Unified Method for Pixel-Level Reference of Four-Source Sensors in Short-Range UAVs (Electro-optical-radar pods)

Abstract

This invention relates to the field of sensor calibration technology for UAV-borne optoelectronic reconnaissance equipment, and particularly to a method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod. The method includes establishing a unified spatial coordinate system for the four sources and demonstrating the full-range representativeness of a Ka-band single-range predetermined calibration; determining the residual attitude using IMU rotation modulation sinusoidal fitting demodulation to establish a horizontal inertial reference; deploying a target checkerboard-antenna integrated structure and calibrating the local coordinate system; determining the absolute direction vector of the infrared optical axis in the horizontal inertial frame through joint calculation by PnP and IMU; determining the absolute direction vector of the radar electrical axis through bidirectional scanning in azimuth and elevation; calculating the actual installation deviation and establishing a full-parameter error budget; and directly calling ground calibration parameters for coarse registration and residual pixel deviation verification during the in-flight phase. The advantages are that it eliminates the need for precise field leveling, achieves pixel-level reference unification through pure calculation, and adopts a unidirectional architecture of ground calibration and in-flight verification.

Description

Technical Field

[0001] This invention relates to the field of sensor calibration technology for UAV airborne optoelectronic reconnaissance equipment, and in particular to a method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod. Background Technology

[0002] The integrated electro-optical radar reconnaissance equipment for short-range unmanned aerial vehicles integrates four sensors—visible light, infrared, laser, and synthetic aperture radar (SAR)—into a single pod. Spatial positional deviations exist between the imaging optical axes of these four sensors, making it impossible to accurately match pixel positions for the same target in the imaging results of different sensors. This severely restricts the accuracy of multi-source image information fusion. More significantly, radar is an active coherent imaging system, sensitive to ground roughness and dielectric properties; visible light and infrared are passive optical imaging systems, relying on reflectance spectral characteristics and thermal radiation characteristics, respectively. This fundamental difference in their imaging mechanisms causes traditional grayscale-based registration methods to completely fail under cross-modal conditions. Furthermore, deep learning-based registration techniques rely on generative adversarial networks for mode conversion, which easily introduces texture distortion and geometric deformation, making it difficult to meet the fidelity requirements of military target interpretation.

[0003] In existing technologies, multi-axis consistency calibration mainly relies on methods such as the large-aperture collimator method, the projection target method, the pentaprism method, and the combination of telescope systems. The large-aperture collimator method achieves multi-axis consistency detection by detecting the deviation of the same target image from the centers of each optical axis. It has few error loops and high accuracy, but it is extremely difficult to manufacture and assemble, the equipment is bulky and expensive, and it is generally only used in laboratories, unable to meet the needs of mobile calibration in the field. The projection target method is usually performed in a field of hundreds of meters, inferring parallelism by the mirror relationship between the optical axis interval and the projection interval of the target. It is convenient to operate and low in cost, but the detection accuracy is strongly correlated with the test distance, greatly affected by ambient light and airflow disturbances, and the human eye's interpretation of the light spot position is highly subjective, making it impossible to guarantee high detection accuracy. The pentaprism method achieves optical axis detection of a large-aperture optical system by shifting and splitting the beam, but it involves numerous adjustment links, and the superposition of various mechanical errors limits the detection accuracy. In addition, there is a combined measurement method using a laser rangefinder, a crosshair target, and a telescope system with a reticle. This method involves aiming the telescope at the reticle and measuring the distance to calculate the optical axis deviation. However, the combination of multiple optical systems results in a large device diameter and size, causing inconvenience for installation, debugging, and transportation. Furthermore, it introduces many error factors, leading to low reliability of the measurement results. In recent years, online automatic axis alignment technology has emerged, which achieves automatic calibration by adding a short-wave infrared detector within the photoelectric device to extract the laser spot in real time. However, this method reduces the installation space for other sensors, lowers system reliability, and during rain or snow, uncontrolled scattering from the laser light can cause the short-wave infrared detector to have difficulty detecting the spot in low visibility or strong light conditions, posing a risk of axis alignment process failure.

[0004] The limitations of the above-mentioned traditional calibration methods are: (1) They are all designed for the optical axis. Collimators, pentaprisms, telescopes, etc. cannot generate microwave targets that can be identified by the radar electric axis, and they cannot handle the changes in the near-field and far-field beam characteristics of the radar; (2) They require precise physical leveling. Both the large-aperture collimator method and the projection target plate method require the pod and target to be strictly leveled (verticality better than 0.03°). The implementation in the field depends on a high-precision bubble level and skilled operators. The preparation time is as long as several hours, and the residual leveling error cannot be quantified, becoming a source of systematic error that is not traceable in the calibration results; (3) There is no unified scheme for the four sources. The calibration of visible light, infrared, and laser is carried out independently from the calibration of radar. There is a lack of mathematical framework and engineering means to incorporate the radar electric axis into the unified optical coordinate system.

[0005] For Ka-band radars, with their extremely short wavelengths (approximately 8.6 mm), the far-field boundary distance is only about 21 m when the effective antenna aperture is 0.3 m. Blindly applying multi-range calibration approaches in an attempt to decouple near-field beam spread and phase center offset errors through multi-range observations reveals that, for Ka-band radars, the difference between the beam spread correction factor at 40 m and the typical operating distance of 400 m is less than 3%. Multi-range observations cannot provide effective parameter differentiation estimates, and forcibly applying these approaches exposes deficiencies in physical understanding. Furthermore, traditional methods introduce atmospheric refraction error models into elevation calibration. However, this model is suitable for radar angle measurement scenarios targeting air targets (where radio waves pass through an atmospheric density gradient layer). On the ground, the atmospheric refractive index along a horizontal propagation path of 40–80 m can be considered a homogeneous medium, with no additional curvature caused by a density gradient. Forcibly introducing atmospheric refraction as a parameter to be estimated is neither physically necessary nor reliable, and it also reduces the model's credibility.

[0006] Therefore, there is an urgent need to develop a ground calibration method that is tailored to the physical characteristics of the Ka-band, does not require precise leveling, and can directly calculate the actual installation deviation of the four sources in a horizontal inertial frame. Furthermore, it is necessary to establish a one-way technical architecture for ground calibration and air verification to ensure that the air mission phase does not consume payload computing power or mission time. Summary of the Invention

[0007] To address the aforementioned problems, this invention provides a method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod.

[0008] The purpose of this invention is to provide a method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod, comprising the following steps: S1. Establish a unified spatial coordinate system of four sources with the infrared optical axis exit as the origin, calculate the far-field boundary distance based on the Ka-band radar parameters, select the predetermined calibration distance, verify that the beam spread factor difference is less than 0.03, and confirm that the single-distance ground calibration has full-range representativeness. S2. Control the pod's orientation to rotate one full circle and collect the IMU's pitch and roll angles. Solve the residual attitude through sine fitting and construct the rotation matrix of the pod body to the horizontal inertial frame. S3. Deploy a checkerboard target, rigidly integrate the radar receiving antenna with the checkerboard, and calibrate the offset vector of the antenna phase center relative to the checkerboard center offline; S4. Converge the visible light, laser, and infrared optical axes to the center of the target, and use the PnP algorithm and IMU to jointly calculate the horizontal inertial coordinates of the checkerboard center to obtain the unit direction vector of the infrared optical axis in the horizontal inertial frame. S5. Control radar azimuth and elevation fine scanning, use parabolic interpolation to extract peak pointing, and calculate the absolute direction vector of radar electric axis in horizontal inertial frame; S6. Calculate the actual installation deviation of the radar relative to the optical reference in the horizontal inertial frame, complete the full parameter error budget and unify it with the four-source pixel level reference, and complete the accuracy confirmation by ground calibration and air verification.

