Method and device for positioning non-cooperative unmanned aerial vehicle pilot, electronic equipment and storage medium

Abstract

The application belongs to the technical field of positioning, and discloses an airborne non-cooperative unmanned aerial vehicle pilot positioning method and device, electronic equipment and a storage medium, which mainly comprises the following steps: mapping real-time pose data from a three-dimensional geographic coordinate system to a local station-centered coordinate system; using a multi-level observation data verification mechanism to verify and filter a plurality of azimuth observation sequences; using a triple joint weighted Gauss-Newton iteration to solve the effective observation sequence to obtain a maximum likelihood estimation coordinate and a weighted normal equation matrix; extracting an effective residual and dynamically reconstructing a measurement variance to generate a spatial covariance matrix in combination with the normal equation matrix; driving a Kalman filter with the maximum likelihood estimation coordinate and the spatial covariance matrix to perform optimal state posterior correction to obtain an optimal state estimation coordinate; and performing error confidence boundary reverse projection with a geometric angle penalty to obtain an optimal state estimation coordinate and an error polygon boundary point in the three-dimensional geographic coordinate system.

Description

Technical Field

[0001] This invention relates to the field of positioning technology, and in particular to a method, device, electronic equipment, and storage medium for locating airborne non-cooperative drone pilots. Background Technology

[0002] In recent years, with the gradual opening of low-altitude airspace, drone technology has been widely used in various scenarios. However, unauthorized, non-cooperative drone flights pose a serious threat to important classified units and public safety. To effectively counter this, high-precision passive positioning of ground-based, uncooperative drone pilots has become a core technological requirement in the field of low-altitude security.

[0003] Currently, passive direction finding and positioning for non-cooperative drone pilots typically involves using an airborne platform equipped with radio direction finding equipment. The drone maneuvers to acquire multiple frames of azimuth observation data, which are then used to calculate the target's position using a cross-location algorithm. However, in actual deployments, airborne reconnaissance platforms often face complex urban operating environments. Dense buildings, large metal structures, and complex electromagnetic backgrounds in cities lead to severe signal blockage and multipath effects. The reflection and refraction of radio waves during propagation inevitably introduce large-scale anomalous observation errors into the direction finding data acquired by the airborne equipment.

[0004] Specifically, under the dual challenges of high-maneuverability flight on airborne platforms and complex urban environments, existing airborne positioning solutions generally suffer from the following significant drawbacks: 1. Traditional cross-location algorithms fail to account for spatiotemporal dynamic differences, resulting in large positioning errors. Existing technologies mostly employ traditional cross-location algorithms such as pseudo-linear least squares for position calculation. However, due to the high-speed movement of airborne platforms, the signal-to-noise ratio of long-distance observations deteriorates sharply, and outdated observation data has very little indicative significance for the current target state. This linearized processing method, lacking spatial distance decay and temporal forgetting mechanisms, leads to severe interference from historical anomalies in the positioning results. Especially under long baseline conditions, the error will be amplified nonlinearly, ultimately resulting in extremely slow positioning convergence or even divergence.

[0005] 2. The lack of robust handling mechanisms for observational "gross errors" in complex environments and the failure to effectively utilize outliers result in poor robustness. As mentioned earlier, positioning in urban environments is extremely difficult. Existing solution systems either directly input these raw measurement data or directly remove large-deviation measurement outliers, lacking dynamic closed-loop anomaly detection and filtering mechanisms, and failing to effectively utilize such outlier observations. Furthermore, under harsh geometric topologies with small observation angles, the baseline configuration of multi-frame observations is poor. Traditional methods lack configuration detection of linear correlations in the column vectors of the underlying Jacobian matrix and matrix singularity avoidance, leading to ill-conditioned degradation or singularities in the normal equation matrix, which directly causes numerical computation collapse and loss of positioning trajectory.

[0006] 3. Independent Gaussian-Newton solutions lack kinematic smoothing between adjacent frames, failing to reflect the pilot's true micro-movements. In existing technologies, some high-precision solutions employ the Gaussian-Newton method for frame-by-frame independent nonlinear solutions. This approach treats the positioning in each observation cycle as an isolated static optimization problem, ignoring the temporal continuity and kinematic characteristics of the target pilot in the physical world. Furthermore, due to the lack of cross-frame cascaded state estimation and kinematic smoothing mechanisms, the coordinates output from each independent solution are extremely sensitive to environmental noise, resulting in significant high-frequency jumps and oscillations in the target trajectory in the time domain. This not only masks the target's true micro-movement trajectory but also prevents downstream early warning and guidance systems from obtaining stable target velocity and heading parameters.

