Urban low-altitude unmanned aerial vehicle trajectory fusion method and system based on multi-source sensing data

Abstract

The embodiment of the invention provides an urban low-altitude unmanned aerial vehicle trajectory fusion method and system based on multi-source sensing data. According to the method, a data preprocessing method based on motion models is provided, and a uniform-speed motion model, a uniform-acceleration motion model and a steering motion model are connected in parallel in each time window. Evaluating and selecting an optimal model according to the result; and carrying out interpolation on the optimal value, and eliminating abnormal values through a confidence interval. According to the multi-source perception data set state, a conditional Kalman filtering strategy based on space and time thresholds is provided to define unmanned aerial vehicle track continuity characteristics of different quality data sets. The working principle of a multi-source sensing heterogeneous device is analyzed, positioning data characteristics and error characteristics of the device on horizontal latitude and longitude and vertical height are analyzed, and a multi-feature-column adaptive weighted fusion multi-source heterogeneous sensing data method is provided. In this way, the sensing positioning error of the unmanned aerial vehicle is reduced by fusing the advantage characteristics of different sensing data.

Description

Technical Field

[0001] This application relates to the field of multi-sensor data fusion technology, and in particular to a method and system for fusion of urban low-altitude unmanned aerial vehicle trajectories based on multi-source sensing data. Background Technology

[0002] Establishing a safety management and control system for low-altitude areas involves more than just target detection. It requires obtaining continuous, reliable, and high-precision three-dimensional trajectories in real time from multi-source heterogeneous observations. This necessitates achieving wide-area coverage and high refresh rate in the horizontal direction, maintaining stable accuracy in the vertical direction, and preserving track continuity and consistency even when target signals are obstructed or certain sensors fail. This requirement dictates that collaborative and fused sensing using multi-source heterogeneous sensors is the core technological path for low-altitude safety management and control.

[0003] 3D trajectory fusion technology primarily relies on multi-sensor data fusion to compensate for the limitations of single devices. Typical solutions include: Time Difference of Arrival (TDOA) positioning systems, which calculate the time difference of arrival of signals using distributed receiving stations to determine the target's location; and Remote ID devices, which resolve the target's real-time location. These two systems differ in their capabilities: TDOA is more advantageous for continuous horizontal (XY) tracking and wide-area coverage, while Remote ID offers higher reliability in identification and direct location reporting, but is constrained by factors such as loading and compliance conditions, coverage environment, and channel quality. The independently generated trajectory data from both systems are complementary in time and space, providing a foundation for building multi-source surveillance networks.

[0004] However, ensuring real-time performance while maintaining high 3D accuracy remains the primary bottleneck in multi-source fusion. Observations from different sensors exhibit significant time delay differences and varying uncertainty characteristics, especially when UAV maneuvers are drastic or signals are blocked, leading to a sharp decline in correlation reliability. Research needs to introduce trajectory prediction compensation, multi-layer thresholding, and adaptive threshold adjustment strategies to achieve highly reliable spatiotemporal alignment in asynchronous measurements. Furthermore, during multi-source fusion, the reliability of different sensors changes with environmental variations and signal quality fluctuations. How to adjust the horizontal / vertical weighting coefficients in real-time based on quantitative measures such as observation covariance or information entropy to maximize fusion accuracy and system robustness remains a challenging problem that urgently needs to be solved. Summary of the Invention

[0005] This application provides a method and system for fusion of urban low-altitude unmanned aerial vehicle (UAV) trajectories based on multi-source sensing data, which can solve the technical problem of poor multi-source trajectory data fusion capability in related technologies.

