Multi-scanning laser radar collaborative calibration method and system based on three-dimensional space grid

Abstract

The invention discloses a multi-scanning laser radar collaborative calibration method and system based on a three-dimensional space grid, relates to the field of laser radars, and is used for improving the high-precision collaborative consistency of spatial orientation of multiple radars. According to the method, a multi-radar relative coordinate system is constructed and initialized; constructing a three-dimensional calibration space in each radar detection intersection area, and dividing a space node set; planning a space moving path of the captive balloon in the space node set, and synchronizing system time; executing space node traversal and observation data acquisition according to the planned space moving path; based on the constructed multi-radar relative coordinate system and the collected observation data, constructing a global nonlinear error model and performing joint optimization; and performing system parameter compensation and correction according to the optimized calibration parameters. The method effectively solves the problems that a traditional plane calibration method cannot decouple radar installation inclination errors and pointing zero deviation, and the consistency of multi-radar beam long-distance space rendezvous is poor.

Description

Technical Field

[0001] This invention relates to the field of lidar technology, and in particular to a collaborative calibration method and system for multi-scan lidar based on a three-dimensional spatial grid. Background Technology

[0002] Scanning laser wind radar is widely used in wind field detection. It often requires multiple scanning radars to work together to obtain high-precision wind speed vectors, so beam pointing accuracy is crucial.

[0003] Known radar calibration methods mostly utilize fixed hard targets or solar trajectories. While UAV-assisted calibration has become a trend in recent years, it is largely based on single-point or single-altitude planar flight to calibrate radars. This two-dimensional sampling method cannot effectively decouple the radar's installation tilt error and pointing null error geometrically. Furthermore, known calibration techniques mostly calibrate individual radars independently, ignoring the spatial geometric constraints of the network system. This can lead to poor beam convergence consistency at long distances, data aliasing, and erroneous data.

[0004] Therefore, there is an urgent need for a calibration method that can decouple multidimensional errors using a three-dimensional grid and improve the accuracy of multi-radar coordination. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-scan lidar collaborative calibration method and system based on a three-dimensional spatial grid, addressing all or part of the aforementioned problems, in order to solve the difficulty of decoupling attitude tilt and zero-position deviation, and improve the high-precision collaborative consistency of spatial pointing of multiple radars.

[0006] The technical solution adopted in this invention is as follows: A collaborative calibration method for multi-scan lidar based on a three-dimensional spatial grid includes the following steps: S1. Construct and initialize the multi-radar relative coordinate system; S2. Construct a three-dimensional calibration space in the intersection area of ​​each radar detection, and divide the spatial node set; S3. Plan the spatial movement path of the tethered balloon at the set of spatial nodes and synchronize the system time; S4. Perform spatial node traversal and observation data collection according to the planned spatial movement path; S5. Based on the constructed multi-radar relative coordinate system and the collected observation data, a global nonlinear error model is constructed and jointly optimized; the global nonlinear error model characterizes the difference between the theoretical observation value and the actual measurement value of the tethered balloon at each space node; S6. Perform system parameter compensation and calibration based on the optimized calibration parameters.

[0007] Furthermore, step S5 includes: S51. Define the error state vector to be solved for each radar; S52. Construct a forward observation physical model under the error state vector; S53. Construct the residual objective function based on the aforementioned forward observation physical model; S54. Optimize the objective function with the goal of minimizing it, and obtain the calibration parameters of each radar. S55. Estimate the accuracy of the calibration parameters.

[0008] Furthermore, the error state vector includes azimuth zero deviation, pitch zero deviation, roll angle error, and pitch angle error.

[0009] Furthermore, a forward observation physical model under the error state vector is constructed, including: Based on the principle of rigid body transformation, a mapping function from the spatial coordinates of the tethered balloon to the theoretical observation angle of the radar is established under the error state vector.

[0010] Furthermore, based on the principle of rigid body transformation, a mapping function from the spatial coordinates of the tethered balloon to the theoretical radar observation angle is established under the aforementioned error state vector, including: Calculate the local relative vector between the tethered balloon and the radar; Perform an inverse coordinate system rotation transformation on the relative vector; The transformed relative vector is projected onto spherical coordinates and superimposed with the deviation.