[0007] 4. The lack of high-confidence error indications based on actual measurement variance and real-time geometric penalties can easily mislead tactical decisions. When presenting results, some conventional positioning terminals rely solely on the fixed direction-finding variance specified by the sensor manufacturer, drawing a static error circle around the estimated point. While some higher-precision solutions introduce Cramer-Rao lower bound (CRLB) to calculate the error ellipse, they often rely solely on the inversion of the underlying normal equation matrix. This purely theoretical matrix calculation has significant shortcomings when facing actual airborne reconnaissance: when the observation angle is extremely small (i.e., poor baseline configuration), a simple CRLB often fails to adequately reflect the huge error stretching caused by such extreme geometric degradation. This error expression method, detached from real-world interference noise and lacking specific directional geometric penalties, can lead to falsely high-confidence outputs (i.e., smaller error ellipses) under adverse observation configurations. This makes it impossible for operators to accurately assess the true reliability of the current positioning coordinates, easily misleading commanders into making incorrect tactical judgments.

[0008] In summary, existing technologies have systematic deficiencies in areas such as optimization of the underlying solution engine, robust utilization of anomalous observation information, cross-frame kinematic cascade state estimation, and calculation of dynamic confidence error boundaries. Summary of the Invention

[0009] The purpose of this invention is to provide a method, device, electronic device, and storage medium for locating non-cooperative drone pilots on airborne platforms, in order to solve the problem of high-precision and robust tracking of non-cooperative drone pilots in complex urban environments and under high-maneuverability flight conditions.

[0010] The technical solution adopted by this invention to solve its technical problem is as follows: In a first aspect, the present invention provides a method for locating airborne non-cooperative drone pilots, comprising the following steps: Acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data from the airborne platform; Mapping real-time pose data from a 3D geographic coordinate system to a local station center coordinate system; A multi-level observation data verification mechanism was used to verify and filter out multiple frames of azimuth observation sequences to obtain valid observation sequences. The effective observation sequence is solved by using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and weighted normal equation matrix; The effective residuals are extracted and the measurement variance is dynamically reconstructed. The spatial covariance matrix is ​​then generated by combining the normal equation matrix. The optimal state estimated coordinates are obtained by driving the Kalman filter with the maximum likelihood estimated coordinates and the spatial covariance matrix to perform optimal state posterior correction. By performing inverse projection of the error confidence boundary with geometric angle penalty, the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system are obtained.

[0011] In some embodiments, the multi-level observation data verification mechanism refers to a three-level observation data verification mechanism, which includes information threshold access and state reset, dynamic filtering of historical windows, and geometric angle degradation detection. The method of using a multi-level observation data verification mechanism to verify and filter multi-frame azimuth observation sequences includes the following steps: During the initial convergence period, calculate the expected azimuth angle of the current airborne platform position relative to the non-cooperative drone pilot, and calculate the information residual between the observed azimuth angle and the expected azimuth angle; The absolute value of the innovation residual is taken to obtain the absolute residual. It is then determined whether the absolute residual is greater than the innovation interception angle threshold. If so, the current frame data is determined to be a gross error and isolated. If the number of consecutive frames containing gross errors exceeds the consecutive frame drop tolerance threshold, the Kalman filter state is reset; otherwise... The corresponding frame data is allowed to pass and treated as valid observation data. It is then pushed into a sliding time window based on the first-in, first-out principle. During the steady-state tracking period, the sliding window is backtracked and it is determined whether the absolute residual of a historical observation point within the window is greater than the historical cleaning angle threshold. If so, abnormal observations are filtered out; otherwise, During the cold start initialization phase, the range angle of all observed azimuth angles within the sliding window is statistically analyzed, and it is determined whether the range angle is lower than the minimum geometric angle threshold to avoid baseline degradation. If so, it is determined that the spatial geometry configuration has degraded. In this case, subsequent solution updates are blocked, and the airborne platform is instructed to maintain its current state and increase the geometric angle of the paper orbital flight. Otherwise, Perform subsequent solution updates.

[0012] In some embodiments, a first tracking threshold is set when the information threshold is admitted and the state is reset, and a second tracking threshold is set when the history window is dynamically filtered out. The second tracking threshold is greater than the first tracking threshold. During the initial convergence period, if the number of historical stable tracking frames is greater than the first tracking threshold and less than the second tracking threshold, then: using the prior predicted coordinates fed back by the Kalman filter, calculate the expected azimuth angle of the current airborne platform position for the non-cooperative drone pilot. During steady-state tracking, if the number of historical stable tracking frames is greater than the second threshold, then: backtrack through the sliding window.

[0013] In some embodiments, prior to solving the effective observation sequence using triple-weighted Gauss-Newton iteration, the method further includes: For the valid observation sequence obtained after verification and filtering, a pseudo-linear overdetermined system of equations is constructed using the data within the valid window, and the initial coordinates are obtained by analytical solution; Using the initial coordinates as a reference, initiate a triple-weighted Gauss-Newton iteration.

[0014] In some embodiments, initiating a triple-weighted Gauss-Newton iteration includes: A weighted normal equation matrix is ​​constructed by integrating space physics slant range weighting, time forgetting factor weighting, and Gaussian robustness weighting. The coordinates are estimated by approximating the maximum likelihood using the Gauss-Newton iteration.