[0006] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0007] Firstly, embodiments of this application provide a method for fusion of urban low-altitude UAV trajectories based on multi-source sensing data. This method includes: performing bidirectional completion operations based on a motion model for missing moments using a unified time axis to obtain completion results; the unified time axis is used to characterize the relationship between remote identification and arrival time difference; performing a prediction scoring operation on the completion results based on at least one of a constant velocity motion model, a constant acceleration motion model, and a constant steering motion model to obtain state prediction results; removing outlier data from the state prediction results based on the confidence interval of the position difference between the two sources to obtain the data to be observed; and limiting the candidate observation set by thresholding the observation timestamps using time and altitude thresholds. Finally, a cost matrix is ​​constructed based on the horizontal residuals of the predicted and observed data to solve for global matching, obtaining matched and unmatched observation pairs. The matched and unmatched observation pairs are input into a Conditionally Extended Kalman Filter (CEKF). CEKF includes three update branches: RemoteID only, TDOA only, and RemoteID and TDOA matched observation pairs. It adaptively selects the update path based on a Mahalanobis distance threshold, performing single-source updates for unmatched observation pairs to maintain trajectory continuity and reliability. In the fusion phase, the 3D observations and covariance are decomposed into horizontal and vertical subspaces, respectively, and covariance-consistent fusion is performed, using the covariance gain of each information source data as the adaptive weighting criterion. In the vertical subspace, variance inflation is applied to TDOA based on the residuals to stabilize the vertical estimation. Variance inflation in the vertical subspace is primarily based on RemoteID and timeliness-related to the broadcast frequency and residuals to stabilize the vertical estimation. Furthermore, the amplitude and direction of the fusion residual are fed back to the correlation and fusion layer, and the time, height threshold, and horizontal and vertical subspace weights are adjusted online to form a closed loop. When the information gain is insufficient or the measurement geometry degrades, the fusion weight of the information source data is automatically adjusted, and finally the three-dimensional fusion trajectory of the target and its uncertainty are output.

[0008] Based on the above description of the urban low-altitude UAV trajectory fusion method based on multi-source sensing data provided in this application embodiment, it can be seen that this method utilizes the complementarity of two types of sensor information to reduce the positioning error of a single data source, significantly improving the spatial positioning accuracy and temporal synchronization accuracy of the target trajectory. During the fusion process, when one track data is missing or the sampling interval is long, the data from the other track can fill the gap, resulting in a smoother and more continuous trajectory after fusion, avoiding target loss. Therefore, this method does not require strict synchronization of different data sources, does not make prior assumptions about the correlation of sensor data, and uses ICI fusion to ensure the consistency of results. It is suitable for asynchronous, multi-source sensing scenarios, and its tracking effect is significantly improved, especially for low-altitude, slow-moving, small targets (such as UAVs). The multi-source asynchronous trajectory fusion method solves the problems of track matching difficulties and low fusion accuracy caused by asynchronous and unknown association of multi-source trajectory data. Through optimized trajectory matching and fusion algorithms, this method can improve the spatiotemporal accuracy and continuity of target trajectory positioning, and is particularly suitable for three-dimensional trajectory fusion tasks of low-altitude, slow-moving targets such as UAVs. Through the above steps, the effective fusion of asynchronously sampled multi-source trajectory data is achieved, improving the multi-source trajectory data fusion capability, and thus improving the accuracy and completeness of UAV target trajectory data.

[0009] In the feasible implementation of the first aspect, the method further includes: constructing a score based on the robust scale and robust loss of the prediction residuals of each candidate model, and determining the optimal model by combining the prior penalty term of UAV dynamics; the optimal model is one of the constant velocity motion model, constant acceleration motion model and constant steering motion model; when the scores of multiple models meet the preset threshold, the model with lower complexity is selected as the optimal model.

[0010] In the feasible implementation of the first aspect, the method further includes: taking the median of the three-dimensional vector of the difference between the two source locations as a robust center, calculating the centered Euclidean norm, and setting an adaptive threshold using the absolute deviation of the median as a scale; when the sample norm exceeds the adaptive threshold, it is judged as outlier data.

[0011] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: performing one-step prediction and calculating robustness scores for three types of models—constant velocity (CV), constant acceleration (CA), and constant rotational speed (CT)—in parallel at each time step, and selecting the model with the best score for state propagation and completion of that frame. When the scores of multiple models are close, the model with lower complexity is preferred. CT only introduces constant rotational speed in the horizontal plane, and the z-axis is independently described using either CV or CA.

