A control and positioning method of an active variable-angle laser radar unmanned aerial vehicle

Abstract

The application discloses a kind of active variable-angle laser radar unmanned plane control and positioning method, including variable-angle laser radar dynamic control module, two-stage external parameter calibration module, radar point cloud distortion removal module and fusion positioning algorithm module, by real-time monitoring point cloud distribution standard deviation and obstacle density, the optimization number of times of iterative Kalman filtering is dynamically adjusted, the positioning accuracy under extreme working conditions such as complex electromagnetic interference, dense vegetation shelter is significantly improved.Combined with robust estimation strategy and covariance convergence threshold, the system can quickly suppress sensor noise and environmental dynamic interference, ensure that horizontal positioning error and elevation error value are small, while avoiding the convergence shock caused by overfitting.The mechanism realizes the dynamic balance of computing resources and positioning accuracy by sliding window statistics point cloud dispersion and rasterization density mapping, suitable for stable operation demand of heterogeneous scene such as power transmission tower, tree crown layer.

Description

Technical Field

[0001] This invention designs an autonomous return system for a multi-rotor unmanned aerial vehicle (UAV), specifically involving a control and positioning method for an active variable-angle lidar UAV. Background Technology

[0002] With the increasing demand for drones in complex environments, their positioning and perception capabilities face severe challenges. Traditional positioning technologies rely on global navigation satellite systems, but signals are prone to failure indoors, in environments with dense obstacles, or with electromagnetic interference, leading to decreased positioning accuracy or even loss of control. Therefore, simultaneous localization and mapping (SLAM) technology based on lidar has become a research hotspot, but it still suffers from the following problems:

[0003] Vertical field of view limitation: Traditional fixed lidar has a limited vertical scanning angle, resulting in insufficient pitch characteristics, which can easily lead to positioning degradation in narrow or tall buildings.

[0004] Insufficient motion distortion compensation: When UAVs move at high speed or change attitude, point cloud data are easily affected by spatiotemporal distortion. Existing methods mostly rely on IMU high-frequency data interpolation, but do not fully combine the pose compensation mechanism of radar dynamic angle adjustment.

[0005] Poor stability of external parameter calibration: The external parameters of the lidar and UAV body coordinate system are easily affected by factors such as mechanical vibration and temperature deformation. Existing calibration methods are mostly static offline calibration, which is difficult to adapt to dynamic flight scenarios.

[0006] A search revealed a Chinese patent document that discloses an autonomous positioning method, device, and drone for unmanned aerial vehicles (UAVs) [Application No.: 202011204006.2, Publication No.: CN112596071B]. This document proposes a multi-sensor fusion method for autonomous positioning of UAVs, but it does not extend the vertical field of view, lacks an online calibration mechanism, and relies on visual and IMU data, making it unsuitable for increasingly demanding requirements.

[0007] In summary, this invention achieves low horizontal error and high efficiency in defect identification through dynamic radar field of view expansion, two-stage calibration, and sub-millimeter point cloud synchronization, thus ensuring high-precision operation of UAVs in complex environments. Summary of the Invention

[0008] A control and positioning method for an active variable-angle lidar UAV, characterized by being achieved through the coordinated operation of the following modules:

[0009] Variable-angle lidar dynamic control module: Based on a servo motor-driven active rotating gimbal, it adjusts the pitch angle of the lidar in real time and expands the vertical scanning field of view;

[0010] Two-stage extrinsic parameter calibration module: Through offline calibration of mechanical zero position and online dynamic calibration, combined with shaft length compensation, the extrinsic parameter relationship between the lidar and the UAV body coordinate system is corrected in real time;

[0011] Radar point cloud distortion correction module: Based on IMU high-frequency data and feedback from the rotating shaft angle encoder, it compensates point by point for spatiotemporal distortion caused by radar angle adjustment and UAV movement;

[0012] Fusion positioning algorithm module: By tightly coupling IMU recursive data and distortion-free point cloud, combined with incremental KD-Tree data association and iterative Kalman filter optimization, the position and attitude of the UAV are controlled in real time.