[0015] In some embodiments, the step of driving the Kalman filter to perform optimal state posterior correction using the maximum likelihood estimated coordinates and the spatial covariance matrix to obtain the optimal state estimated coordinates means: The Kalman filter uses the maximum likelihood estimated coordinates as virtual measurement input, the spatial covariance matrix as the measurement noise matrix, and performs the following standard filtering loop: Perform prior state prediction and covariance prediction to complete prior deduction; Perform Kalman optimal gain calculation; Calculate the new observation information and perform posterior state correction and posterior covariance correction to complete the measurement update.

[0016] In some embodiments, the step of performing inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system includes the following steps: Perform eigenvalue decomposition on the spatial covariance matrix and calculate the theoretical semi-axis length of the error ellipse; Extract the maximum physical subtended angle formed by all observed rays and introduce a penalty mechanism based on the geometric subtended angle penalty factor; If the maximum physical angle is less than the set penalty trigger angle threshold, then calculate the penalty amplification factor; For observation rays whose maximum physical angle is less than the set penalty trigger angle threshold, a penalty amplification factor is applied, and a graphics rendering limit protection threshold is set to obtain the semi-axis length after error ellipse correction. Calculate the rotation angle of the principal axis of the error ellipse, discretize the ellipse into a polygonal point matrix with a set number of sides at equal intervals, and generate the coordinates of the boundary points of the error polygon in the local stationary coordinate system. Invoke the reverse mapping transformation sequence to map the target smooth estimate corresponding to the optimal state estimate coordinates and the error polygon boundary points corresponding to the error polygon boundary point coordinates to the three-dimensional geographic coordinate system.

[0017] Secondly, the present invention also provides an airborne non-cooperative drone pilot positioning device, comprising: The data acquisition module is configured to acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data from the airborne platform; The coordinate system mapping module is configured to map real-time pose data from a 3D geographic coordinate system to a local station center coordinate system; The data verification and filtering module is configured to use a multi-level observation data verification mechanism to verify and filter multiple frames of azimuth observation sequences to obtain valid observation sequences. The iterative solution module is configured to solve the effective observation sequence using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and the weighted normal equation matrix; The measurement variance assessment and spatial covariance derivation module is configured to extract effective residuals and dynamically reconstruct the measurement variance, and generate a spatial covariance matrix by combining the normal equation matrix. The cascaded state smoothing filter module is configured to drive the Kalman filter to perform optimal state posterior correction using the maximum likelihood estimated coordinates and the spatial covariance matrix, so as to obtain the optimal state estimated coordinates. The error confidence boundary indication module based on adaptive penalty is configured to perform inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system.

[0018] Thirdly, the present invention also provides an electronic device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the airborne non-cooperative UAV pilot positioning method.

[0019] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the steps of the airborne non-cooperative UAV pilot positioning method.

[0020] The beneficial effects of this invention are as follows: First, unlike traditional algorithms that directly solve in a three-dimensional geographic coordinate system, this invention maps real-time pose data from a three-dimensional geographic coordinate system to a local station-centered coordinate system to establish a subsequent solution space without geometric projection distortion, thereby improving the accuracy of the positioning solution. Second, this invention reduces the probability of ill-conditioned degradation of the normal equation matrix through a multi-level observation data verification mechanism, thereby improving the overall robustness of the positioning algorithm. Third, this invention solves the effective observation sequence through triple joint weighted Gauss-Newton iteration and drives the Kalman filter with the maximum likelihood estimated coordinates and spatial covariance matrix for optimal state posterior correction, obtaining the optimal state estimated coordinates. Ultimately, it can obtain a more stable and continuous estimated trajectory for non-cooperative UAV pilots and improve the robustness of the positioning method in multipath interference environments. Finally, by back-projecting the error confidence boundary with geometric angle penalty, it can avoid the problem of easily outputting small error boundaries under adverse observation configurations such as small angles, thereby improving the confidence of the current positioning results. Attached Figure Description

[0021] Figure 1 A flowchart illustrating an example embodiment of the airborne non-cooperative drone pilot localization method of the present invention; Figure 2 This is a flowchart of multi-level dynamic observation verification and nonlinear weighted solution in one example embodiment of the present invention; Figure 3 The flowchart illustrates variance evaluation, cascaded filtering, and confidence boundary in one exemplary embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] See Figure 1 In one example implementation, this embodiment provides an airborne non-cooperative drone pilot positioning method. After continuously acquiring the azimuth measurement data of the non-cooperative drone pilot using radio reconnaissance or direction-finding equipment mounted on the drone platform, and simultaneously acquiring the real-time pose data of the airborne platform itself, the core positioning process can be divided into two collaborative stages: "front-end nonlinear spatial robustness calculation" and "back-end dynamic state smoothing and boundary characterization." Therefore, the positioning method of this embodiment, in its specific implementation, may include the following steps: S1. Acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data of the airborne platform; S2. Map the real-time pose data from the three-dimensional geographic coordinate system to the local station center coordinate system; S3. Use a multi-level observation data verification mechanism to verify and filter out multiple frames of azimuth observation sequences to obtain valid observation sequences; S4. Solve the effective observation sequence using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and weighted normal equation matrix; S5. Extract the effective residuals and dynamically reconstruct the measurement variance, and generate the spatial covariance matrix by combining the normal equation matrix; S6. Drive the Kalman filter with the maximum likelihood estimated coordinates and the spatial covariance matrix to perform optimal state posterior correction, and obtain the optimal state estimated coordinates. S7. Perform inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system.