[0011] Furthermore, the objective function represents the set of residuals between the theoretical radar observations and the actual measurements of the tethered balloon at all spatial grid points under the error state vector.

[0012] Furthermore, estimating the accuracy of the calibration parameters includes: Calculate the root mean square error between the theoretical radar observations and the actual measurements from the tethered balloon under the calibration parameters; If the root mean square error is less than the preset convergence threshold, the corresponding radar calibration is determined to be successful; otherwise, the calibration parameters of the corresponding radar are re-optimized.

[0013] Furthermore, system parameter compensation and calibration are performed based on the optimized calibration parameters, including: The calibration parameters are written into the parameter configuration file or underlying control algorithm of the multi-radar system so that real-time reverse compensation correction can be performed based on the calibration parameters in subsequent wind field collaborative observation tasks. The reverse compensation correction includes: when the radar main control system receives an observation command pointing to the coordinates of a predetermined spatial target, the control algorithm first performs coordinate pre-rotation on the relative vector based on the calibrated tilt angle error, and superimposes the calibrated zero position deviation to calculate the corrected encoder angle command that the servo motor actually needs to execute.

[0014] Furthermore, the partitioned spatial node set includes: The three-dimensional calibration space is discretized according to the equidistant partitioning rule to construct a structure containing... A set of spatial nodes, n This indicates the number of spatial nodes in each dimension of the three-dimensional calibration space; The height layer span, which is divided vertically in the three-dimensional calibration space, covers the radar observation height.

[0015] The present invention also provides a multi-scan lidar collaborative calibration system based on a three-dimensional spatial grid, which includes a processor and a storage medium; the storage medium stores computer instructions, and the processor executes the computer instructions to perform the above-described multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This invention addresses multi-radar systems by constructing and initializing a multi-radar relative coordinate system to unify the coordinate system; constructing a three-dimensional calibration space in the intersection area of ​​each radar detection, and dividing it into spatial node sets; planning the spatial movement path of a tethered balloon within the spatial node sets, and synchronizing system time; performing spatial node traversal and observation data acquisition according to the planned spatial movement path; constructing a global nonlinear error model based on the constructed multi-radar relative coordinate system and the acquired observation data, and jointly optimizing it to obtain the calibration parameters of each radar; finally, compensating and correcting system parameters based on the optimized calibration parameters. This invention effectively solves the problems of traditional planar calibration methods failing to decouple radar installation tilt errors (roll / pitch angles) and pointing zero-position deviations (azimuth / pitch), as well as the poor consistency of long-distance spatial intersection of multi-radar beams. It is highly suitable for applications such as high-precision three-dimensional wind field inversion, complex terrain wind resource assessment, and multi-radar network collaborative observation. Attached Figure Description

[0017] The present invention will be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of a multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid.

[0018] Figure 2 This is a flowchart illustrating the implementation process of constructing and jointly optimizing a global nonlinear error model.

[0019] Figure 3 This is a schematic diagram of the spatial geometry of lidar networking and three-dimensional calibration space grid division.

[0020] Figure 4 It is a structured diagram of the observation data set.

[0021] Figure 5 This is a comparison chart of radar observation residual distribution before and after joint optimization using a global nonlinear error model. Detailed Implementation

[0022] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0023] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0024] This application proposes a collaborative calibration method for multi-scan lidar based on a three-dimensional spatial grid, such as... Figure 1 As shown, the method includes the following steps: S1. Construct and initialize the multi-radar relative coordinate system.

[0025] Multiple scanning laser wind radars are deployed in the area to be detected. For each of the multiple radars, one radar is used as a reference to determine the three-dimensional position information of the other radars relative to that radar, and a radar network position set is constructed.

[0026] For example, suppose three radars, denoted as Radar A, Radar B, and Radar C, are deployed in the area to be detected. The geometric center of Radar A is selected as the origin of the global coordinate system, establishing a local Cartesian coordinate system. High-precision positioning equipment (such as D-RTK GNSS) is used to determine the three-dimensional position information of Radar B and Radar C relative to Radar A. In specific measurements, the D-RTK base station is set up at Radar A, and the mobile stations are placed at the geometric centers of Radar B and Radar C, respectively, to obtain the centimeter-level relative coordinates of Radar B and Radar C relative to Radar A.