[0009] Preferably, in step S1, the formula for calculating the far-field boundary distance is: In the formula, Indicates the distance to the far-field boundary of the radar antenna; Indicates the effective aperture diameter of the radar antenna; Indicates the radar's operating wavelength; The predetermined calibration distance is 40~80m; the formula for calculating the beam spread correction factor at the predetermined calibration distance is as follows: ; In the formula: Indicates the predetermined calibration distance Beam spread correction factor at the location; Indicates the predetermined calibration distance; Indicates the near-field beam spread intensity coefficient; Indicates the distance to the far-field boundary of the radar antenna; Beam spread factor difference Represented as: ; Indicates typical working distance. This represents the beam spread correction factor at a typical operating distance.

[0010] Preferably, in step S2, the pod rotates one revolution around the azimuth axis, and static data is collected at no less than 12 evenly distributed azimuth angle positions to establish a sinusoidal fitting demodulation observation equation: ; ; In the formula: Indicates the first Azimuth position Measured IMU pitch angle output; Indicates the first Azimuth position Measured IMU roll angle output; Indicates a constant pitch angle; Indicates a constant roll angle; Indicates the perpendicularity deviation of the azimuth axis; Indicates the first One sampling azimuth angle; Indicates measurement noise; Take the residual attitude angle of the pod body relative to the horizontal inertial frame as Construct the rotation matrix from the pod's body coordinate system to the horizontal inertial frame: ; In the formula: This represents the residual pitch angle of the pod body relative to the horizontal inertial frame; This represents the residual roll angle of the pod body relative to the horizontal inertial frame; This represents the 3×3 rotation matrix from the pod's body coordinate system to the horizontal inertial frame; Indicates rotation around the roll axis of the pod body rotation matrix; Indicates rotation about the pitch axis of the pod body The rotation matrix.

[0011] Preferably, in step S3, the integrated target adopts a coplanar rigid integrated structure of a checkerboard grid and a broadband radar receiving antenna, with the main lobe direction of the antenna perpendicular to the front of the target and facing outwards; the three-dimensional offset vector of the antenna phase center relative to the center of the checkerboard grid is calibrated offline using a coordinate measuring machine, and the calibration error is no greater than 0.5mm.

[0012] Preferably, step S4 specifically includes: using the infrared optical axis as a guiding reference, converging the visible light optical axis and the laser optical axis sequentially to the center of the target checkerboard at a predetermined calibration distance; acquiring an infrared checkerboard image and using a sub-pixel corner extraction algorithm to obtain the pixel coordinates of no less than four non-collinear corner points; combining the camera intrinsic parameter matrix, the physical coordinates of the target corner points, and the rotation matrix of the pod body to the horizontal inertial frame, using the PnP algorithm to calculate the three-dimensional coordinates of the checkerboard center in the horizontal inertial frame and the rotation matrix of the target attitude in the horizontal inertial frame; using the pod translation vector obtained by laser ranging, calculating the unit direction vector of the infrared optical axis in the horizontal inertial frame, and simultaneously calculating the variance of the measurement uncertainty of the direction vector.

[0013] Preferably, the camera intrinsic parameter matrix is ​​represented as follows: ; In the formula: Represents the camera intrinsic parameter matrix; , These represent the camera's focal length in the horizontal and vertical directions of the image, respectively. , Represents the coordinates of the camera's principal point on the image plane; The three-dimensional coordinates of the target checkerboard center in the horizontal inertial frame Rotation matrix of the target attitude in the horizontal inertial frame It is expressed as follows:

[0014] ; In the formula: Indicates the center of the chessboard square The three-dimensional coordinates m in the horizontal inertial frame; Represents the three-dimensional coordinates of the center of the checkerboard grid in the coordinate system of the pod body; This represents the translation vector of the pod's origin in the horizontal inertial frame; The rotation matrix represents the target's attitude in the horizontal inertial frame; The rotation matrix representing the target's attitude in the pod's body coordinate system is directly output by PnP; The unit direction vector of the infrared optical axis in the horizontal inertial frame is represented as follows: ; In the formula: This represents the unit direction vector of the infrared optical axis in the horizontal inertial frame; This represents the components of the unit direction vector along the X, Y, and Z axes of the horizontal inertial frame.

[0015] Preferably, step S5 specifically includes the following sub-steps: S51. Lock the pod's pitch angle to the pitch encoder reading when the three optical axes converge, scan in 0.01° steps within ±5° of the azimuth direction, record the received power, and extract the radar's electric axis peak azimuth angle through parabolic interpolation; S52. Lock the pod azimuth angle to the peak azimuth angle, scan in 0.01° steps in the pitch direction, record the received power and extract the peak pitch angle of the radar electric axis through parabolic interpolation; S53. Based on the antenna phase center offset vector calibrated in step S3 and the target pose information calculated in step S4, calculate the three-dimensional coordinates of the radar receiving antenna phase center in the horizontal inertial frame. The radar electric axis direction vector is independently calculated from the azimuth and elevation scan results. The vectors are then combined and weighted with the inverse of the variance of the fitted residual as the weight and normalized to obtain the final absolute direction vector of the radar electric axis.

[0016] Preferably, the peak azimuth angle of the radar electrical axis The expression is as follows: ; In the formula: This represents the peak azimuth angle of the radar electrical axis extracted by interpolation; This represents the azimuth angle corresponding to the point of maximum power in the discrete sampling sequence; Indicates the azimuth scan step angle; The first The received power at each sampling point; Radar electric axis peak elevation angle The expression is as follows: ; In the formula: This represents the peak elevation angle of the radar electrical axis extracted by interpolation; This represents the pitch angle corresponding to the point of maximum power in the discrete sampling sequence; Indicates the pitch scan step angle; The first The received power at each sampling point; The three-dimensional coordinates of the radar receiving antenna phase center in the horizontal inertial frame are as follows: ; In the formula: This represents the three-dimensional coordinates of the phase center of the radar receiving antenna in the horizontal inertial frame. Represents the three-dimensional coordinates of the center of the chessboard square in a horizontal inertial frame; The rotation matrix representing the target's attitude in the pod's body coordinate system; This represents the offset vector of the calibrated antenna phase center from the origin of the target's local coordinate system.

[0017] Preferably, step S6 specifically includes the following sub-steps: S61. In a horizontal inertial frame, based on the absolute direction vector of the radar's electric axis and the unit direction vector of the infrared optical axis, calculate the true azimuth installation deviation and the true elevation installation deviation of the radar relative to the optical reference. S62. Fit the IMU attitude measurement error, PnP calculation error, antenna offset calibration error, and parabolic interpolation fitting error using the covariance propagation law, incorporate them into the full parameter error budget model, and synthesize the total covariance matrix. S63. Based on actual installation deviations, sensor offsets, and beam spread correction factors, establish pixel-level spatial positional relationships for visible light-infrared, laser-infrared, and radar-infrared, and complete the construction of a unified model for the four-source spatial reference. S64. Adopting a ground calibration and air verification architecture, the ground calibration parameters are called to calculate the coarse registration offset, and the validity of the calibration result is determined based on the residual pixel deviation and the 3σ confidence interval.