[0012] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: using the position residual between the current observation and the prior prediction as an error source, mapping it to a small-amplitude offset correction for the velocity channel and a feedback correction for the acceleration channel, respectively, and adaptively adjusting the residual magnitude through an observation-driven residual adjustment mechanism. This is used to suppress systematic deviations and instantaneous disturbances caused by external forces such as wind.

[0013] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data also includes: calculating the position difference between the two sources on a unified time axis, using the median of the components as the confidence center and the absolute deviation of the median as the robust interval, setting an error threshold to remove outliers and prevent them from affecting matching and fusion.

[0014] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: adaptively tightening or relaxing the time threshold based on the predicted covariance and chi-square statistics; using the altitude threshold to apply a relaxed range to TDOA in order to suppress vertical mismatch; using the altitude threshold to apply a normal range to Remote ID and adaptively adjusting it with innovation statistics; and using the horizontal residual to construct a cost matrix to solve for the global optimal match on the candidate point pairs obtained by filtering through the time threshold and altitude threshold; the candidate point pairs include matched observation pairs and unmatched observation pairs.

[0015] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data also includes: on the candidate point pairs obtained by filtering through time gate and altitude gate, the cost matrix is ​​constructed using the horizontal residual to solve for the global optimal match; the candidate point pairs include matched observation pairs and unmatched observation pairs.

[0016] In the feasible implementation of the first aspect, the branch selection of CEKF satisfies the following: only the Remote ID branch performs real-time adaptive expansion of the Remote ID measurement covariance according to the observation interval and height transition; and only the TDOA branch performs vertical consistency correction and shrinks the vertical measurement variance in the vertical dimension with the most recent Remote ID posterior height as a reference; and the Remote ID and TDOA matched observation pair branch jointly updates the two sources after passing the consistency criterion; and each branch removes outliers using Mahalanobis distance threshold.

[0017] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: normalizing the covariance gain of each information source data to obtain the fusion weight of the horizontal subspace; and adaptively reducing the fusion weight of the information source data when the gain is lower than the threshold or the condition number of the measurement geometry exceeds the threshold.

[0018] Among the feasible implementation methods in the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data also includes: using Remote ID as the main factor to fuse the vertical subspace; and introducing TDOA vertical information as an auxiliary to improve continuity, provided that the vertical consistency threshold and residual stability conditions are met.

[0019] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data also includes: the target fusion trajectory includes the final target state estimate and its uncertainty.

[0020] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: the method further includes: generating a first target trajectory according to the TDOA positioning system; obtaining a second target trajectory according to the Remote ID device parsing; both the first target trajectory and the second target trajectory are asynchronously sampled and have different timestamps; the first target trajectory and the second target trajectory are used to construct a cost matrix.

[0021] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data also includes: dynamically setting a time threshold and combining it with an altitude threshold to screen candidate observations based on the observed arrival time, predicted covariance, and innovation statistics; and solving the global matching problem by constructing a cost matrix with horizontal residuals on the set that has passed the screening.

[0022] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: determining the target state vector based on the three-dimensional position and velocity components of the matched trajectory pair at a certain moment; and constructing a state transition equation based on the target state vector, the state transition matrix, and the process noise covariance matrix.

[0023] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data also includes: obtaining the observation estimate based on the state transition matrix, state transition equation, and prediction covariance in the absence of observations.

[0024] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: determining the target state vector based on the three-dimensional position and velocity components of the matching result at a certain moment; making predictions based on the state transition matrix and process noise covariance; outputting predicted estimates to maintain trajectory continuity in the absence of observations; and updating the corresponding branches of CEKF when observations exist. Among these, only the Remote ID branch performs time-adaptive expansion of the measurement covariance related to broadcast frequency and altitude transitions, and only the TDOA branch performs vertical consistency correction on the z-axis with the most recent Remote ID posterior altitude as a reference and appropriately shrinks the vertical measurement variance. The Remote ID and TDOA matched observations are jointly updated after passing the consistency criterion, and each branch removes outlier data using Mahalanobis distance thresholds.