[0013] Preferably, the two-stage extrinsic calibration module includes:

[0014] Phase 1: In the mechanical zero position, the initial external parameter matrix between the lidar and the body coordinate system is estimated offline by using known reference objects and IMU data at the calibration site;

[0015] The second stage involves dynamically adjusting the external parameter matrix during flight by combining the shaft length, motor encoder angle feedback, and IMU data to compensate for mechanical deformation and wear errors.

[0016] The above technical solution systematically achieves full lifecycle management of lidar extrinsic parameters in both static and dynamic scenarios. In the static calibration phase, a reference coordinate system mapping relationship is established through multi-sensor data fusion. In the dynamic calibration phase, an adaptive compensation network is constructed based on a kinematic chain model and a real-time feedback mechanism, effectively suppressing parameter drift caused by mechanical vibration, temperature deformation, and component wear. In scenarios such as power transmission line sag detection and high-tower bolt loosening identification, this solution ensures stable detection of millimeter-level defects through dynamic correction of the extrinsic parameter matrix. Simultaneously, it is compatible with the continuous modeling requirements under complex flight maneuvers such as rapid acceleration and large-angle maneuvers of UAVs, significantly improving the spatiotemporal consistency of 3D reconstruction.

[0017] Preferably, the radar point cloud distortion correction module includes: estimating the radar pose of each laser point in real time based on IMU data and feedback from the axis angle encoder; transforming the point cloud from the radar coordinate system to the global coordinate system; eliminating spatiotemporal errors caused by radar angle adjustment and UAV movement; and using a linear interpolation algorithm to compensate for motion distortion caused by time lag.

[0018] Through the above technical solution, the system achieves spatiotemporal consistency correction of radar point clouds during dynamic scanning. Based on the joint pose calculation of IMU and shaft encoder, the discrete point cloud acquired by lidar is uniformly mapped to the global coordinate system, effectively eliminating non-rigid deformation caused by UAV attitude changes and gimbal active rotation. Linear interpolation algorithms are used to compensate for the time delay of sensor data, combined with dynamic matching of radar scanning frequency and motor angular velocity, ensuring that the spatial position of each laser point is strictly aligned with the actual radar pose at the time of acquisition. In scenarios such as power line sag detection and insulator defect identification, this solution significantly improves the geometric accuracy and detail integrity of 3D model reconstruction through real-time point cloud distortion removal, providing a centimeter-level precision perception data foundation for autonomous UAV inspection.

[0019] Preferably, the state variables to be optimized in the fusion positioning algorithm module for:

[0020] ;

[0021] Where G represents the world coordinate system (global coordinate system), and I represents the IMU coordinate system. These represent the IMU's position, velocity, and attitude in the global coordinate system, respectively. , These represent angular velocity and acceleration zero bias, respectively. Represents the gravity vector in the world frame. , , These represent the rotation, translation extrinsic parameter, and rotation axis length between the radar and IMU coordinate systems, respectively, and are all optimized state variables of the system.

[0022] The state variables are optimized based on iterative Kalman filters, and the residual is calculated as the distance between the projection point and the local plane in the IKD-Tree map;

[0023] The state variables are optimized by backward optimization of the residual gradient, the covariance matrix is ​​updated, and the optimal pose is output.

[0024] Through the above technical solutions, a tightly coupled optimization framework incorporating the error characteristics of multiple sensor sources is constructed using the integrated positioning algorithm module. The global pose parameters of the IMU and the radar extrinsic parameters in the state vector are updated iteratively without singularity through Lie group manifold modeling. The rotation parameters are represented using quaternions or rotation vectors to avoid Euler angle deadlock. During the iterative Kalman filtering process, the distortion-free radar point cloud is projected onto the ikd-Tree dynamic map. Local plane equations are matched using k-nearest neighbor search, and Mahalanobis distance residuals are calculated. When calculating the residual gradient, partial derivatives are taken with respect to IMU pose, extrinsic parameter offset, and rotation axis length. The Jacobian matrix of the extrinsic parameter rotation is derived using the chain rule combined with the radar scanning geometry model, while the translation component is linearly dependent based on the kinematic chain relationship of the rotation axis. In the covariance update stage, an adaptive noise covariance estimation method is used to dynamically adjust the observation noise weights according to the residual distribution, suppressing the interference of dynamic obstacles or anomalies on positioning accuracy. The final optimal pose output is smoothed through a sliding window to ensure the temporal continuity of the state estimation and is fed back to the flight control system in real time to achieve centimeter-level trajectory tracking. This scheme maintains decimeter-level positioning accuracy even in complex electromagnetic interference and dynamic obstacle scenarios through full-state joint optimization and closed-loop correction of map matching residuals, meeting the autonomous obstacle avoidance and precision operation requirements of UAVs in confined spaces such as power line corridors.