[0018] Preferably, the actual azimuth installation deviation and the actual pitch installation deviation are expressed as follows: ; ; In the formula: This indicates the actual azimuth installation deviation of the radar's electrical axis relative to the infrared optical axis; These represent the components of the radar's final unit vector along the X and Y axes in the horizontal inertial frame, respectively. These represent the components of the infrared optical axis unit vector along the X and Y axes in the horizontal inertial frame, respectively. This indicates the actual pitch installation deviation of the radar electrical axis relative to the infrared optical axis; This represents the component of the final unit vector of the radar electric axis along the Z-axis of the horizontal inertial frame; This represents the component of the infrared optical axis unit vector along the Z-axis of the horizontal inertial frame; The total covariance matrix is ​​as follows: ; In the formula: The 2×2 covariance matrix represents the final azimuth-pitch deviation; Represents a 2×M Jacobian matrix with elements of . , This represents the M independent error input variables that affect the deviation calculation; Represents the M×M total covariance matrix with all parameters; This represents the total number of independent error sources that participate in error propagation; The unified model for four-source space references is as follows: ; In the formula: Indicates the optimal calibration distance; These represent the horizontal and vertical physical offsets of visible light relative to infrared light at the optimal calibration distance, respectively. These represent the horizontal and vertical physical offsets of the laser relative to the infrared at the optimal calibration distance, respectively. These represent the horizontal and vertical physical offsets of the radar relative to the infrared at the optimal calibration distance, respectively. These represent the projected coordinates of the visible light axis at the optimal calibration distance; These represent the projected coordinates of the laser optical axis at the optimal calibration distance; These represent the projected coordinates of the infrared optical axis at the optimal calibration distance, with a value of 0; , These represent the horizontal and vertical installation tilt angles of the radar's electric axis relative to the pod's mechanical reference, respectively. This represents the beam spread correction factor.

[0019] Compared with the prior art, the present invention can achieve the following beneficial effects: (1) No need for precise physical leveling of the field pod and target. The attitude error algorithm is compensated by IMU rotation modulation and PnP visual calculation, which greatly reduces the calibration conditions and operation difficulty, shortens the field preparation time, and improves the engineering practicality. (2) Based on the far-field characteristics of Ka-band radar, single-range calibration can cover the entire working range, eliminating the need for multi-range setup and measurement, simplifying the calibration process, while avoiding near-field errors and multipath interference, and ensuring stable and reliable calibration results; (3) By using an integrated target, the optical reference and the radar receiving reference are rigidly unified, eliminating the systematic error caused by the relative displacement between the traditional separate target and the antenna, and significantly improving the registration accuracy of the four-source sensor; (4) The actual installation deviation is directly calculated in the horizontal inertial frame, which is not affected by the attitude of the pod or the tilt of the target. It realizes the unification of the spatial reference of four sources: visible light, infrared, laser and radar at the pixel level. The measurement results are traceable and have higher accuracy. (5) Establish a one-way architecture of ground calibration and air verification. In the air, only the calibration parameters are called, without occupying airborne computing power and task time; (6) Construct a full-parameter error budget system to realize error quantification assessment and validity determination. The calibration results are more reliable and the system is more robust. It can be widely applied to the rapid field calibration of various short-range UAV optoelectronic radar pods. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod according to an embodiment of the present invention. Detailed Implementation

[0021] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0023] This invention provides a unified method for pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod, targeting the far-field characteristics of Ka-band radar, and using a single-range predetermined calibration distance R. cal(Preferred 40~80m) Ground calibration achieves full-range coverage; residual attitude is determined by rotation modulation using the inertial measurement unit (IMU) built into the pod, and residual target tilt is determined by checkerboard PnP visual calculation. All attitude and tilt errors are compensated by algorithm in the horizontal inertial frame without the need for precise physical leveling; a complete correction chain is established from attitude error, PnP calculation error, antenna offset calibration error, parabolic interpolation fitting error to full parameter error budget; a unidirectional architecture of ground calibration and air verification is established, in which only ground calibration parameters are called in the air for coarse registration and fusion accuracy confirmation, without any parameter re-estimation.

[0024] See flowchart Figure 1 Specifically, it includes the following steps: S1. Establishment of a unified coordinate system for four sources and verification of the Ka-band far-field calibration distance: Taking the center point of the optical axis exit of the pod's infrared sensor as the origin, a unified three-dimensional Cartesian coordinate system is established. The optical axes, phase centers, and electrical axis directions of the visible light sensor, laser sensor, and radar sensor are all incorporated into the coordinate system for unified description. The unified coordinate system for four sources is represented as follows: ; in, This represents the position vector of the origin of the infrared optical axis, in meters. This represents the position vector of the origin of the visible light axis, in meters. This represents the position vector of the origin of the laser optical axis, in meters. This represents the position vector of the phase center of the radar antenna, in meters. These represent the horizontal and vertical offsets of the visible light axis relative to the origin of the infrared light axis, respectively, in meters (m). These represent the horizontal and vertical offsets of the laser optical axis relative to the origin of the infrared optical axis, respectively, in meters (m). These represent the horizontal and vertical offsets of the radar antenna phase center relative to the origin of the infrared optical axis, respectively, in meters (m). This indicates the longitudinal offset of the radar antenna phase center along the longitudinal axis (optical axis) of the pod, in meters (m). Calculate the far-field boundary distance: In the formula, This indicates the distance to the far-field boundary of the radar antenna, in meters (m). This indicates the effective aperture diameter of the radar antenna, in meters (m). Indicates the radar operating wavelength. ,in c Represents the speed of light in a vacuum. c =3×10 8 m / s, This indicates the radar center frequency, measured in Hz. At the predetermined calibration distance R cal Calculate the beam spread correction factor η(R) within the range of 40m to 80m. cal The beam spread factor difference between the predetermined calibration distance and the typical working distance was verified. <0.03, confirming that the single-distance ground calibration has full-distance representativeness and that multi-distance calibration is not required; after completing the coordinate system construction and far-field verification, the relevant geometric parameters will be used as the reference input for the subsequent whole process; The formula for calculating the beam spread correction factor at the predetermined calibration distance is as follows: ; In the formula: Indicates the predetermined calibration distance The beam spread correction factor at the location is dimensionless. This represents the predetermined calibration distance, with a value range of [value range missing]. The unit is meters (m). This represents the near-field beam spread intensity coefficient, with a typical value. =0.05, dimensionless; This indicates the distance to the far-field boundary of the radar antenna, in meters (m). Beam spread factor difference Represented as: ; Indicates typical working distance. This represents the beam spread correction factor at a typical operating distance. A value less than 0.03 indicates that the predetermined calibration distance is representative of the entire distance.

[0025] For a center frequency of 35 GHz, wavelength λ≈8.6 mm, and effective antenna aperture... Typical parameters = 0.3m, far-field boundary distance ≈21m, predetermined calibration distance R cal The beam spread correction factor at 40m differs from that at the typical working distance of 400m by approximately 2%, a difference that is negligible in engineering practice; therefore, 40m is representative of the full range. Similarly, R... cal The difference is even smaller at 80m, and it still has full-range representativeness. For the Ka band, 40~80m is sufficient to enter the far-field stable region. The beamwidth, phase center pointing, and antenna gain remain constant during subsequent propagation. The single-range calibration results have full-range coverage representativeness. There is no need to introduce beam spread and phase center offset terms that attenuate with distance as parameters to be estimated, which fundamentally simplifies the calibration model and eliminates the engineering complexity of multi-point field deployment. This coordinate system, together with the far-field demonstration, constitutes the foundation for the full-link geometric calculation of subsequent pod attitude determination, target deployment, three-light convergence, and radar scanning.