[0025] In the feasible implementation of the first aspect, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: performing covariance consistency fusion in the horizontal and vertical subspaces respectively, adaptively weighting the data from each information source with covariance gain, and automatically adjusting the fusion weights when the information gain is insufficient or geometric degradation occurs. Furthermore, the fusion residual is used to simultaneously adjust the upper and lower limits of the time and altitude thresholds, as well as the horizontal and vertical weights of different observation source data. The target fused trajectory represents the final target state estimate and its uncertainty.

[0026] Secondly, embodiments of this application provide an urban low-altitude unmanned aerial vehicle (UAV) trajectory fusion system based on multi-source sensing data. The urban low-altitude UAV trajectory fusion system based on multi-source sensing data includes: at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method provided in the first aspect.

[0027] Thirdly, embodiments of this application provide a computer-readable medium having computer program instructions stored thereon, which can be executed by a processor to implement the method provided in the first aspect. Attached Figure Description

[0028] Figure 1 A schematic diagram of the structure of an urban low-altitude unmanned aerial vehicle trajectory fusion system based on multi-source sensing data provided in this application embodiment;

[0029] Figure 2 A flowchart illustrating a method for fusing urban low-altitude unmanned aerial vehicle (UAV) trajectories based on multi-source sensing data, provided in an embodiment of this application;

[0030] Figure 3A flowchart illustrating a method for fusing urban low-altitude unmanned aerial vehicle (UAV) trajectories based on multi-source sensing data, provided in an embodiment of this application;

[0031] Figure 4 A flowchart illustrating the EKF process in an urban low-altitude UAV trajectory fusion method based on multi-source sensing data, provided in an embodiment of this application.

[0032] Figure 5 This is a flowchart illustrating the ICI process in an urban low-altitude UAV trajectory fusion method based on multi-source sensing data, provided in an embodiment of this application. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the embodiments of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0034] Furthermore, to facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" are not necessarily different. Meanwhile, in the embodiments of the present invention, the terms "exemplary" or "for example" are used to indicate that something is being used as an example, illustration, or description. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.

[0035] The principles and features of this application are described below. The examples given are only for explaining this application and are not intended to limit the scope of this application.

[0036] Time Difference of Arrival (TDOA) positioning systems and Remote ID devices are two methods for detecting low-altitude targets.

[0037] The TDOA system uses multiple receiving stations to measure the time difference of arrival of the target signal to calculate the target's position.

[0038] Remote ID devices parse the location information carried by the target.

[0039] The trajectory data acquired by these two types of devices are relatively independent, with different sampling rates and time bases. As a result, the generated target track (trajectory) sequences are sampled asynchronously and are not synchronized in time, making direct one-to-one correspondence impossible. In addition, the trajectory accuracy measured by each sensor is affected by its own noise and bias. TDOA data has many altitude jumps, while Remote ID data has many gaps.

[0040] Traditional multi-sensor fusion methods mostly require data synchronization or the introduction of a unified time base, and assume that the sensor error characteristics are known. If asynchronous trajectories are forcibly fused, trajectory mismatch, data gaps, or inconsistencies often occur, failing to leverage the complementary advantages of multiple sensors. Specifically, the track timestamps of the TDOA system (millisecond-level sampling) and the Remote ID device (second-level sampling) cannot be aligned, and traditional interpolation alignment introduces timing distortion. The spatial deviation between the two trajectories is time-varying due to the complex electromagnetic environment, making it impossible to establish a fixed correlation model. Differences in track point density lead to data gaps, and forced fusion can cause trajectory mismatches (such as associating different target trajectories) or confidence conflicts.

[0041] The lack of an effective solution in the current technology to handle asynchronous and unknown correlation of multi-source trajectory fusion has created a technical bottleneck in the field of three-dimensional trajectory fusion for low-altitude UAVs.

[0042] This application provides a method for fusion of urban low-altitude UAV trajectories based on multi-source sensing data. It is applicable to fields such as airspace UAV target perception, low-altitude security, and multi-UAV cooperative positioning, and has strong robustness and engineering practicality.

