: longitude λ0, latitude φ0, altitude h0; E =(λ-λ0)×cosφ0×R e N=(φ-φ0)×R n U=h-h0 Among them, R e is the mean radius of the Earth, R n is the radius of curvature of the meridian at the reference latitude.

6. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 2, characterized in that: In S14, the method for calculating the external parameters from the VIO world coordinate system to the ENU coordinate system through the position P11 and the position P12 is: Using the lower-frequency GNSS timestamp as a benchmark, linear interpolation is performed on the VIO timestamp to achieve time alignment. Construct a sliding window, establish a least squares model with coordinate system alignment in the sliding window, and solve the least squares model to obtain the initial extrinsic parameters; The improved temporal difference method is used to transform the optimization of the external parameters into the following state estimation problem: Among them, α is the learning rate, x n+1 is the current initial external parameter, and They are the estimated values ​​of the external parameters at the current moment and the estimated values ​​of the external parameters at the previous moment respectively.

7. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 6, characterized in that: The least squares model is: Among them, {R EV ,T EV } is an external parameter, and N is the sliding window size, which is set to 20.

8. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 2, characterized in that: The method of converting the position P13 to the position P14 in the ENU coordinate system by using the external reference in S16 is: First, determine whether the external parameter converges. If it converges successfully, record the convergence value R of the external parameter. in and T in ; P14=R in P13+T in If it does not converge, then the average value is obtained by removing the outliers of the historical external parameters and obtaining R ave and T ave ; Among them, t s is the time when the quadrotor drone to be positioned is moved back to the room, t k is the current moment, Δt is the time it takes to linearly transition the quadrotor drone to be positioned from outdoor to indoor; P10 is the position of the quadrotor drone to be positioned when it moves back to the indoor space.

9. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 3, characterized in that: In step S26, the method of fusing the position P22 and the position P23 by the sampled federated Kalman filter is: The federated Kalman filter includes two sub-filters and a main filter; the two sub-filters are sub-filter a and sub-filter b; Use sub-filter a to filter the noise at position P22; Use sub-filter b to filter the noise at position P23; The main filter is used to fuse the noise-filtered position P22 and the noise-filtered position P23; Distribute the global optimization information of the main filter to the sub-filters.

10. The method for seamless positioning of a quadrotor drone based on multi-source fusion according to claim 9, characterized in that: The mathematical models of the two sub-filters are the same, and the state space equations of the sub-filters are: in, is the state vector at time k, is the state vector at time k-1, Z k is the observation vector, A is the state transfer matrix, H is the observation matrix, w k and v k All are Gaussian white noise, w k is called process noise, v k It is called measurement noise; in, is the position in the ENU coordinate system, Velocity in the ENU coordinate system; The observation vector of sub-filter a is: Among them, V 22 Speed ​​at position P22; The observation vector of sub-filter b is: Among them, V 23 Speed ​​at position P23; The state transfer matrix A is: Where δt is the time interval between the outputs of sub-filter a and sub-filter b; I 3×3 is the identity matrix of size 3×3; Observation matrix H = [I 6×6 ]; The formula for fusion in the main filter is: in, is the error covariance matrix of sub-filter a at time k, is the error covariance matrix of sub-filter b at time k, is the state estimate of sub-filter a at time k, is the state estimate of sub-filter b at time k; k is the fusion result at the moment; The allocation principle of the fused global information to the sub-filters is: in, is the error covariance matrix of sub-filter a at time k+1, is the error covariance matrix of the sub-filter b at time k+1, is the state estimate of the k+1 sub-filter a, is the state estimate of the sub-filter b at time k+1, and are the information allocation coefficients of sub-filter a and sub-filter b, respectively, satisfying the following principles: Define the performance indicators of sub-filter a and sub-filter b and They are: The distribution coefficient is calculated as:

Images