[0026] To verify the full-range representativeness of the single-range calibration, the far-field projection model of this invention can be expressed as: ; In the formula: Indicates the convergence distance of the radar's electric axis in the far field. The projected coordinates at the location, in meters; This represents the far-field convergence distance, with a typical value of 4000m, and the unit is meters. , These represent the horizontal and vertical installation tilt angles of the radar's electric axis relative to the mechanical reference of the pod, respectively, in rad; , These represent the horizontal and vertical coordinate components of the radar antenna phase center in the pod's mechanical reference coordinate system, respectively, in meters; The core function of this step is to establish a unified spatial mathematical benchmark for four-source sensors: visible light, infrared, laser, and radar, and to rigorously verify the predetermined calibration distance R for a single distance in the Ka-band. cal The physical completeness of the ground calibration provides an unambiguous coordinate language and physical basis for all subsequent absolute direction vector determinations and horizontal inertial frame calculations; the established unified coordinate system provides the coordinate origin and reference axis definition for the pod IMU attitude determination in step S2; the far-field verification directly simplifies the subsequent radar calibration model, allowing the radar electric axis pointing at the predetermined calibration distance to represent the full-range characteristics, laying the physical foundation for the subsequent establishment of an absolute direction vector difference model in the horizontal inertial frame. The calibration distance R selected in this step... cal That is, the optimal calibration distance D cal .

[0027] S2. Pod IMU Rotation Modulation Attitude Measurement and Horizontal Inertial Reference Establishment: The optoelectronic pod is rigidly mounted on a standard adjustable base without any precision physical leveling operations, ensuring only stable installation without shaking; the pod is controlled to rotate one revolution around the azimuth axis, at at least 12 evenly distributed azimuth angle positions. Static data acquisition was performed, continuously recording the pitch angle output by the IMU at each location. With roll angle Establish the sinusoidal fitting demodulation observation equation: ; ; In the formula: Indicates the first Azimuth position The measured IMU pitch angle output, in degrees (°); Indicates the first Azimuth position The measured IMU roll angle output is in degrees (°). This represents a constant pitch angle (to be estimated), in degrees (°). This represents a constant roll angle (to be estimated), in degrees (°). This indicates the perpendicularity deviation of the azimuth axis (to be estimated), in degrees (°). Indicates the first Each sampling azimuth angle is in degrees (°). The measurement noise is assumed to follow a zero-mean Gaussian distribution and is expressed in degrees (°). Take the residual attitude angle of the pod body relative to the horizontal inertial frame as Construct the rotation matrix from the pod's body coordinate system to the horizontal inertial frame: ; In the formula: The residual pitch angle of the pod body relative to the horizontal inertial frame is expressed in degrees (°). The residual roll angle of the pod body relative to the horizontal inertial frame is expressed in degrees (°). The dimensionless 3×3 rotation matrix represents the rotation from the pod's body coordinate system to the horizontal inertial frame. This indicates rotation around the X-axis (roll axis) of the pod body. The rotation matrix is ​​dimensionless; This indicates rotation around the Y-axis (pitch axis) of the pod body. The rotation matrix is ​​dimensionless; Simultaneously, the variance of the IMU attitude measurement uncertainty estimate is calculated, serving as a partial error source for subsequent full-parameter error budgeting; the variance of the IMU attitude measurement uncertainty estimate is expressed as follows: ; In the formula: This represents the variance of the residual pitch angle estimate, in degrees squared (°). 2 ); This represents the variance of the residual roll angle estimate, in degrees squared (°). 2 ); This represents the variance of the sinusoidal fitting residuals, expressed in degrees squared (°). 2 ); This indicates the total number of azimuth sampling locations. ≥12, dimensionless.

[0028] Since the coupling error between the IMU installation zero bias and the Earth's rotation exhibits sinusoidal modulation characteristics during azimuth rotation, this step innovatively introduces a sinusoidal fitting demodulation model, which simultaneously solves for the constant pitch angle through least squares fitting. Constant roll angle perpendicularity to azimuth axis The random noise, installation zero bias and azimuth axis tilt error are completely separated mathematically. The key innovation is that the field calibration no longer pursues the absolute level of the mechanical structure, but incorporates the residual attitude as a known system quantity into the subsequent PnP solution and direction vector compensation, so that the horizontal inertial frame becomes an absolute geometric reference that does not depend on physical leveling.

[0029] The rotation matrix output in this step With Estimated Variance Directly input the following steps: In step S4, when calculating the three-dimensional coordinates of the target using PnP, it is necessary to utilize... The camera measurements are transformed from the pod's body coordinate system to the horizontal inertial frame; step S5 also relies on this matrix to complete attitude compensation when calculating the absolute direction of the radar's electric axis.

[0030] S3. Integrated Target Deployment and Precise Calibration of Target Local Coordinate System: A high-flatness aluminum or carbon fiber plate is selected as the target substrate. A black and white alternating square checkerboard pattern is printed on the front of the target facing the pod. The number of squares on each side of the checkerboard is no less than 11, and the physical side length of each square is no less than 50mm (e.g., an 11×11 array, with each square measuring 50mm×50mm). The geometric center point of the checkerboard is defined as the origin of the target local coordinate system. A complete local coordinate system for the target is established, with the horizontal and vertical directions within the target plane as the X and Y axes, and the target's frontal normal as the Z axis. The local coordinate system is represented as follows: ; In the formula: Represents the target's local coordinate system; The geometric center of the chessboard is represented by the origin of the target's local coordinate system, and the unit is meters. This represents a horizontal and vertical unit vector within the target plane, and is dimensionless. This represents the frontal normal vector of the target, which is aligned with the direction of the antenna's main lobe and is dimensionless. This represents the frontal normal vector of the target, which is dimensionless. A wideband radar receiving antenna is rigidly embedded on the outer edge or corner of the checkerboard array to ensure that there is no relative displacement between the antenna and the target. The antenna aperture is less than 1 / 5 of the side length of a single checkerboard grid and does not obstruct the corner points of the checkerboard. The main lobe of the antenna is perpendicular to the front of the target and faces outward (pointing towards the pod). The gain is ≥10dBi and the polarization is matched with the radar transmission polarization of the pod. The antenna phase center was determined through precision machining and offline calibration using a coordinate measuring machine (CMM). relative to the origin of the local coordinate system of the target 3D offset vector , and These represent the horizontal and vertical offsets of the antenna phase center relative to the center of the checkerboard grid, respectively, in meters (m); calibration error ≤ 0.5 mm; The physical coordinates of all corner points in the target's local coordinate system are calculated based on the checkerboard parameters, as shown below: ; In the formula: Indicates the first The three-dimensional coordinates of each corner point of the chessboard grid in the local coordinate system of the target are given in meters. This represents the number of squares on one side of the chessboard grid; it is dimensionless. This represents the physical side length of a single square, in meters (m). Represents the row and column index of a corner point; it is dimensionless. The target is set up in the calibration area and roughly aligned with the pod. No precise leveling is required. Its residual tilt will be jointly measured by the checkerboard PnP and the pod IMU in step S4. This integrated design ensures a rigid geometric relationship between the optical aiming reference and the radar receiving reference, eliminating the uncontrollable error introduced by the relative displacement of the target and antenna in traditional separate deployments.