[0025] Preferably, in the second stage of calibration, the shaft length is defined as the preset distance from the gimbal's rotation axis to the lidar mounting point. Combining the motor encoder angle feedback and the shaft length, the extrinsic parameter matrix is ​​dynamically corrected to ensure a calibration error ≤1mm. (A correction formula can be added). Assuming the extrinsic parameter for the mechanical zero position is... The angular correction amount generated during the rotation of the shaft is Then the rotation matrix and translation correction of the corresponding extrinsic parameters can be obtained. , Compensated external parameters for

[0026] .

[0027] The core implementation process of the second-stage dynamic extrinsic parameter calibration, based on the above technical solution, can be summarized as follows: Based on the preset shaft length parameter from the gimbal's rotation axis to the radar mounting point, combined with the real-time feedback of rotation angle changes from the motor encoder, the rotation and translation components of the extrinsic parameter matrix are dynamically corrected using a kinematic chain model. The initial mechanical zero-position extrinsic parameter matrix is ​​incrementally updated through Lie group multiplication of the shaft angle compensation. The correction of the rotation component is achieved through the exponential mapping relationship of the shaft axis, while the correction of the translation component is based on the displacement vector constructed according to the relationship between the shaft length and trigonometric functions. The system verifies the effectiveness of the correction parameters in real time through backpropagation of the point cloud projection residuals. When abnormal deviations are detected, online recalibration based on gradient descent is triggered to ensure that the dynamic calibration error is strictly constrained within a preset threshold. This solution, through the synergistic effect of the shaft kinematic model and data-driven optimization, continuously ensures the geometric consistency of the extrinsic parameter matrix during UAV flight.

[0028] Preferably, time lag compensation includes: recording the acquisition timestamp of each laser point, calculating the UAV pose corresponding to the timestamp based on IMU data interpolation, and back-projecting the point cloud data to the global coordinate system.

[0029] Through the above technical solution, the system achieves high-precision spatiotemporal synchronization between laser point cloud acquisition and carrier motion status. Based on a hardware-level timestamp synchronization mechanism, a strict temporal correspondence is established between the emission time of each laser point and the IMU data stream. The Bezier curve interpolation algorithm is used to reconstruct the continuous pose change trajectory of the UAV within any small time interval. The point cloud data is inversely projected to the global coordinate system through a kinematic inversion model, eliminating the layered distortion effect caused by carrier displacement and attitude vibration during the scanning cycle. In scenarios such as power line galloping monitoring and tower bolt loosening detection, this compensation mechanism effectively restores the real spatial geometric relationships under complex motion states with sub-millimeter-level point cloud alignment accuracy, providing reliable underlying data support for defect identification and quantitative analysis.

[0030] Preferably, the IKD-Tree map is dynamically updated in the following way: the newly scanned point cloud is divided into local blocks, duplicate points are merged and noise points are removed, and the map update frequency is synchronized with the radar scan frame rate.

[0031] Through the above technical solution, the dynamic update module of the IKD-Tree map achieves real-time fusion and management of high-frequency environmental perception data. Newly acquired point clouds are segmented into local blocks through spatial density clustering, and corresponding areas in existing maps are quickly matched using a geometric hash algorithm, eliminating duplicate points and optimizing the uniformity of point cloud distribution. The system combines scan line temporal information and the curvature characteristics of neighboring points to identify and filter outliers introduced by dynamic obstacles or sensor noise. Simultaneously, a sliding window mechanism is used to constrain the map size, retaining only recently valid point cloud data to reduce memory usage. The map update process is strictly synchronized with the radar scan frame rate. Parallel threads handle point cloud preprocessing and incremental tree structure optimization operations, ensuring real-time reflection of environmental changes along the UAV inspection path while maintaining sub-millisecond query efficiency. This solution significantly improves the consistency of map reconstruction in densely vegetated areas during high-voltage corridor tree obstacle detection, providing a highly reliable dynamic spatial benchmark for refined flight path planning and obstacle avoidance decisions.