[0024] It should be noted that, in the above-described positioning method of this embodiment, after acquiring the multi-frame azimuth observation sequence of the non-cooperative UAV pilot and the real-time pose data of the airborne platform (step S1), the optimal state estimation coordinates and error boundaries of the non-cooperative UAV pilot are finally obtained through the processing in steps S2 to S7. Specifically, the specific implementation steps of the above positioning method are as follows: S1. Acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data of the airborne platform.

[0025] In this embodiment, the airborne platform is It continuously acquires its real-time pose data in a three-dimensional geographic coordinate system (WGS84), including: latitude. ,longitude Elevation .

[0026] S2. Map the real-time pose data from the three-dimensional geographic coordinate system to the local station center coordinate system to eliminate long-distance projection distortion.

[0027] To eliminate the two-dimensional plane projection error introduced by the Earth's curvature in long-distance long-baseline positioning, it is necessary to establish a relay transformation space using the standard constant of the Earth ellipsoid. In this step, the WGS84 geographic coordinates are first converted to geocentric rectangular coordinates (ECEF). ,in: ; ; ; ; in, The radius of curvature of the circle is denoted as . For the Earth's semi-major axis, It is the square of the first eccentricity.

[0028] Subsequently, a priori physical constraint was introduced to ensure the target's proximity to the Earth's surface. Upon the initial acquisition of a valid observation, the onboard geographic coordinates were locked, and a reference elevation was set on the ground. Establish a local tangent plane as the absolute reference origin of the local station-centered coordinate system (ENU). Its corresponding ECEF coordinates are Here, the three-dimensional coordinates of the current airborne platform in the station-centered coordinate system. (Representing distances to the east, north, and sky respectively) are calculated using the following rotation matrix: ; Therefore, through step S2, the subsequent nonlinear solutions are all reduced in dimension and isolated in the two-dimensional plane of the station center system.

[0029] Compared to directly performing local linearization approximation in the latitude and longitude coordinate system, the coordinate mapping method in this embodiment can effectively avoid large-scale planar projection distortion caused by the curvature of the Earth, ensuring the accuracy of spatial geometric relationships under deep observation conditions, thereby improving the accuracy of positioning calculation.

[0030] S3. The multi-level observation data verification mechanism is used to verify and filter out multiple azimuth observation sequences to obtain valid observation sequences.

[0031] See Figure 2 To address the multipath interference and mobile aircraft movement characteristics in complex urban environments, this embodiment constructs a three-level judgment mechanism that spans the entire observation sequence period: Level 1 Judgment (New Information Threshold Access and Status Reset): After initializing the convergence period (setting the judgment threshold to the number of historically stable tracking frames), ,in (As the first tracking threshold), using the prior predicted coordinates fed back by the Kalman filter. Calculate the expected azimuth angle of the current airborne position relative to the target. And calculate the information residual between the observed azimuth and the desired azimuth. (through (Phase unwinding process).

[0032] If absolute residual ( (For the set information interception angle threshold), determine when the frame data is gross and isolate it; if the number of consecutive rejected frames is counted... ( If the threshold for continuous frame loss is set, the system determines that the target has undergone large-scale maneuvering or filter divergence. At this time, the filter state is reset, that is, the initialization flag is set to invalid, the filter state is reset, but the historical observation cache is retained.

[0033] Level 2 Judgment (Dynamic Filtering of Historical Windows): The valid observation data is pushed into a sliding time window based on the first-in, first-out (FIFO) principle. At this point, the maximum capacity is set to... To control computing power boundaries, a minimum starting threshold is set as follows: To ensure the stability of the overdetermined equations, when the system is in the steady-state tracking period, i.e. ,in The second steady-state tracking threshold and At that time, backtrack through the sliding window. If the absolute value of the residual at a historical observation point within the window is greater than... , If the historical cleaning angle threshold is used, then anomaly observation filtering operation is performed to ensure the validity of the solution data.

[0034] Level 3 Judgment (Geometric Angle Degradation Detection): During the system cold start initialization phase, the range of all observed azimuth angles within the statistical sliding window is calculated. If the angle ( To avoid baseline degradation (minimum geometric angle threshold), it is determined that the spatial geometry is degraded at this point (the direction finding rays tend to be parallel), which may lead to divergence in subsequent matrix inversion. At this point, subsequent solution updates are blocked, and the airborne platform continues to fly around in the current state until the geometric angle increases.