[0031] The core function of this step is to design and deploy an integrated structure on the front of the target, rigidly integrating the checkerboard visual reference and the radar receiving antenna onto the same plane facing the pod. This establishes a local coordinate system for the target and precisely calibrates the three-dimensional offset of the radar receiving antenna's phase center relative to the checkerboard center, providing a geometric reference for the target end in subsequent absolute direction vector calculations in the horizontal inertial frame. (Calibration) The physical coordinates of the checkerboard corner points are directly input into the PnP calculation in step S4; step S4 calculates the target's pose in the camera coordinate system by identifying the pixel coordinates of the checkerboard corner points in the image and combining them with the known 3D coordinates of the corner points from this step; then, it is combined with the pose from step S2. The antenna phase center is then switched to the horizontal inertial frame.

[0032] S4. Three-axis pixel-level convergence and optical reference absolute direction vector determination: Using the infrared optical axis as the guiding reference, the visible light optical axis and the laser optical axis are sequentially converged to the calibration distance R through the precision servo closed-loop control of the pod. cal At the center of the target's checkerboard pattern, complete the three-axis pixel-level alignment; acquire checkerboard images through an infrared imaging system, and use a sub-pixel corner extraction algorithm to obtain the pixel coordinates of no less than four non-collinear corner points. Ensure corner point positioning accuracy is better than 0.1 pixels; load camera intrinsic parameter matrix. (focal length Main point Camera intrinsic parameter matrix Represented as: ; In the formula: Represents the camera intrinsic parameter matrix; These represent the camera's focal length in the horizontal and vertical directions of the image, respectively, in pixels; Represents the coordinates of the camera's principal point on the image plane, in pixels; Combined with the known corner point physical coordinates from step S3 And the 3×3 rotation matrix from the pod body coordinate system to the horizontal inertial frame measured in step S2. A PnP solution model in a horizontal inertial frame is constructed to solve for the three-dimensional coordinates of the target checkerboard center in the horizontal inertial frame. Rotation matrix of the target attitude in the horizontal inertial frame ;

[0033] ; In the formula: Indicates the center of the chessboard square Three-dimensional coordinates in a horizontal inertial frame, in meters (m). The coordinates of the center of the checkerboard grid in the coordinate system of the pod body are represented by the three-dimensional coordinates of PnP in the pod system, and the unit is m. This represents the translation vector of the pod's origin in the horizontal inertial frame, determined by laser ranging, and is expressed in meters (m). The rotation matrix representing the target's attitude in the horizontal inertial frame is dimensionless. The rotation matrix representing the target attitude in the pod body coordinate system is directly output by PnP and is dimensionless. The translation vector of the pod's origin in the horizontal inertial frame, measured by a laser rangefinder. Calculate the unit direction vector of the infrared optical axis in the horizontal inertial frame: ; In the formula: This represents the unit direction vector of the infrared optical axis in a horizontal inertial frame; it is dimensionless. This represents the components of the unit vector along the X, Y, and Z axes in the horizontal inertial frame. It is dimensionless and independent of the pod's residual attitude. It only reflects the true orientation of the infrared optical axis in the absolute horizontal frame. The variance of the measurement uncertainty estimate for the optical reference direction vector is calculated, incorporated into the subsequent error budget, and encoder readings corresponding to visible light and laser convergence are recorded to complete the establishment of the optical reference. The variance of the measurement uncertainty estimate for the optical reference direction vector is expressed as: ; In the formula: Represents the estimated variance of the unit vector of the infrared optical axis, which is dimensionless; This represents the unit direction vector of the infrared optical axis in a horizontal inertial frame; it is dimensionless. Indicates the center of the chessboard square Three-dimensional coordinates in a horizontal inertial frame, in meters (m). This represents the target center coordinate covariance matrix output by the PnP solution, with dimensions 3×3 and units of m. 2 ; This represents the IMU attitude angle covariance matrix, with dimensions 3×3 and units of degrees squared (°). 2 This variance characterizes the absolute uncertainty of the optical reference and participates in the error synthesis in step S6.

[0034] Since the IMU has converted the pod's attitude to a horizontal inertial frame, the PnP solution results... Within a horizontal inertial frame, no additional attitude compensation is required; the absolute direction unit vector of the infrared optical axis is the direction pointing from the pod's origin. normalized vector After the visible light and laser channels converge using the same criteria, their respective encoder readings are recorded, but the infrared optical axis... As the sole optical benchmark for subsequent radar calibration; the calculated target attitude. Directly input step S5 to transform the phase center of the radar receiving antenna from the target local coordinate system to the horizontal inertial system. This step eliminates the coupling between the residual attitude of the pod and the residual tilt of the target on the optical reference through "visual-inertial joint attitude determination", realizing sub-milliradian level absolute orientation determination.

[0035] S5. Radar electrical axis azimuth and elevation joint scanning and absolute direction vector determination; specifically including the following sub-steps: S51. Fine-grained azimuth scanning and peak azimuth angle extraction: The pod's elevation angle is locked to the elevation direction at the point of convergence of the three optical axes in step S4, and preset to the elevation encoder reading at the point of optical aiming in step S4, so that the radar elevation direction is consistent with the optical reference elevation direction; scanning is performed in 0.01° steps within ±5° of the azimuth direction, and the receiving antenna on the front of the target records the Ka-band received power in real time. The expression is as follows: ; In the formula: Indicates the azimuth angle The radar received power measured by the target receiving antenna, in W; This represents the received power corresponding to the radar's peak transmitted power, expressed in watts (W). The true peak azimuth angle of the radar electronic axis (in the coordinate system of the pod body) is expressed in degrees (°). Indicates the radar azimuth beamwidth, in degrees (°). This represents the receiver link noise and ground clutter residue during azimuth scanning, expressed in W. Parabolic interpolation is performed on discrete power sampling points to extract the peak azimuth angle of the radar electrical axis. This improves peak extraction resolution to the order of 0.001°. ; In the formula: This represents the peak azimuth angle of the radar electric axis extracted by interpolation, in degrees (°). This represents the azimuth angle corresponding to the point of maximum power in the discrete sampling sequence, in degrees (°). This represents the azimuth scan step angle, which is taken in some embodiments. °, the unit is degrees (°); The first The received power at each sampling point is expressed in watts (W).

[0036] S52. Fine-grained elevation scanning and peak elevation angle extraction: The pod azimuth angle is locked to the peak azimuth angle extracted in step S51 to ensure that the radar main lobe is perfectly aligned with the target receiving antenna in the azimuth direction; scanning is performed in 0.01° increments within a ±3° range in the elevation direction, and the Ka-band received power is recorded in real time by the receiving antenna in front of the target. The expression is as follows: ; In the formula: Indicates pitch angle The radar received power measured by the target receiving antenna, in W; This represents the received power corresponding to the radar's peak transmitted power, expressed in watts (W). The true peak elevation angle of the radar electrical axis (in the pod's coordinate system) is expressed in degrees (°). This indicates the radar elevation beamwidth, in degrees (°). This represents the residual noise from the receiver link and ground clutter during elevation scanning, expressed in W. Parabolic interpolation is performed on discrete power sampling points to extract the peak elevation angle of the radar electrical axis. This improves peak extraction resolution to the order of 0.001°. ; In the formula: This represents the peak elevation angle of the radar electric axis extracted by interpolation, in degrees (°). This represents the pitch angle corresponding to the point of maximum power in the discrete sampling sequence, in degrees (°). This represents the pitch scan step angle, which is taken in some embodiments. °, the unit is degrees (°); The first The received power at each sampling point is expressed in watts (W).