[0032] Preferably, in the iterative Kalman filter optimization, the state variable update frequency is ≥30Hz, and the covariance matrix convergence threshold is set to 0.01 to ensure the stability of the real-time pose output.

[0033] Through the above technical solutions, a tightly coupled positioning optimization architecture is constructed. Iterative Kalman filtering incrementally updates the global state variables at a frequency of 30Hz, balancing computational load and parameter convergence speed through a sliding window constraint mechanism, achieving millisecond-level single-iteration time on an embedded processor. During the optimization process, the covariance matrix employs an eigenvalue threshold truncation strategy; when the standard deviation of the main diagonal elements is below 0.01, it is considered to be in a convergent state, and the weights of historical observation data within the current window are frozen to suppress overfitting risk. For complex operating conditions such as rapid UAV acceleration and strong electromagnetic interference, an adaptive noise estimation module is designed to dynamically adjust the process noise covariance matrix, detecting abnormal residuals through Mahalanobis distance and triggering a robust weighting mechanism. This solution achieves stable output with horizontal positioning errors ≤3cm and elevation errors ≤5cm in refined inspection of power transmission lines. Even in GPS-denied environments, it can maintain sub-decimeter-level positioning accuracy through point cloud-map matching, significantly improving the reliability of autonomous operation of UAVs in scenarios with strong winds and dense obstacles.

[0034] Preferably, in the dynamic extrinsic parameter correction, the mechanical deformation compensation term is calculated using a linear regression model of historical calibration data and the shaft angle; the regression coefficients are calibrated through offline experiments and used to correct the extrinsic parameter matrix online. Assume the shaft angle is... The deformation compensation observed in the historical calibration data is (Changes in the length of the rotating shaft) can be modeled as an angle. Linear functions:

[0035] ;

[0036] in For regression coefficients, This is a Gaussian noise term. The shaft length was obtained from offline calibration. (Compensation-derived length) for

[0037] .

[0038] Through the above technical solutions, the system constructs an adaptive and tightly coupled positioning framework by fusing multi-sensor data and dynamic environmental feedback in real time. The iterative Kalman filter module, based on high-frequency state updates and combined with a covariance convergence threshold determination mechanism, ensures the stability and real-time performance of pose estimation. Through a dynamic extrinsic parameter correction model and mechanical deformation compensation strategy, the system effectively suppresses parameter drift under complex working conditions, ensuring the consistency of global coordinate system mapping. The point cloud distortion correction module, based on spatiotemporal synchronization and back projection algorithms, eliminates geometric deformation errors caused by carrier motion, providing high-fidelity input for 3D reconstruction. In scenarios such as UAV autonomous inspection and dynamic obstacle avoidance, this solution significantly improves the collaborative accuracy of environmental perception and pose control through end-to-end closed-loop optimization, providing technical support for reliable operation in complex engineering scenarios.

[0039] Preferably, in the fusion positioning algorithm module, the number of optimization iterations of the Kalman filter is dynamically adjusted according to the point cloud density and environmental complexity. In complex scenarios, the number of optimization iterations is increased to 10 to enhance positioning accuracy, while in simple scenarios, the number of optimization iterations is reduced to 3 to reduce computational latency. The adjustment of the number of optimization iterations is achieved by real-time monitoring of the standard deviation of the point cloud distribution and the density of environmental obstacles.