[0027] The final radar network location set is as follows: Equation (1): ; in, This represents the set of radar network locations for a multi-radar system. Indicates the coordinates of radar A, which serves as the reference station, and and Let T represent the relative three-dimensional coordinates of radar B and radar C in a coordinate system with radar A as the origin, and let T denote matrix transpose. For ease of explanation, the following description will use three radars as an example to illustrate the technical solution. Those skilled in the art should understand that the solution in this application is also applicable to other multi-radar systems.

[0028] For the constructed radar network location set In one alternative implementation, the coordinates are stored in a structured form to complete the initialization of the relative coordinate system.

[0029] For example, the radar network location set The data is formatted and stored as a standard CSV file (e.g., named radar_positions_data.CSV). The file contains four columns of data: radar ID, x-coordinate, y-coordinate, and z-coordinate, which serve as static configuration parameters for the initialization of the multi-radar system.

[0030] S2. Construct a three-dimensional calibration space in the intersection area of ​​each radar detection and divide the spatial node set.

[0031] Based on the radar network location set obtained in step S1 Based on the effective detection range of each radar, the common intersection airspace of the three radar beams is determined. A virtual three-dimensional calibration space is then established within this common intersection airspace. The established three-dimensional calibration space is discretized according to an equal-interval (each dimension is equally spaced, meaning the step size is equal for each individual dimension, but not necessarily for different dimensions; of course, the step size can be equal for all dimensions) partitioning rule, constructing a system containing... A three-dimensional mesh set of spatial nodes (i.e., grids).

[0032] Equation (2): ; In the formula, Represents a three-dimensional mesh set; Represents the set of elements. line, number Column, No. The three-dimensional spatial coordinates of the layer space node. These are the three-dimensional coordinates of the spatial node in the global coordinate system. This indicates the number of spatial nodes in each of the three dimensions (orthogonal X-axis, Y-axis, and Z-axis) of the three-dimensional calibration space.

[0033] In specific implementation, as a preferred embodiment, the above-mentioned division of the three-dimensional calibration space... n The value should be no less than 3 to ensure sufficient geometric constraints are formed in space. For example, setting... That is, to build A three-dimensional grid; and the height layer span of this three-dimensional grid in the vertical direction (Z-axis) needs to cover the radar observation height (e.g., 100m to 500m) in order to fully decouple the pitch and roll tilt errors of the radar installation.

[0034] In addition, to ensure that the observation angles of the three radars on the spatial nodes are evenly distributed, in one optional implementation, the center position of the set three-dimensional calibration space is selected directly above the center of the geometric topology structure formed by the three radars.

[0035] S3. Plan the spatial movement path of the tethered balloon at the spatial node set and synchronize the system time.

[0036] The three-dimensional mesh set constructed based on step S2 The plan outlines the spatial movement trajectory of a tethered balloon traversing each spatial node, setting the dwell time at each node to ensure sequential traversal. For observing the tethered balloon's spatial movement trajectory, a tethered balloon equipped with an optical corner reflector and a real-time dynamic differential positioning (D-RTK) rover can be used as a calibration target. The D-RTK base station is positioned at the origin of radar A to ensure that the positioning coordinates fed back by the target are strictly consistent with the local coordinate system of the multi-radar system.

[0037] Before executing the mobile mission, the target positioning system and the data acquisition systems of the three radars are synchronized using a unified time synchronization service to ensure that the time deviation of each system meets the following constraints: Equation (3): ; In the formula, The timestamp represents the true location of the tethered balloon. Indicates the i-th radar The timestamp of the collected observation data (i.e., target location). This indicates the maximum allowable time synchronization error threshold for the system.

[0038] In an optional implementation, the time synchronization method in step S3 employs Network Time Protocol (NTP) or GPS Pulse Per Second (PPS) hardware-triggered synchronization to ensure... Better than 10 milliseconds (or other values ​​within 10 milliseconds) to eliminate position-observation asynchrony errors caused by target movement in a non-stationary state.

[0039] For the planning of the target's spatial movement trajectory, as a preferred implementation method, the spatial movement trajectory planning adopts an "S"-shaped layered traversal strategy or a spiral ascending traversal strategy. This involves controlling the tethered balloon from the bottom spatial node through ground vehicle traction and cable deployment / retraction, traversing all nodes layer by layer upwards. n × n × n Each space node has a preset dwell time (e.g., 10s to 20s) so that the multi-radar system has enough time to perform micro-scans and capture stable corner reflector echo peaks.