[0037] S53. Joint weighted determination of radar receiving antenna phase center coordinates and electrical axis absolute direction vector: combined with the target attitude rotation matrix in the horizontal inertial frame obtained in step S4. and the offset vector of the radar receiving antenna phase center relative to the checkerboard grid center obtained from offline calibration in step S3. Calculate the phase center of the radar receiving antenna Three-dimensional coordinates in a horizontal inertial frame: ; In the formula: This represents the three-dimensional coordinates of the phase center of the radar receiving antenna in the horizontal inertial frame, in meters (m). Represents the three-dimensional coordinates of the center of the chessboard grid in a horizontal inertial frame, in meters; The rotation matrix (horizontal inertial frame) representing the target's attitude in the pod's body coordinate system is dimensionless. This represents the offset vector of the antenna phase center calibrated in step S3 from the origin of the target's local coordinate system, in meters. Based on the results of step S51 (azimuth scan) and step S52 (elevation scan), the direction vector of the radar electric axis in the horizontal inertial frame is calculated independently. With redundant check vector ; ; In the formula: This represents the unit direction vector of the radar's electric axis in the horizontal inertial frame; it is dimensionless. This represents the three-dimensional coordinates of the phase center of the radar receiving antenna in the horizontal inertial frame, in meters (m). This represents the translation vector of the pod's origin in the horizontal inertial frame, determined by laser ranging, and is expressed in meters (m). This represents the components of the unit vector along the X, Y, and Z axes of the horizontal inertial frame; it is dimensionless. This vector is determined by the absolute coordinates of the antenna phase center and the absolute coordinates of the pod origin, and is independent of the encoder readings. ; In the formula: This represents the radar electrical axis redundancy check vector based on elevation scan verification; it is dimensionless. This represents the three-dimensional coordinates of the phase center of the radar receiving antenna in the horizontal inertial frame, in meters (m). This represents the translation vector of the pod's origin in the horizontal inertial frame, determined by laser ranging, and its unit is meters; this vector is related to... Theoretically consistent, used for consistency testing; Using the reciprocals of the variances of the fitting residuals from azimuth and elevation scans as weights, a joint weighted average is performed to obtain the final absolute unit direction vector of the radar electric axis: ; In the formula: This represents the final absolute unit direction vector of the radar electric axis after the azimuth-elevation joint weighted average, and is dimensionless. These represent the weights of the azimuth and elevation scans, respectively, and are dimensionless. These represent the residual variances of the azimuth and elevation parabolic interpolation fitting, respectively, in degrees squared (°). 2 ); The weighted average vector is normalized to ensure that its magnitude is 1, thus completing the determination of the absolute direction vector of the radar electric axis in the horizontal inertial frame.

[0038] The final radar electric axis absolute unit vector output in this step unit vector of infrared optical axis In step S6, the azimuth / pitch difference angle between the two is directly calculated in the horizontal inertial frame to obtain the actual installation deviation. The redundancy estimate of the pitch scan is also used to verify the consistency of the results of step S5. If the difference exceeds the threshold, it indicates that the target attitude is drifting or the measurement is abnormal.

[0039] S6. Real Installation Deviation Calculation, Full-Parameter Error Budgeting, and Aerial Pixel-Level Fusion Verification: This step is used to complete the installation deviation calculation of the four-source sensors and the full-process error synthesis in a horizontal inertial frame. Dynamic verification is achieved by directly using ground calibration results and avoiding recalculation in the air, thus realizing a closed loop for unified spatial reference at the entire pixel level. Specifically, this includes: S61. Calculation of actual installation deviation in horizontal inertial frame: based on the final absolute unit vector of radar electric axis. unit vector of infrared optical axis The actual installation deviation can be directly calculated in a unified horizontal inertial frame: Actual orientation installation deviation: ; In the formula: This indicates the actual azimuth installation deviation of the radar's electrical axis relative to the infrared optical axis, expressed in degrees (°). These represent the components of the final unit vector of the radar electric axis along the X and Y axes of the horizontal inertial frame, respectively, and are dimensionless. These represent the components of the infrared optical axis unit vector along the X and Y axes in the horizontal inertial frame, respectively, and are dimensionless; this difference angle is calculated directly in the horizontal inertial frame and is independent of the pod attitude and target tilt. Actual pitch installation deviation: ; In the formula: This indicates the actual elevation installation deviation of the radar electrical axis relative to the infrared optical axis, expressed in degrees (°). It represents the component of the final unit vector of the radar electric axis along the Z-axis of the horizontal inertial frame, and is dimensionless. This represents the dimensionless component of the unit vector of the infrared optical axis along the Z-axis of the horizontal inertial frame.

[0040] S62. Synthesis of Total Parameter Error Budget and Total Covariance Matrix: The IMU attitude determination error, PnP calculation error, antenna offset calibration error, and parabolic interpolation fitting error are fitted using the covariance propagation law and incorporated into the total parameter error budget model to synthesize the total covariance matrix: ; In the formula: The 2×2 covariance matrix representing the final azimuth-pitch deviation, in degrees squared (°). 2 ); Represents a 2×M Jacobian matrix with elements of . , This represents the M independent error input variables that affect the deviation calculation; they are dimensionless or mixed units. This represents an M×M total covariance matrix with all parameters, the diagonal of which contains... The variance of each component varies in units depending on the combination of parameters; This represents the total number of independent error sources involved in error propagation, and is dimensionless.

[0041] S63. Establish the spatial positional relationship of the four-source sensors at the pixel level: based on the optimal calibration distance. The installation offset of each sensor, the actual installation deviation, and the beam spread correction factor are used to establish the pixel-level spatial position relationship between visible light-infrared, laser-infrared, and radar-infrared, and to complete the construction of a unified model for the four-source spatial reference. ; In the formula: Indicates the optimal calibration distance; These represent the horizontal and vertical physical offsets of visible light relative to infrared light at the optimal calibration distance, respectively, in meters (m). These represent the horizontal and vertical physical offsets of the laser relative to the infrared at the optimal calibration distance, respectively, in meters (m). These represent the horizontal and vertical physical offsets of the radar relative to the infrared at the optimal calibration distance, respectively, in meters (m). These represent the projected coordinates of the visible light axis at the optimal calibration distance, determined by the three-light convergence geometry mentioned above, with units in meters; These represent the projected coordinates of the laser optical axis at the optimal calibration distance, determined by the three-beam convergence geometry mentioned earlier, with units in meters; These represent the projected coordinates of the infrared optical axis at the optimal calibration distance, with the origin set to 0, and the unit being meters. , These represent the horizontal and vertical installation tilt angles of the radar's electric axis relative to the mechanical reference of the pod, respectively, in rad; This represents the beam spread correction factor.