[0035] Therefore, step S3 in this embodiment differs from the traditional single data filtering mechanism. It constructs a three-level progressive judgment mechanism that includes "predictive information threshold verification", "historical window dynamic secondary filtering" and "geometric angle degradation detection". It isolates abnormal measurement data of different dimensions layer by layer, which greatly reduces the probability of ill-conditioned degradation of the normal equation matrix and greatly improves the overall robustness of the positioning method.

[0036] S4. Solve the effective observation sequence using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and the weighted normal equation matrix.

[0037] See Figure 2 First, to address the problem of lacking points for differentiation and expansion, this embodiment can utilize the data within the effective window to construct a pseudo-linear overdetermined system of equations to extract analytical seed points. The logic for accumulating the elements of the block matrix of the normal equations is as follows: ; ; When its determinant is not zero, the deterministic initial coordinates can be obtained by analytical solution. .

[0038] Subsequently, using the initial coordinates as a reference, a joint triple-weighted Gauss-Newton iteration was initiated.

[0039] In each iteration (with an upper limit set for the number of iterations to balance computing power and accuracy), this embodiment calculates triple physical weights for each observation point within the window: Space Physics Slant Range Weighted ( Considering the slope distance attenuation caused by elevation, let the square of the horizontal distance between the current iteration coordinates and the observation point on the horizontal plane be... Among these measures, a lower limit for numerical protection is established to restrict... Then the three-dimensional slant distance in space In this embodiment, the distance weight can be constructed using the inverse square decay law, i.e., spatial physical slant range weighting: ; in, The standard reference slope distance, This is the lower limit of the tolerance.

[0040] Time-forgetting factor weighted ( This embodiment can use an exponential function to reduce the weight of historical observations, which is expressed as: ; in, The current system time. This is the time memory constant.

[0041] Gaussian robust weighting, i.e. Gaussian M-estimation robust weighting ( This embodiment can first calculate the nonlinear geometric residual. Then, a reduced-weight tolerance kernel based on the Gaussian kernel function is constructed and used as a Gaussian robust weight, which is expressed as: ; in, and The prior standard deviation parameter characterizes the error tolerance boundary.

[0042] Therefore, the comprehensive weight in this embodiment .

[0043] It should be noted that, after obtaining the above comprehensive weights, this embodiment can sequentially calculate the partial derivatives of the column vectors of the Jacobian matrix. and And by accumulating these, a weighted normal equation approximation Hessian matrix is ​​constructed. With gradient vector .

[0044] In addition, this embodiment requires a singularity check before solving the normal equation to update the step size: calculating the matrix determinant. If its absolute value is less than the minimum threshold If the matrix is ​​well-state (non-singular), then the current iteration terminates. If the matrix is ​​well-state (non-singular), then the corrected step size is calculated. And update the target coordinates.

[0045] After the iteration is complete, the maximum likelihood estimated coordinates are output, and the weighted normal equation matrix is ​​retained for reuse in subsequent steps.

[0046] Through step S4 of this embodiment, firstly, unlike existing technologies that typically employ simple direct rejection or inclusion strategies when dealing with "outliers" deviating from the normal range, this embodiment introduces a Gaussian robust kernel function, which elevates the processing of outlier observations to a "weight reduction approach." Therefore, while suppressing large-deviation noise interference, it can maximize the extraction and retention of usable weak observational state information during sudden changes in the target state. Secondly, unlike existing technologies with equal or single-weight processing models, this embodiment proposes a triple comprehensive weighting mechanism combining "spatial physical slant range attenuation," "time forgetting factor," and "error robust distribution" during the solution process. Because it fits the attenuation characteristics of spatial electromagnetic waves and the timeliness of measurement data, it can enable coordinate calculation to actively focus on high-value observational data. Thirdly, unlike the huge nonlinear truncation error caused by the pseudo-linearized equation single-step solution method commonly used in traditional cross-location, this embodiment adopts a Gauss-Newton nonlinear iterative weighted least squares algorithm. With the assistance of analytical initial coordinates, through refined nonlinear iteration, it can effectively approximate the maximum likelihood estimate coordinates of the target spatial parameters.

[0047] S5. Extract the effective residuals and dynamically reconstruct the measurement variance, and generate the spatial covariance matrix by combining the normal equation matrix.

[0048] See Figure 3 In step S5, the most recent sliding time window can be selected (the maximum number of observation points in the window is set to...). Extract the residuals within the effective convergence range to assess the angular variance of the current environment.

[0049] Specifically, the generation of the spatial covariance matrix can be achieved through the following steps: Instantaneous residual calculation and clipping: For each observation point within the window Calculate its coordinates with respect to the current optimal estimate. The posterior residuals are obtained, and phase unwinding is performed: ; in Normalization to The interval is then adjusted. A clipping threshold is then applied. Only retain The data is used for variance evaluation to reduce the interference of gross errors on the Gaussian distribution.