[0040] S4. Perform spatial node traversal and observation data collection according to the planned spatial movement path.

[0041] The tethered balloons are controlled to sequentially reach the three-dimensional grid set according to the planned spatial movement path. The system maintains a fixed position at each spatial node. During this period, three lidar units (radar A, radar B, and radar C) are controlled to perform a coordinated micro-scan of the corner reflector position on the tethered balloon, locking onto the beam direction with the strongest echo intensity. The system simultaneously acquires observation data, namely the true spatial coordinates of the tethered balloon relative to radar A (i.e., the origin of the coordinate system) and the observation angles of the three radars, constructing a calibration observation dataset containing nine data dimensions.

[0042] Equation (4): ; In the formula, This represents the calibration observation dataset; This represents the total number of sampling points, i.e., the total number of spatial nodes. m For sampling point index; In the first m The three-dimensional coordinates of the tethered balloon relative to radar A at each sampling point; For the first i Taiwan Radar In the m Azimuth angles measured at each sampling point With pitch angle The data at each sampling point is aligned using synchronized timestamps.

[0043] Similarly, the calibration observation dataset can be... Stored in a structured format for easy management and computation. For example, calibration observation datasets... The data is formatted and stored as a standard CSV file (filename for example: calibration_flight_data.CSV), which strictly contains the following 9 columns of data (i.e., 9 attributes): target_x: The X-axis coordinate of the tethered balloon relative to radar A; target_y: The Y-axis coordinate of the tethered balloon relative to radar A; target_z: The Z-axis coordinate of the tethered balloon relative to radar A; Radar_A_az: The observation azimuth angle of radar A; Radar_A_el: The observation elevation angle of radar A; Radar_B_az: The observation azimuth angle of radar B; Radar_B_el: The observation elevation angle of radar B; Radar_C_az: The observation azimuth angle of radar C; Radar_C_el: The observation elevation angle of radar C.

[0044] The first three columns of data serve as the spatial reference (true value) for geometric calibration, while the last six columns serve as the observations to be calibrated, together forming the input source for joint calibration.

[0045] S5. Based on the constructed multi-radar relative coordinate system and the collected observation data, a global nonlinear error model is constructed and jointly optimized.

[0046] To set the location of the radar network separately and calibration observation dataset Taking the standardization process as an example, the radar network location set generated in step S1 is read. The file (radar_positions_data.CSV) and the calibration observation dataset generated in step S4 The file (calibration_flight_data.CSV) contains data for each radar in the multi-radar system. i A geometric calibration model (i.e., the aforementioned global nonlinear error model) is constructed and solved based on the nonlinear least squares method. This global nonlinear error model characterizes the difference between the theoretical observation and actual measurement values ​​of the tethered balloon at each space node. The results of the joint optimization are the calibration parameters of each radar, including null position deviation and tilt angle error.

[0047] Specifically, such as Figure 2 As shown, step S5 includes the following sub-steps: S51. Define the error state vector to be solved for each radar.

[0048] For any radar Define the error state vector to be optimized. This includes zero-position deviation and tilt angle error: Equation (5): ; In the formula, i For radar indexing, For radar i azimuth zero position deviation, For radar i pitch angle zero position deviation, For radar i The roll angle error, For radar i The pitch angle error. Initialize this difference state vector. It is a zero vector or an empirical value.

[0049] S52. Construct a forward observation physical model under the error state vector.

[0050] Based on the principle of rigid body transformation, and established on the error state vector Below is the mapping function from "tethered balloon spatial coordinates" to "radar theoretical observation angle". ,in This indicates the coordinates of the tethered balloon's probe.

[0051] Specifically, for each sampling point m : (1) Calculate the local relative vector between the tethered balloon and the radar.

[0052] Calculate the coordinates of the tethered balloon D-RTK probe radar coordinates The difference is used to obtain the relative vector in the world coordinate system. .

[0053] (2) Perform a coordinate system inverse rotation transformation on the relative vector.