[0042] S64. In-flight dynamic verification and pixel-level fusion accuracy confirmation: A one-way architecture of ground calibration and in-flight verification is adopted, without any parameter re-estimation during flight, only directly calling the ground calibration results: The distance between the pod and the target is obtained in real time by a laser rangefinder. Calculate the coarse registration offset: ; ; In the formula: : Coarse registration offset of radar image relative to infrared image in azimuth and elevation directions, in meters; This indicates the straight-line distance between the pod and the target, measured in real time by the laser rangefinder, in meters (m). The actual azimuth and elevation installation deviation obtained from ground calibration are expressed in degrees (°). A dimensionless coefficient representing the conversion of angular units (°) to radians. After initially aligning the radar image according to the offset, perform a pixel-level comparison with the infrared image to calculate the residual pixel deviation. , : ; ; In the formula: This represents the residual deviation between the radar image and the infrared image in the pixel coordinate system after coarse registration, expressed in pixels. This represents the coordinates of the radar image feature points after coarse registration in the infrared image pixel coordinate system, in pixels. This represents the pixel coordinates of the corresponding feature point in the infrared image, in pixels. The ground calibration covariance matrix is ​​converted into pixel-level total uncertainty through error propagation: ; ; In the formula: These represent the standard deviation of the total uncertainty in the aerial azimuth and pitch directions at the pixel level, respectively, in pixels. They represent the values ​​derived from the covariance matrix. The given azimuth and pitch deviation variances are in degrees squared (°). 2 ); This represents the variance of laser ranging error, with typical values ​​corresponding to... =0.5m, unit is m 2 ; These represent the pixel resolution in the azimuth and pitch directions, respectively, in m / pixel. Execution validity determination: If the residual bias is within the range determined by the ground calibration covariance matrix Pixel-level total uncertainty calculated using the error propagation law If the calibration is within the confidence interval, the ground calibration is considered valid; if it exceeds this interval, an anomaly is detected, such as calibration parameter drift, drastic changes in atmospheric conditions, or structural deformation, triggering a re-execution of the ground calibration process steps S1-S6. The anomaly alarm conditions are as follows: ,or ; This ground calibration and air verification architecture completely avoids the impact of in-flight dynamic calibration on payload computing power and mission time, ensuring that the combat mission process is not interfered with by calibration operations.

[0043] Example 1 Taking a certain type of short-range UAV's integrated electro-optical radar pod as an example, this pod integrates an infrared high-definition camera (resolution 1920×1080 pixels, field of view 2°×1.5°), a visible light camera, a laser rangefinder, and a Ka-band synthetic aperture radar (center frequency 35GHz, wavelength λ=0.0086m, antenna aperture...). =0.3m), the method of the present invention is used to unify the spatial reference at the pixel level of four-source sensors. The specific implementation process is as follows: S1. Establishment of a unified coordinate system for four sources and verification of far-field calibration distance: Establish a unified three-dimensional coordinate system with the infrared optical axis exit as the origin; calculate the radar far-field boundary distance: =2×0.3 2 / 0.0086≈21m; Select calibration distance R cal =40m, calculated to be: ≈1.022, at a typical working distance of 400m ≈1.003; Beam spread factor difference: ≈∣1.022 1.003 |≈0.019<0.03; This satisfies the full-distance representativeness condition. This difference is negligible in engineering, confirming that the 40m single-distance calibration has full-distance representativeness. It can also be implemented at 60m or 80m, with the calibration process remaining unchanged.

[0044] S2. The pod IMU samples at 12 azimuth positions, and the residual pitch angle is obtained by sinusoidal fitting and demodulation. °, residual roll angle °, azimuth axis perpendicularity °, build Attitude estimation uncertainty °, °.

[0045] S3. The front of the target is printed with an 11×11 checkerboard pattern, with each grid measuring 50mm. A receiving antenna (8mm diameter) is embedded in the outer corner of the checkerboard pattern for CMM calibration. .

[0046] S4. The infrared camera aims at the center of the checkerboard pattern, and the attitude is calculated using PnP and IMU combined. Determine the absolute unit direction vector of the infrared optical axis. =(0.99987,0.00300,0.00200).

[0047] S5. Azimuth Scan Extraction of Peak Azimuth Angle °, combined with target attitude and antenna offset, determine the radar electrical axis azimuth component; elevation scan extracts the peak elevation angle. °, combined with weighted average, yields the final absolute unit direction vector of the radar electric axis. =(0.99980,0.00560, 0.00310).

[0048] S6. Calculate the true deviation °, °, covariance matrix The estimated accuracy is better than 0.005°; the spatial positional relationship of the four sources is established, and the radar is horizontally offset relative to the infrared at 40m. m, vertical offset m; Dynamic verification results during aerial application: Flight altitude 500m, laser ranging m, directly call the ground calibration parameters to calculate the coarse registration offset m, m; Residual pixel deviation verification results pixel pixel, and 3σ total uncertainty pixel Pixel comparison shows that if the residual deviation is within the confidence interval, the calibration is deemed valid, and the four-source sensors achieve pixel-level spatial reference unification.

[0049] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0050] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod, characterized in that: Includes the following steps: S1. Establish a unified spatial coordinate system of four sources with the infrared optical axis exit as the origin, calculate the far-field boundary distance based on the Ka-band radar parameters, select the predetermined calibration distance, verify that the beam spread factor difference is less than 0.03, and confirm that the single-distance ground calibration has full-range representativeness. S2. Control the pod's orientation to rotate one full circle and collect the IMU's pitch and roll angles. Solve the residual attitude through sine fitting and construct the rotation matrix of the pod body to the horizontal inertial frame. S3. Deploy a checkerboard target, rigidly integrate the radar receiving antenna with the checkerboard, and calibrate the offset vector of the antenna phase center relative to the checkerboard center offline; S4. Converge the visible light, laser, and infrared optical axes to the center of the target, and use the PnP algorithm and IMU to jointly calculate the horizontal inertial coordinates of the checkerboard center to obtain the unit direction vector of the infrared optical axis in the horizontal inertial frame. S5. Control radar azimuth and elevation fine scanning, use parabolic interpolation to extract peak pointing, and calculate the absolute direction vector of radar electric axis in horizontal inertial frame; S6. Calculate the actual installation deviation of the radar relative to the optical reference in the horizontal inertial frame, complete the full parameter error budget and unify it with the four-source pixel level reference, and complete the accuracy confirmation by ground calibration and air verification.

2. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod according to claim 1, characterized in that: In step S1, the formula for calculating the far-field boundary distance is: In the formula, Indicates the distance to the far-field boundary of the radar antenna; Indicates the effective aperture diameter of the radar antenna; Indicates the radar's operating wavelength; The predetermined calibration distance is 40~80m; the formula for calculating the beam spread correction factor at the predetermined calibration distance is as follows: ； In the formula: Indicates the predetermined calibration distance Beam spread correction factor at the location; Indicates the predetermined calibration distance; Indicates the near-field beam spread intensity coefficient; Indicates the distance to the far-field boundary of the radar antenna; Beam spread factor difference Represented as: ; Indicates typical working distance. This represents the beam spread correction factor at a typical operating distance.

3. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod according to claim 1, characterized in that: In step S2, the pod rotates one revolution around the azimuth axis and performs static data acquisition at no less than 12 evenly distributed azimuth angle positions to establish a sinusoidal fitting demodulation observation equation: ； ； In the formula: Indicates the first Azimuth position Measured IMU pitch angle output; Indicates the first Azimuth position Measured IMU roll angle output; Indicates a constant pitch angle; Indicates a constant roll angle; Indicates the perpendicularity deviation of the azimuth axis; Indicates the first One sampling azimuth angle; Indicates measurement noise; Take the residual attitude angle of the pod body relative to the horizontal inertial frame as Construct the rotation matrix from the pod's body coordinate system to the horizontal inertial frame: ； In the formula: This represents the residual pitch angle of the pod body relative to the horizontal inertial frame; This represents the residual roll angle of the pod body relative to the horizontal inertial frame; This represents the 3×3 rotation matrix from the pod's body coordinate system to the horizontal inertial frame; Indicates rotation around the roll axis of the pod body rotation matrix; Indicates rotation about the pitch axis of the pod body The rotation matrix.

4. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod according to claim 1, characterized in that: In step S3, the integrated target adopts a coplanar rigid integrated structure of a checkerboard grid and a broadband radar receiving antenna, with the main lobe of the antenna perpendicular to the front of the target and facing outwards; the three-dimensional offset vector of the antenna phase center relative to the center of the checkerboard grid is calibrated offline by a coordinate measuring machine, and the calibration error is no greater than 0.5mm.

5. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod according to claim 1, characterized in that: Step S4 specifically includes: using the infrared optical axis as a guiding reference, converging the visible light optical axis and the laser optical axis sequentially to the center of the target checkerboard at a predetermined calibration distance; acquiring an infrared checkerboard image and using a sub-pixel corner extraction algorithm to obtain the pixel coordinates of no less than four non-collinear corner points; combining the camera intrinsic parameter matrix, the physical coordinates of the target corner points, and the rotation matrix of the pod body to the horizontal inertial frame, using the PnP algorithm to calculate the three-dimensional coordinates of the checkerboard center in the horizontal inertial frame and the rotation matrix of the target attitude in the horizontal inertial frame; using the pod translation vector obtained by laser ranging, calculating the unit direction vector of the infrared optical axis in the horizontal inertial frame, and simultaneously calculating the variance of the measurement uncertainty of the direction vector.

6. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod according to claim 5, characterized in that: The camera intrinsic parameter matrix is ​​represented as follows: ； In the formula: Represents the camera intrinsic parameter matrix; , These represent the camera's focal length in the horizontal and vertical directions of the image, respectively. , Represents the coordinates of the camera's principal point on the image plane; The three-dimensional coordinates of the target checkerboard center in the horizontal inertial frame Rotation matrix of the target attitude in the horizontal inertial frame It is expressed as follows: ； In the formula: Indicates the center of the chessboard square The three-dimensional coordinates m in the horizontal inertial frame; Represents the three-dimensional coordinates of the center of the checkerboard grid in the coordinate system of the pod body; This represents the translation vector of the pod's origin in the horizontal inertial frame; The rotation matrix represents the target's attitude in the horizontal inertial frame; The rotation matrix representing the target's attitude in the pod's body coordinate system is directly output by PnP; The unit direction vector of the infrared optical axis in the horizontal inertial frame is represented as follows: ； In the formula: This represents the unit direction vector of the infrared optical axis in the horizontal inertial frame; This represents the components of the unit direction vector along the X, Y, and Z axes of the horizontal inertial frame.

7. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod according to claim 6, characterized in that: Step S5 specifically includes the following sub-steps: S51. Lock the pod's pitch angle to the pitch encoder reading when the three optical axes converge, scan in 0.01° steps within ±5° of the azimuth direction, record the received power, and extract the radar's electric axis peak azimuth angle through parabolic interpolation; S52. Lock the pod azimuth angle to the peak azimuth angle, scan in 0.01° steps in the pitch direction, record the received power and extract the peak pitch angle of the radar electric axis through parabolic interpolation; S53. Based on the antenna phase center offset vector calibrated in step S3 and the target pose information calculated in step S4, calculate the three-dimensional coordinates of the radar receiving antenna phase center in the horizontal inertial frame. The radar electric axis direction vector is independently calculated from the azimuth and elevation scan results. The vectors are then combined and weighted with the inverse of the variance of the fitted residual as the weight and normalized to obtain the final absolute direction vector of the radar electric axis.

8. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV electro-optical radar pod according to claim 7, characterized in that: Radar electric axis peak azimuth angle The expression is as follows: ; In the formula: This represents the peak azimuth angle of the radar electrical axis extracted by interpolation; This represents the azimuth angle corresponding to the point of maximum power in the discrete sampling sequence; Indicates the azimuth scan step angle; The first The received power at each sampling point; Radar electric axis peak elevation angle The expression is as follows: ； In the formula: This represents the peak elevation angle of the radar electrical axis extracted by interpolation; This represents the pitch angle corresponding to the point of maximum power in the discrete sampling sequence; Indicates the pitch scan step angle; The first The received power at each sampling point; The three-dimensional coordinates of the radar receiving antenna phase center in the horizontal inertial frame are as follows: ； In the formula: This represents the three-dimensional coordinates of the phase center of the radar receiving antenna in the horizontal inertial frame. Represents the three-dimensional coordinates of the center of the chessboard square in a horizontal inertial frame; The rotation matrix representing the target's attitude in the pod's body coordinate system; This represents the offset vector of the calibrated antenna phase center from the origin of the target's local coordinate system.

9. The method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod according to claim 1, characterized in that: Step S6 specifically includes the following sub-steps: S61. In a horizontal inertial frame, based on the absolute direction vector of the radar's electric axis and the unit direction vector of the infrared optical axis, calculate the true azimuth installation deviation and the true elevation installation deviation of the radar relative to the optical reference. S62. Fit the IMU attitude measurement error, PnP calculation error, antenna offset calibration error, and parabolic interpolation fitting error using the covariance propagation law, incorporate them into the full parameter error budget model, and synthesize the total covariance matrix. S63. Based on actual installation deviations, sensor offsets, and beam spread correction factors, establish pixel-level spatial positional relationships for visible light-infrared, laser-infrared, and radar-infrared, and complete the construction of a unified model for the four-source spatial reference. S64. Adopting a ground calibration and air verification architecture, the ground calibration parameters are called to calculate the coarse registration offset, and the validity of the calibration result is determined based on the residual pixel deviation and the 3σ confidence interval.

10. A method for unifying the pixel-level reference of four-source sensors in a short-range UAV optoelectronic radar pod according to claim 9, characterized in that: The actual azimuth installation deviation and the actual elevation installation deviation are expressed as follows: ； ； In the formula: This indicates the actual azimuth installation deviation of the radar's electrical axis relative to the infrared optical axis; These represent the components of the radar's final unit vector along the X and Y axes in the horizontal inertial frame, respectively. These represent the components of the infrared optical axis unit vector along the X and Y axes in the horizontal inertial frame, respectively. This indicates the actual pitch installation deviation of the radar electrical axis relative to the infrared optical axis; This represents the component of the final unit vector of the radar electric axis along the Z-axis of the horizontal inertial frame; This represents the component of the infrared optical axis unit vector along the Z-axis of the horizontal inertial frame; The total covariance matrix is ​​as follows: ； In the formula: The 2×2 covariance matrix represents the final azimuth-pitch deviation; Represents a 2×M Jacobian matrix with elements of . , This represents the M independent error input variables that affect the deviation calculation; Represents the M×M total covariance matrix with all parameters; This represents the total number of independent error sources that participate in error propagation; The unified model for four-source space references is as follows: ； In the formula: Indicates the optimal calibration distance; These represent the horizontal and vertical physical offsets of visible light relative to infrared light at the optimal calibration distance, respectively. These represent the horizontal and vertical physical offsets of the laser relative to the infrared at the optimal calibration distance, respectively. These represent the horizontal and vertical physical offsets of the radar relative to the infrared at the optimal calibration distance, respectively. These represent the projected coordinates of the visible light axis at the optimal calibration distance; These represent the projected coordinates of the laser optical axis at the optimal calibration distance; These represent the projected coordinates of the infrared optical axis at the optimal calibration distance, with a value of 0; , These represent the horizontal and vertical installation tilt angles of the radar's electric axis relative to the pod's mechanical reference, respectively. This represents the beam spread correction factor.

Images