[0040] Through the above technical solutions, the system achieves adaptive optimization capabilities for the fusion positioning algorithm. Based on real-time point cloud distribution standard deviation calculation and obstacle density assessment, the iteration depth of the Kalman filter is dynamically adjusted: when the standard deviation of sparse areas of the point cloud exceeds a threshold or obstacle density increases significantly, the number of optimization iterations is automatically increased to the upper limit, enhancing pose convergence accuracy under complex geometric conditions through refined gradient descent; conversely, in open scenes, the number of iterations is reduced to release computing resources. This mechanism uses a sliding window to statistically analyze the spatial dispersion of the point cloud, combined with rasterized density mapping to quantify environmental complexity in real time, and constructs a nonlinear weight function to drive the smooth switching of iterations. The system synchronously monitors the trace change rate of the pose covariance matrix, and when the convergence speed is lower than a dynamic threshold, it actively triggers an iteration number compensation strategy to avoid local optimum traps. When UAVs traverse heterogeneous environments such as power transmission towers and tree canopies, this solution achieves synergistic optimization of centimeter-level positioning stability and millisecond-level response speed through dynamic balance between computing resources and positioning accuracy, significantly improving the system robustness in complex operating scenarios.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] 1. This patent significantly improves positioning accuracy under extreme conditions such as complex electromagnetic interference and dense vegetation obstruction by dynamically adjusting the optimization times of iterative Kalman filtering through real-time monitoring of the point cloud distribution standard deviation and obstacle density. Combined with a robust estimation strategy and a covariance convergence threshold, the system can quickly suppress sensor noise and dynamic environmental interference, ensuring small horizontal and vertical positioning errors while avoiding convergence oscillations caused by overfitting. This mechanism achieves a dynamic balance between computational resources and positioning accuracy through a sliding window statistical mapping of point cloud discreteness and raster density, making it suitable for stable operation requirements in heterogeneous scenarios such as power transmission towers and tree canopies.

[0043] 2. This patent, based on feedback from the shaft angle encoder and historical calibration data, uses a linear regression model to correct the rotation and translation components of the extrinsic parameter matrix in real time, compensating for mechanical deformation and wear errors. Through reverse verification using point cloud projection residuals, combined with temperature-deformation coupling correction coefficients, long-term stability with small calibration errors is achieved. This technology solves the problem of extrinsic parameter drift under conditions such as rapid acceleration and large-angle maneuvers of UAVs, providing reliable assurance for millimeter-level 3D modeling. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the process of this invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Those skilled in the art will understand that embodiments of the present invention can be implemented as a system, apparatus, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0047] Example 1: High-precision 3D reconstruction of urban building facades

[0048] 1. Implementation of variable angle lidar structure

[0049] An active pitch-adjustable gimbal is installed at the bottom of the hexacopter drone to drive the lidar for angle adjustment. The gimbal base integrates a three-axis vibration sensor to monitor and compensate for high-frequency mechanical vibrations in real time.

[0050] 2. Two-stage external parameter calibration implementation

[0051] Phase 1 Laboratory Calibration:

[0052] Sixteen ceramic calibration spheres, each 25cm in diameter (with 300 feature points on their surface), were arranged in a 10m×10m calibration field. Their coordinates were determined using a laser tracker (error <0.1mm). A UAV hovered at the center of the field, with the gimbal zeroed, and collected 30 minutes of static data. Calibration results:

[0053] ;

[0054] ;

[0055] The shaft length A = 0.325m (measured by a laser tracker).

[0056] Second-stage online calibration:

[0057] During flight, the encoder angle θ is read in real time (example value: +35° corresponds to the encoded value 0x8FFF), and the extrinsic parameters are dynamically updated.

[0058] ;

[0059] Verification is performed every 5 minutes using facade plan fitting, and the shaft length error is controlled within ±1.2mm.

[0060] 3. Implementation of Radar Point Cloud Distortion

[0061] Time synchronization system:

[0062] The IMU (200Hz), encoder (1kHz), and radar (10Hz) achieve μs-level synchronization via PPS pulses.

[0063] A single frame of point cloud spans 100ms and is divided into 20 5ms time slices.

[0064] Motion compensation calculation (example point):

[0065] Laser dot timestamp: t=1523.6ms

[0066] The corresponding encoder angle θ = 28.7° (linear interpolation)

[0067] The IMU integration yields the body displacement Δp = [0.12, -0.03, 0.05] m.

[0068] Coordinates after compensation:

[0069] ;

[0070] The original coordinates of a corner point of a window frame are [5.213, 3.078, 22.456]m, which are then corrected to [5.216, 3.081, 22.451]m.

[0071] 4. Tightly Coupled Fusion Positioning Implementation

[0072] State vector initialization:

[0073] ;

[0074] Covariance matrix:

[0075] ;

[0076] Optimize the iterative process:

[0077] IMU Pre-integration:

[0078] Integrating over Δt=0.1s yields:

[0079] ;

[0080] ;

[0081] ;

[0082] Point cloud matching:

[0083] The plane equation of a point cloud on a certain wall surface is: 0.707x + 0.707y + 0.000z - 10.000 = 0

[0084] Point-to-plane residual:

[0085] ;

[0086] Jacobian matrix calculation:

[0087] ;

[0088] ;

[0089] Real-time output:

[0090] Optimal pose:

[0091] ;

[0092] ;

[0093] Positioning accuracy: horizontal error 2.1cm RMS, elevation error 3.5cm RMS.