[0054] This transformation operation is performed based on tilt angle error. An Euler angle rotation matrix is ​​introduced. Transform the relative vector in the world coordinate system into a vector in the sensor coordinate system with installation tilt. : Equation (6): ; vector The three-dimensional coordinate form is .

[0055] (3) Project the transformed relative vector onto spherical coordinates and superimpose the deviation.

[0056] The superimposed deviation is the zero-position deviation. The vector Convert to theoretical distance, azimuth, and elevation angles in spherical coordinates, and superimpose zero offset. difference Theoretical observation values ​​were obtained. : Equation (7):

[0057] Indicates the first m The theoretically observed azimuth angle of each sampling point. Indicates the first m The theoretically observed pitch angle at each sampling point.

[0058] In practice, arctan2 (usually denoted as ata2, a two-parameter arctangent function used to calculate the coordinates of a point relative to its azimuth starting axis) needs to be adjusted according to the radar's defined azimuth starting axis. x The parameter order and normalization logic (including the angle of the axis and automatic quadrant determination) are determined.

[0059] S53. Construct the residual objective function based on the forward observation physical model.

[0060] Construct the objective function Its definition is all The set of residuals between the "theoretical radar observations" and the "actual measurements of the tethered balloon" at each sampling point (i.e., the spatial grid) under the corresponding error state vector: Equation (8): ; In the formula, and For the first Radar at each sampling point i The actual measured values ​​(azimuth and elevation angles); and For the first m Radar at each sampling point i The theoretical observations (azimuth and elevation angles), i.e., substituted into The theoretical azimuth and elevation angles are calculated subsequently. When calculating the azimuth residual, it is necessary to perform... Periodic processing is used to eliminate the effects of phase jumps, and the pitch angle will not exceed this range, so no conversion is required.

[0061] S54. Optimize the objective function to minimize it, and obtain the calibration parameters for each radar.

[0062] The Levenberg-Marquardt (LM) algorithm is used to evaluate the objective function. Iterative optimization is performed by continuously adjusting the error parameter vector. This causes the sum of squared residuals to converge to a minimum. When the iteration satisfies the convergence condition (e.g., the change in residuals is less than...), the iteration continues. Output the final optimized solution by determining the time (or maximum number of iterations). That is, the radar. i calibration parameters : Equation (9): .

[0063] These represent the radars respectively. i The calibrated azimuth zero deviation, pitch zero deviation, roll angle error, and pitch angle error.

[0064] S55. Estimate the accuracy of calibration parameters.

[0065] Using the calculated optimal calibration parameters Substitute these values ​​back into the forward observation physical model and recalculate the theoretical observations for all sampling points. Define the root mean square error (RMSE) as a quantitative indicator for evaluating calibration quality; its calculation formula is as follows: Equation (10): .

[0066] If the calculated result The root mean square error of the i-th radar is less than a preset convergence threshold (preferably...). (It can also be other values), then the radar is determined to be... i Calibration successful, output final calibration parameters. Otherwise, the calibration is deemed divergent, requiring radar recalibration. The system may prompt the user to check the coordinate system definition of the input data or to reacquire observation data. The calibration process is then repeated until successful.

[0067] S6. Perform system parameter compensation and calibration based on the optimized calibration parameters.

[0068] The optimal calibration parameters of each radar output in step S5 (Including azimuth zero deviation) Pitch angle zero position deviation Roll angle error and pitch angle error This calibration parameter is written into the parameter configuration file or underlying control algorithm of the multi-radar system. In subsequent wind field collaborative observation tasks, the multi-radar system performs real-time reverse compensation correction based on this calibration parameter.

[0069] Specifically, in one optional implementation, the reverse compensation correction includes active beam pointing compensation: when the radar main control system receives the coordinates pointing to a predetermined spatial target... When receiving observation commands, the control algorithm first uses the calibrated tilt angle error, derived from the tilt angle error matrix. right The vector is pre-rotated in coordinates and the calibrated zero-position deviation is superimposed. and The corrected encoder angle command that the servo motor actually needs to execute is calculated, thereby eliminating the pointing offset caused by the installation posture and zero position error at the physical level, and ensuring that multiple radar beams can maintain micro-arc level spatial convergence consistency outside the long-distance calibration airspace.