[0050] Time-domain weighted measurement variance estimation: This embodiment uses an exponential decay function to assign higher weights to recent data, and sets the evaluation time constant to be... Construct time-domain weights: ; Calculate the weighted measurement variance in angular space : ; If the number of valid points is insufficient to meet the set minimum evaluation point threshold If the weights are too small, the default system variance will be used. As an alternative value.

[0051] Spatial covariance matrix generation: Based on the Cramer-Rao lower bound (CRLB) theorem, the angular variance is projected onto rectangular coordinate space to generate the observation noise covariance matrix. , which is the spatial covariance matrix, is expressed as: ; Expanding the observation noise covariance matrix into a specific two-dimensional matrix inversion formula, let the elements of the normal equation matrix be represented as: ; And its determinant is The spatial covariance matrix can then be analytically derived as follows: .

[0052] It should be noted that step S5 of this embodiment establishes a low-level data transfer mechanism between "static nonlinear optimization" and "temporal state-space filtering". It not only extracts the maximum likelihood estimation coordinates after Gauss-Newton iteration convergence, but also directly transfers the weighted normal equation approximation Hessian matrix generated by the iteration to the downstream module. Combined with the measurement variance based on the dynamic reconstruction of the sliding time window, and using the Cramer-Rhodes lower bound (CRLB) theorem to project the variance to generate a global spatial covariance matrix, the spatial covariance matrix can synchronously drive the EKF to solve for the optimal filter gain and be directly applied to the generation of the error confidence polygon. Therefore, the data transfer and reuse mechanism mentioned in step S5 of this embodiment can ensure a tight closed loop of cascaded solution while controlling the computational overhead.

[0053] S6. Drive the Kalman filter with the maximum likelihood estimated coordinates and the spatial covariance matrix to perform optimal state posterior correction, and obtain the optimal state estimated coordinates.

[0054] See Figure 3 Since the model simplifies the representation by removing higher-order control variables when the target pilot is on the ground, this embodiment can construct a 4-dimensional state vector. , representing the eastward position, northward position, eastward velocity, and northward velocity, respectively. In this embodiment, the Kalman filter (EKF) can output the following in step S4: As a virtual measurement input, the output of step S5 As a measurement noise array, the following standard filtering loop is executed: State and covariance prediction (prior deduction): Based on a discrete uniform motion model, state transition matrix The formula for predicting prior states is as follows ( (for time steps) ; Setting the process noise matrix The diagonal elements represent the system's position and velocity walk uncertainties, updating the prior covariance. : .

[0055] Kalman optimal gain solution: Since virtual measurements only contain location information, an observation matrix is ​​constructed. In this embodiment, firstly, calculation The new covariance matrix : ; Then, solve Kalman gain matrix : .

[0056] Measurement Update (Posterior Correction): Extract the output of the current Gauss-Newton iteration to construct the measurement vector. Calculate observational information : ; Finally, the corrections for the system's posterior state and posterior covariance are completed, and they are expressed as follows: ; ; at this time, The first two components are the optimal state estimation coordinates of the final output. .

[0057] It should be noted that step S6 in this embodiment differs from traditional filters that use random initial points or directly use high-frequency fluctuating original direction-finding angles as input. This embodiment can use the high-precision coordinates obtained by the Gauss-Newton algorithm after precise iteration as the virtual measurement input of EKF. Therefore, it achieves physical isolation between the underlying original direction-finding noise and the macroscopic state update, avoiding the direct contamination of the filter covariance matrix by invalid observation noise, thereby obtaining a more stable and continuous non-cooperative UAV pilot estimated trajectory. In addition, unlike traditional algorithms that rely on static constant variance, this embodiment evaluates the optimal window period of recent environmental real measurement noise and calculates the variance by statistically analyzing the residuals within effective convergence, eliminating the interference of out-of-limit gross errors on the variance distribution. Therefore, the measurement variance calculated by the system can be highly sensitive to the current environmental interference changes, thereby significantly improving the accuracy of the EKF gain allocation mechanism and the final filtering accuracy.

[0058] S7. Perform inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system.

[0059] See Figure 3 To provide an intuitive error confidence interval, this embodiment needs to extract the spatial covariance matrix and convert it into a discrete polygon point matrix. Specifically, step S7 of this embodiment may include the following: Eigenvalue decomposition for principal axis parameter extraction: Solving the characteristic equation of the covariance matrix yields the eigenvalues ​​related to the major and minor semi-axes of a two-dimensional Gaussian distribution on a horizontal plane. , : ; Set scaling factor Calculate the theoretical semi-axis length of the error ellipse by mapping a specific probability confidence interval: ; .