[0094] Example 2: Detailed Inspection of Power Transmission Line Corridors

[0095] 1. Implementation of variable angle lidar structure

[0096] Hardware configuration:

[0097] An active pitch-adjustable gimbal is installed at the bottom of the hexacopter drone to drive the lidar to achieve an angle adjustment from -45° to +60°. The gimbal base integrates a dual-axis vibration monitoring module to suppress high-frequency mechanical vibration in real time.

[0098] 2. Two-stage external parameter calibration implementation

[0099] Phase 1 Laboratory Calibration:

[0100] Sixteen ceramic calibration spheres, each 25cm in diameter (with 300 feature points on their surface), were arranged in an 8m×8m calibration field. Their coordinates were determined using a laser tracker (error <0.1mm). A UAV hovered at the center of the field, with the gimbal zeroed, and collected 20 minutes of static data. Calibration results:

[0101] ;

[0102] ;

[0103] The shaft length A = 0.362m (measured by a laser tracker).

[0104] Second-stage online calibration:

[0105] During flight, the encoder angle θ is read in real time (example value: +40° corresponds to the encoded value 0xA3D7), and the extrinsic parameters are dynamically updated.

[0106] ;

[0107] The shaft length is optimized every 10 minutes by checking the axial consistency of the conductor, with a maximum compensation of ΔA = 1.2 mm.

[0108] 3. Implementation of Radar Point Cloud Distortion

[0109] Time synchronization system:

[0110] IMU (200Hz), encoder (1kHz), and radar (10Hz) achieve μs-level synchronization via PTP protocol.

[0111] A single frame of point cloud spans 200ms and is divided into 20 time slices of 10ms each.

[0112] Motion compensation calculation (example point):

[0113] Laser dot timestamp: t=2845.3ms

[0114] The corresponding encoder angle θ = 32.5° (linear interpolation)

[0115] The IMU integral yields the body displacement Δp = [0.15, −0.04, 0.08] m.

[0116] Coordinates after compensation:

[0117] ;

[0118] The original coordinates of a certain conductor splice pipe are [158.72, 92.34, 25.16]m, which are then corrected to [158.73, 92.33, 25.15]m.

[0119] 4. Tightly Coupled Fusion Positioning Implementation

[0120] State vector initialization:

[0121] ;

[0122] Covariance matrix:

[0123] ;

[0124] Optimize the iterative process:

[0125] IMU Pre-integration:

[0126] Integrating over Δt=0.1s yields:

[0127] ;

[0128] ;

[0129] ;

[0130] Point cloud matching:

[0131] Search for 5 neighboring points within a 0.5m radius.

[0132] The equation of the plane is: 0.866x + 0.500y + 0.000z - 50.000 = 0

[0133] Point-to-plane residual:

[0134] ;

[0135] Jacobian matrix calculation:

[0136] ;

[0137] ;

[0138] Optimal pose:

[0139] ;

[0140] ;

[0141] Positioning accuracy: horizontal error 2.3cm RMS, elevation error 3.7cm RMS.

[0142] 5. Implementation effect verification

[0143] Test environment:

[0144] Gear distance: 250m

[0145] Maximum wind speed: 10 m / s

[0146] Quantitative indicators:

[0147] Scan coverage:

[0148] Vertical range: 36m one way (12m for traditional method)

[0149] Horizontal efficiency: 3.2 km / h (240% improvement over fixed radar)

[0150] Positioning accuracy:

[0151] Horizontal error: 2.5cm RMS

[0152] Elevation error: 3.0cm RMS

[0153] Attitude angle error: 0.02°RMS

[0154] Defect detection:

[0155] Minimum identifiable defect:

[0156] Wire strands: 2mm

[0157] Insulator flashover damage: 5cm²

[0158] Bolt loosening: 1.5mm displacement

[0159] System stability:

[0160] Parameter drift under continuous vibration conditions: <0.005%

[0161] Communication interruption recovery time: <500ms.