[0070] This application effectively solves the problems of traditional planar calibration methods failing to decouple radar installation tilt error (roll / pitch angle) and pointing zero deviation (azimuth / pitch), as well as the poor consistency of long-distance spatial intersection of multiple radar beams, by constructing a three-dimensional mesh for tethered balloon traversal and a global nonlinear error model joint optimization technique. It is very suitable for application scenarios such as high-precision three-dimensional wind field inversion, wind resource assessment in complex terrain, and multi-radar network collaborative observation.

[0071] The following examples illustrate the collaborative calibration method for multi-scan lidar based on three-dimensional spatial meshes proposed in this application: S1. Construct and initialize the multi-radar relative coordinate system.

[0072] Three scanning laser wind-measuring radars were deployed in the area to be detected, labeled as radar A, radar B, and radar C, respectively. The geometric center of radar A was selected as the origin of the global coordinate system, and a local Cartesian coordinate system was established. The three-dimensional position information of radar B and radar C relative to radar A was determined using high-precision positioning equipment such as D-RTKGNSS, and the radar network position set was constructed as shown in equation (1).

[0073] like Figure 3 The diagram shows the spatial geometry of a multi-lidar network and the 3D calibration trajectory of a tethered balloon. The measured deployment coordinates of the three radars are as follows: Radar A: Base station; Radar B: It is located approximately 3.7 km east-southeast of radar A; Radar C: It is located approximately 3.6 km northeast of radar A.

[0074] The coordinate data mentioned above is formatted and stored as a standard CSV file named radar_positions_data.CSV, which serves as the static spatial reference parameters for the system's calculation.

[0075] S2. Construct a three-dimensional calibration space in the intersection area of ​​each radar detection and divide the spatial node set.

[0076] Based on the radar network location set obtained in step S1 Based on the effective detection range of each radar, the common intersection airspace of the three radar beams is determined. Within this intersection airspace, a virtual three-dimensional calibration space is established and discretized according to an equal-interval division rule, constructing a system containing... A three-dimensional mesh set with a total of M=125 spatial nodes. ,like Figure 3 As shown by the purple scatter dots, the center of this three-dimensional mesh (i.e., the three-dimensional calibration space) is located above the geometric center of the three radars, and horizontally covers the area surrounding the center point. The area covers vertical height layers from 100m to 500m (with a 100m spacing between layers). This multi-height layer span design provides sufficient vertical spatial geometric constraints for subsequent calculations, effectively decoupling the pitch and roll tilt errors of the radar installation.

[0077] S3. Plan the spatial movement path of the tethered balloon at the spatial node set and synchronize the system time.

[0078] The 3D mesh set constructed based on step S2 The spatial movement trajectory of the tethered balloon is planned so that it can traverse each spatial node (i.e., sampling point) in sequence, including but not limited to traversing in an "S"-shaped spatial movement trajectory, and setting the stay at each spatial node for 10s-20s. The tethered balloon equipped with an optical corner reflector and a real-time dynamic differential positioning (D-RTK) mobile station is selected as the calibration target, and the D-RTK base station is set at the coordinate origin of radar A to ensure that the positioning coordinates fed back by the target are strictly consistent with the local coordinate system of the radar system. Before the target performs the movement task, the target positioning system and the data acquisition system of the three radars are synchronized through unified time service such as NTP protocol to ensure that the time deviation of each system meets the constraint of time (3), in which, Set as .

[0079] S4. Perform spatial node traversal and observation data collection according to the planned spatial movement path.

[0080] Control the tethered balloons to reach the 3D grid set sequentially according to the planned path. Each sampling point in the system is maintained at its position. During the tethered balloon's stay, three lidars are controlled to perform a coordinated and precise scan of the corner reflector position carried by the balloon. By monitoring and locking the peak direction of the echo signal-to-noise ratio (CNR) in real time, the precise pointing angle of the target center is obtained, thereby eliminating measurement errors caused by beam edge effects. The system simultaneously acquires the true spatial coordinates relative to the base station (i.e., radar A) measured by the tethered balloon's D-RTK positioning system, as well as the observation angles fed back by the encoders of the three radars.