[0060] GDOP Adaptive Penalty Based on Local Physical Angle: Dynamically Extracting the Maximum Physical Angle Formed by All Observed Rays To prevent spurious convergence of the error boundary under narrow opening angle conditions, this embodiment introduces a penalty mechanism based on GDOP, namely: when ( When the set penalty trigger angle threshold is reached, calculate the penalty amplification factor. (Set the minimum opening angle lower limit as) ): ; Apply penalties and set a graphics rendering limit protection threshold. The corrected half-axis is obtained: ; .

[0061] Inverse projection and discretization of coordinate system: Calculating the rotation angle of the principal axis of the error ellipse : ; Discretize the ellipse at equal intervals into a set number of sides. Polygonal lattice (index) ), generating polygon boundary point coordinates in the local station-centered coordinate system : ; ; Finally, the reverse mapping transformation sequence is called ( The target smooth trajectory composed of the optimal state estimation coordinates and this Each error polygon boundary point is fully mapped back to the three-dimensional geographic coordinate system and output to the front end for situation display.

[0062] Therefore, unlike traditional error boundaries that rely solely on theoretical CRLB calculations and are prone to outputting smaller error boundaries under adverse observation configurations such as small opening angles, this embodiment can apply GDOP penalty to the spatial covariance extracted by CRLB in step S7 for small opening angle cases, correcting the error indication boundary under spatial configuration degradation, thereby improving the authenticity and reliability of the final tactical confidence score output by the system.

[0063] In another example implementation, this disclosure also provides an airborne non-cooperative drone pilot positioning device, comprising: The data acquisition module is configured to acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data from the airborne platform; The coordinate system mapping module is configured to map real-time pose data from a 3D geographic coordinate system to a local station center coordinate system; The data verification and filtering module is configured to use a multi-level observation data verification mechanism to verify and filter multiple frames of azimuth observation sequences to obtain valid observation sequences. The iterative solution module is configured to solve the effective observation sequence using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and the weighted normal equation matrix; The measurement variance assessment and spatial covariance derivation module is configured to extract effective residuals and dynamically reconstruct the measurement variance, and generate a spatial covariance matrix by combining the normal equation matrix. The cascaded state smoothing filter module is configured to drive the Kalman filter to perform optimal state posterior correction using the maximum likelihood estimated coordinates and the spatial covariance matrix, so as to obtain the optimal state estimated coordinates. The error confidence boundary indication module based on adaptive penalty is configured to perform inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system.

[0064] In another example implementation, this disclosure also provides an electronic device including a processor, a memory, and a bus, the memory storing machine-readable instructions executable by the processor, wherein when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of an airborne non-cooperative drone pilot localization method.

[0065] In another example implementation, this disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the airborne non-cooperative drone pilot localization method.

[0066] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for locating airborne non-cooperative drone pilots, characterized in that, Includes the following steps: Acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data from the airborne platform; Mapping real-time pose data from a 3D geographic coordinate system to a local station center coordinate system; A multi-level observation data verification mechanism was used to verify and filter out multiple frames of azimuth observation sequences to obtain valid observation sequences. The effective observation sequence is solved by using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and weighted normal equation matrix; The effective residuals are extracted and the measurement variance is dynamically reconstructed. The spatial covariance matrix is ​​then generated by combining the normal equation matrix. The optimal state estimated coordinates are obtained by driving the Kalman filter with the maximum likelihood estimated coordinates and the spatial covariance matrix to perform optimal state posterior correction. By performing inverse projection of the error confidence boundary with geometric angle penalty, the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system are obtained.

2. The method for locating airborne non-cooperative UAV pilots according to claim 1, characterized in that, The multi-level observation data verification mechanism refers to a three-level observation data verification mechanism, which includes information threshold access and state reset, dynamic filtering of historical windows, and geometric angle degradation detection. The method of using a multi-level observation data verification mechanism to verify and filter multi-frame azimuth observation sequences includes the following steps: During the initial convergence period, calculate the expected azimuth angle of the current airborne platform position relative to the non-cooperative drone pilot, and calculate the information residual between the observed azimuth angle and the expected azimuth angle; The absolute value of the innovation residual is taken to obtain the absolute residual. It is then determined whether the absolute residual is greater than the innovation interception angle threshold. If so, the current frame data is determined to be a gross error and isolated. If the number of consecutive frames containing gross errors exceeds the consecutive frame drop tolerance threshold, the Kalman filter state is reset; otherwise... Release the corresponding frame data and treat it as valid observation data. Push it into a sliding time window based on the first-in-first-out principle. During the steady-state tracking period, backtrack through the sliding window and determine whether the absolute residual of a certain historical observation point in the window is greater than the historical cleaning angle threshold. If so, filter out abnormal observation values. otherwise, During the cold start initialization phase, the range angle of all observed azimuth angles within the sliding window is statistically analyzed, and it is determined whether the range angle is lower than the minimum geometric angle threshold to avoid baseline degradation. If so, it is determined that the spatial geometry configuration has degraded. In this case, subsequent solution updates are blocked, and the airborne platform is instructed to maintain its current state and increase the geometric angle of the paper orbital flight. Otherwise, Perform subsequent solution updates.