[0162] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0163] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0164] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A control and positioning method for an active variable-angle lidar UAV, characterized in that, This is achieved collaboratively through the following modules: Variable-angle lidar dynamic control module: Based on a servo motor-driven active rotating gimbal, it adjusts the pitch angle of the lidar in real time and expands the vertical scanning field of view; Two-stage extrinsic parameter calibration module: Through offline calibration of the mechanical zero position and online dynamic calibration, combined with shaft length compensation, the extrinsic parameter relationship between the lidar and the UAV body coordinate system is corrected in real time; assuming the extrinsic parameter of the mechanical zero position is... The angular correction amount generated during the rotation of the shaft is Then the rotation matrix and translation correction of the corresponding extrinsic parameters can be obtained. , Compensated external parameters for ； The correction of the rotation component is achieved through the exponential mapping relationship of the axis of rotation, while the correction of the translation component is based on the relationship between the axis of rotation and trigonometric functions to construct the displacement vector; the radar point cloud distortion correction module: based on IMU high-frequency data and feedback from the axis angle encoder, it compensates for spatiotemporal distortion caused by radar angle adjustment and UAV movement point by point. Fusion positioning algorithm module: By tightly coupling IMU recursive data and distortion-free point cloud, combined with incremental KD-Tree data association and iterative Kalman filter optimization, the position and attitude of the UAV are controlled in real time; The state variables to be optimized in the fusion positioning algorithm module for: ； Where G represents the world coordinate system and I represents the IMU coordinate system. These represent the IMU's position, velocity, and attitude in the global coordinate system, respectively. , These represent angular velocity and acceleration zero bias, respectively. Represents the gravity vector in the world frame. , , These represent the rotation, translation extrinsic parameter, and rotation axis length between the radar and IMU coordinate systems, respectively, and are all optimized state variables of the system. The state variables are optimized based on iterative Kalman filters, and the residual is calculated as the distance between the projection point and the local plane in the IKD-Tree map; The state variables are optimized by backward optimization of the residual gradient, the covariance matrix is ​​updated, and the optimal pose is output.

2. The control and positioning method for an active variable-angle lidar UAV according to claim 1, characterized in that, The external parameter calibration module includes: With the machine at zero position, the initial external parameter matrix between the lidar and the body coordinate system is estimated offline by using known reference objects and IMU data at the calibration site. During flight, the external parameter matrix is ​​dynamically adjusted by combining the shaft length, motor encoder angle feedback and IMU data to compensate for mechanical deformation and wear errors.

3. The control and positioning method for an active variable-angle lidar UAV according to claim 1, characterized in that, The radar point cloud distortion correction module includes: estimating the radar pose of each laser point in real time based on IMU data and feedback from the axis angle encoder; transforming the point cloud from the radar coordinate system to the global coordinate system; eliminating spatiotemporal errors caused by radar angle adjustment and UAV movement; and using a linear interpolation algorithm to compensate for motion distortion caused by time lag.

4. The method according to claim 3, characterized in that, The time lag compensation includes: recording the acquisition timestamp of each laser point, calculating the UAV pose corresponding to the timestamp based on IMU data interpolation, and back-projecting the point cloud data to the global coordinate system.

5. The method according to claim 1, characterized in that, The IKD-Tree map is dynamically updated in the following way: the newly scanned point cloud is divided into local blocks, duplicate points are merged and noise points are removed, and the map update frequency is synchronized with the radar scan frame rate.

6. The method according to claim 1, characterized in that, In the iterative Kalman filter optimization, the state update frequency is ≥30Hz, and the covariance matrix convergence threshold is set to 0.01 to ensure the stability of the real-time pose output.

7. The method according to claim 1, characterized in that, In the aforementioned fusion positioning algorithm module, the number of optimization iterations of the Kalman filter is dynamically adjusted according to the point cloud density and environmental complexity. In complex scenarios, the number of optimization iterations is increased to 10 to enhance positioning accuracy, while in simple scenarios, the number of optimization iterations is reduced to 3 to reduce computational latency. The adjustment of the number of optimization iterations is achieved by real-time monitoring of the standard deviation of the point cloud distribution and the density of environmental obstacles.

Images