[0081] like Figure 4 As shown, the calibration observation dataset in this embodiment The data structure and some numerical examples are shown in the figure. A total of [number] data were collected during this survey. The valid observation data is formatted and stored as a standard CSV file, calibration_flight_data.CSV. This file strictly contains the following nine data dimensions, which together constitute the input source for joint optimization: Spatial truth (3 columns): Target_x, Target_y, Target_z, which are the X-axis, Y-axis and Z-axis coordinates relative to the radar A origin measured by the tethered balloon D-RTK, respectively; Radar A observations (2 columns): Radar_A_az, Radar_A_el, which are the azimuth and elevation angles measured by Radar A; Radar B observations (2 columns): Radar_B_az, Radar_B_el, which are the azimuth and elevation angles measured by Radar B; Radar C observations (2 columns): Radar_C_az, Radar_C_el, which are the azimuth and elevation angles measured by radar C.

[0082] S5. Based on the constructed multi-radar relative coordinate system and the collected observation data, a global nonlinear error model is constructed and jointly optimized.

[0083] Read the files radar_positions_data.CSV generated in step S1 and calibration_flight_data.CSV generated in step S4, for each radar in the radar network. i A geometric calibration model is constructed and solved based on the nonlinear least squares method. The specific sub-steps are as follows: S51. Define the error state vector to be solved for each radar.

[0084] For any radar Define the four-dimensional error state vector to be optimized. (The format is as shown in equation (5)). Initialize the vector. It is a zero vector or an empirical value.

[0085] S52. Construct a forward observation physical model.

[0086] Based on the principle of rigid body transformation, a mapping function is established from "tethered balloon rectangular coordinates" to "radar theoretical observation angle". For each sampling point m : (1) Calculate the local relative vector: Calculate the D-RTK coordinates of the tethered balloon radar coordinates The difference is used to obtain the relative vector in the world coordinate system. .

[0087] (2) Inverse rotation transformation of coordinate system: Introducing Euler angle rotation matrix Transform the relative vector in the world coordinate system into a vector in the sensor coordinate system with installation tilt. .

[0088] (3) Spherical coordinate projection and deviation superposition: superimposing vectors The theoretical distance, azimuth, and elevation angles are converted to spherical coordinates, and the zero-position deviation is added to obtain the theoretical observation values. .

[0089] S53. Constructing the residual objective function based on the forward observation physical model. .

[0090] S54. Optimize the objective function to minimize it, and obtain the calibration parameters for each radar.

[0091] The Levenberg-Marquardt (LM) algorithm is used to evaluate the objective function. Iterative optimization is performed by continuously adjusting the error parameter vector. This causes the sum of squared residuals to converge to a minimum. When the iteration satisfies that the change in residuals is less than... When the stopping condition is met, output the final optimized solution. These are the calibration parameters of the radar. .

[0092] S55. Estimate the accuracy of calibration parameters.

[0093] Using the calculated optimal calibration parameters Substitute these values ​​back into the forward observation physical model and recalculate the theoretical observations for all sampling points. Define the root mean square error (RMSE) as a quantitative indicator for evaluating the calibration quality.

[0094] In this embodiment, joint optimization is performed based on 125 sets of successfully loaded calibration observation data. Figure 5 This example shows a comparison of the radar observation residual distribution before and after joint optimization using a global nonlinear error model. Figure 5 As shown, the residual distribution before calibration (red scatter) is discrete and significantly deviates from the center, indicating a large systematic error; after calibration (blue mark), the residuals converge closely to the vicinity of zero, verifying the high-precision decoupling capability of the error model.

[0095] The calculated geometric calibration parameters and residual RMSE values ​​for the three radars in this embodiment are as follows: Radar A: Azimuth zero-point deviation Pitch angle zero position deviation Roll angle error Pitch angle error Final residual ; Radar B: Azimuth zero-position deviation Pitch angle zero position deviation Roll angle error Pitch angle error Final residual ; Radar C: Azimuth zero-point deviation Pitch angle zero position deviation Roll angle error Pitch angle error Final residual .

[0096] The above results show that the post-calibration residuals of the three radars are all better than those of the radars in the standard model. The calibration was deemed successful.

[0097] S6. Perform system parameter compensation and calibration based on the optimized calibration parameters.

[0098] The optimal calibration parameters of each radar output in step S5 (Including azimuth zero deviation) Pitch angle zero position deviation Roll angle error and pitch angle error This calibration parameter is written into the parameter configuration file or underlying control algorithm of the multi-radar system. In subsequent wind field collaborative observation tasks, the multi-radar system performs real-time reverse compensation correction based on this calibration parameter.