3. The method for locating airborne non-cooperative UAV pilots according to claim 2, characterized in that, When the new information threshold is used for admission and the status is reset, a first tracking threshold is set. When the history window is dynamically filtered, a second tracking threshold is set. The second tracking threshold is greater than the first tracking threshold. During the initial convergence period, if the number of historical stable tracking frames is greater than the first tracking threshold and less than the second tracking threshold, then: using the prior predicted coordinates fed back by the Kalman filter, calculate the expected azimuth angle of the current airborne platform position for the non-cooperative drone pilot. During steady-state tracking, if the number of historical stable tracking frames is greater than the second threshold, then: backtrack through the sliding window.

4. The method for locating airborne non-cooperative UAV pilots according to claim 1, characterized in that, Before solving the effective observation sequence using the triple joint weighted Gauss-Newton iteration, the following is also included: For the valid observation sequence obtained after verification and filtering, a pseudo-linear overdetermined system of equations is constructed using the data within the valid window, and the initial coordinates are obtained by analytical solution; Using the initial coordinates as a reference, initiate a triple-weighted Gauss-Newton iteration.

5. The method for locating airborne non-cooperative UAV pilots according to claim 4, characterized in that, When initiating the triple joint weighted Gauss-Newton iteration, the following is included: A weighted normal equation matrix is ​​constructed by integrating space physics slant range weighting, time forgetting factor weighting, and Gaussian robustness weighting. The coordinates are estimated by approximating the maximum likelihood using the Gauss-Newton iteration.

6. The method for locating airborne non-cooperative UAV pilots according to claim 1, characterized in that, The phrase "using the maximum likelihood estimated coordinates and the spatial covariance matrix to drive the Kalman filter for optimal state posterior correction, thereby obtaining the optimal state estimated coordinates" refers to: The Kalman filter uses the maximum likelihood estimated coordinates as virtual measurement input, the spatial covariance matrix as the measurement noise matrix, and performs the following standard filtering loop: Perform prior state prediction and covariance prediction to complete prior deduction; Perform Kalman optimal gain calculation; Calculate the new observation information and perform posterior state correction and posterior covariance correction to complete the measurement update.

7. The method for locating airborne non-cooperative UAV pilots according to any one of claims 1-6, characterized in that, The step of performing inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system includes the following steps: Perform eigenvalue decomposition on the spatial covariance matrix and calculate the theoretical semi-axis length of the error ellipse; Extract the maximum physical subtended angle formed by all observed rays and introduce a penalty mechanism based on the geometric subtended angle penalty factor; If the maximum physical angle is less than the set penalty trigger angle threshold, then calculate the penalty amplification factor; For observation rays whose maximum physical angle is less than the set penalty trigger angle threshold, a penalty amplification factor is applied, and a graphics rendering limit protection threshold is set to obtain the semi-axis length after error ellipse correction. Calculate the rotation angle of the principal axis of the error ellipse, discretize the ellipse into a polygonal point matrix with a set number of sides at equal intervals, and generate the coordinates of the boundary points of the error polygon in the local stationary coordinate system. Invoke the reverse mapping transformation sequence to map the target smooth estimate corresponding to the optimal state estimate coordinates and the error polygon boundary points corresponding to the error polygon boundary point coordinates to the three-dimensional geographic coordinate system.

8. An airborne non-cooperative drone pilot positioning device, characterized in that, include: The data acquisition module is configured to acquire multi-frame azimuth observation sequences from non-cooperative drone pilots and real-time pose data from the airborne platform; The coordinate system mapping module is configured to map real-time pose data from a 3D geographic coordinate system to a local station center coordinate system; The data verification and filtering module is configured to use a multi-level observation data verification mechanism to verify and filter multiple frames of azimuth observation sequences to obtain valid observation sequences. The iterative solution module is configured to solve the effective observation sequence using triple joint weighted Gauss-Newton iteration to obtain the maximum likelihood estimation coordinates and the weighted normal equation matrix; The measurement variance assessment and spatial covariance derivation module is configured to extract effective residuals and dynamically reconstruct the measurement variance, and generate a spatial covariance matrix by combining the normal equation matrix. The cascaded state smoothing filter module is configured to drive the Kalman filter to perform optimal state posterior correction using the maximum likelihood estimated coordinates and the spatial covariance matrix, so as to obtain the optimal state estimated coordinates. The error confidence boundary indication module based on adaptive penalty is configured to perform inverse projection of the error confidence boundary with geometric angle penalty to obtain the optimal state estimation coordinates and error polygon boundary points in the three-dimensional geographic coordinate system.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the airborne non-cooperative UAV pilot localization method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the airborne non-cooperative UAV pilot positioning method as described in any one of claims 1-7.

Images