[0099] Based on the concept of this application, this application also provides a multi-scan lidar cooperative calibration system based on a three-dimensional spatial grid. The system includes a processor and a storage medium. The storage medium stores computer instructions, and the processor executes these computer instructions to perform the multi-scan lidar cooperative calibration method based on a three-dimensional spatial grid as described in the above embodiments or their optional embodiments.

[0100] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. A collaborative calibration method for multi-scan lidar based on a three-dimensional spatial grid, characterized in that, Includes the following steps: S1. Construct and initialize the multi-radar relative coordinate system; S2. Construct a three-dimensional calibration space in the intersection area of ​​each radar detection, and divide the spatial node set; S3. Plan the spatial movement path of the tethered balloon at the set of spatial nodes and synchronize the system time; S4. Perform spatial node traversal and observation data collection according to the planned spatial movement path; S5. Based on the constructed multi-radar relative coordinate system and the collected observation data, a global nonlinear error model is constructed and jointly optimized; The global nonlinear error model characterizes the difference between the theoretical observation and actual measurement of the tethered balloon at each space node; S6. Perform system parameter compensation and calibration based on the optimized calibration parameters.

2. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 1, characterized in that, Step S5 includes: S51. Define the error state vector to be solved for each radar; S52. Construct a forward observation physical model under the error state vector; S53. Construct the residual objective function based on the aforementioned forward observation physical model; S54. Optimize the objective function with the goal of minimizing it, and obtain the calibration parameters of each radar. S55. Estimate the accuracy of the calibration parameters.

3. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 2, characterized in that, The error state vector includes azimuth zero deviation, pitch zero deviation, roll angle error, and pitch angle error.

4. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 3, characterized in that, Constructing a forward observation physical model under the error state vector includes: Based on the principle of rigid body transformation, a mapping function from the spatial coordinates of the tethered balloon to the theoretical observation angle of the radar is established under the error state vector.

5. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 4, characterized in that, Based on the principle of rigid body transformation, a mapping function from the spatial coordinates of the tethered balloon to the theoretical observation angle of the radar is established under the aforementioned error state vector, including: Calculate the local relative vector between the tethered balloon and the radar; Perform an inverse coordinate system rotation transformation on the relative vector; The transformed relative vector is projected onto spherical coordinates and superimposed with the deviation.

6. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 2, characterized in that, The objective function represents the set of residuals between the theoretical radar observations and the actual measurements of the tethered balloon at all spatial grid points under the error state vector.

7. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 2, characterized in that, Estimating the accuracy of the calibration parameters includes: Calculate the root mean square error between the theoretical radar observations and the actual measurements from the tethered balloon under the calibration parameters; If the root mean square error is less than the preset convergence threshold, the corresponding radar calibration is determined to be successful; otherwise, the calibration parameters of the corresponding radar are re-optimized.

8. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in claim 3, characterized in that, System parameter compensation and calibration are performed based on the optimized calibration parameters, including: The calibration parameters are written into the parameter configuration file or underlying control algorithm of the multi-radar system so that real-time reverse compensation correction can be performed based on the calibration parameters in subsequent wind field collaborative observation tasks. The reverse compensation correction includes: when the radar main control system receives an observation command pointing to the coordinates of a predetermined spatial target, the control algorithm first performs coordinate pre-rotation on the relative vector based on the calibrated tilt angle error, and superimposes the calibrated zero position deviation to calculate the corrected encoder angle command that the servo motor actually needs to execute.

9. The multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in any one of claims 1-8, characterized in that, Divide the set of spatial nodes, including: The three-dimensional calibration space is discretized according to the equidistant partitioning rule to construct a structure containing... A set of spatial nodes, n This indicates the number of spatial nodes in each dimension of the three-dimensional calibration space; The height layer span, which is divided vertically in the three-dimensional calibration space, covers the radar observation height.

10. A multi-scan lidar collaborative calibration system based on a three-dimensional spatial grid, characterized in that, It includes a processor and a storage medium; the storage medium stores computer instructions, and the processor executes the computer instructions to perform the multi-scan lidar collaborative calibration method based on a three-dimensional spatial grid as described in any one of claims 1-9.

Images