[0043] This application provides an urban low-altitude UAV trajectory fusion system based on multi-source sensing data, which can execute the urban low-altitude UAV trajectory fusion method based on multi-source sensing data provided in this application. Figure 1 This is a schematic diagram of the structure of an urban low-altitude unmanned aerial vehicle trajectory fusion system based on multi-source sensing data, provided in an embodiment of this application.

[0044] like Figure 1 As shown, the urban low-altitude UAV trajectory fusion system 001 based on multi-source sensing data includes at least one processor 011 and a memory 012 communicatively connected to the at least one processor; wherein, the memory 012 stores instructions that can be executed by the at least one processor 011, and the instructions are executed by the at least one processor 011 to enable the at least one processor 011 to execute the urban low-altitude UAV trajectory fusion method based on multi-source sensing data provided in the embodiments of this application.

[0045] Figure 2 This is a flowchart illustrating a method for fusing urban low-altitude UAV trajectories based on multi-source sensing data, provided in an embodiment of this application. Figure 2 and Figure 3 As shown, in some embodiments, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data includes the following steps:

[0046] S1, based on a unified time axis, performs a bidirectional completion operation based on a motion model on the missing moments to obtain the completion result.

[0047] A unified timeline is used to characterize the relationship between remote identity and arrival time difference.

[0048] First, a three-dimensional motion model (CV / CA / CT) of the fixed UAV is provided, giving discrete state propagation and one-step prediction functions to characterize typical motion scenarios such as straight lines, continuous acceleration / deceleration, and horizontal turns.

[0049] S101, in order to closely match the actual flight of the UAV, the model simultaneously considers three-axis velocity and acceleration, introduces wind offset or gust disturbance, and establishes three motion models: CV, CA, and CT.

[0050] CV motion model, state X = [P, v] ∈ R 9 :

[0051] P k+1 =P k +v k Δt, V k+1 =v k +δ v

[0052] Matrix form:

[0053]

[0054] Where, p t+1 p represents the predicted location. t v represents the current observation position. k δ represents the current observation velocity. v This introduces velocity error. The model describes the motion state of the target where both the magnitude and direction of its velocity remain constant.

[0055] CA motion model, state X = [P, v, α] ∈ R 6 :

[0056]

[0057] V k+1 =v k +a k Δt

[0058] ak+1 =a k +δ a

[0059] Matrix form:

[0060]

[0061] Among them, a k For the currently observed acceleration, a k+1 Predict the acceleration at the next moment, δ a This introduces acceleration error. This model is used to describe the motion state of a target with constant velocity and acceleration over a certain time period.

[0062] The CT motion model in this patent only considers multi-rotor UAVs, therefore it assumes that rotation only occurs in the horizontal plane, and the vertical axis does not participate in the rotation of that plane. The nonlinear form in the horizontal plane is as follows:

[0063] x k+1 =x k +v k cosθ k Δt,

[0064] y k+1 =y k +v k cosθ k Δt,

[0065] v k+1 =v k +δ v ,

[0066] θ k+1 =θ k +ωΔt+δ θ

[0067] Where θ represents the horizontal heading angle (rad) δ θ This represents the angular error introduced by the heading angle. This model describes the motion state of the target when it turns.

[0068] S2, based on at least one of the constant velocity motion model, constant acceleration motion model and constant steering motion model, perform a prediction scoring operation on the completion result to obtain the state prediction result.

[0069] For any missing data point from any data source on a unified time axis, the selected motion model and propagation function are used to perform predictive interpolation to fill in the missing data, thereby achieving continuous prediction and completion of the trajectory.

[0070] S201, for the three candidate motion models, perform a one-step prediction at the most recent L observed times, and record the predicted position at time i. Corresponding observation position p i ∈R3 .

[0071] Define the position residual vector and its 2-norm as:

[0072] r i+1 =||R i+1 ||2, i=k-L+1,…k

[0073] Robust scaling estimation using median absolute partial construction The robustness measure of standardized residuals is obtained by using Huber loss:

[0074]

[0075] Where δ>0 is a constant used as the judgment threshold, and combined with the prior knowledge of UAV dynamics, a model feasibility penalty term is constructed:

[0076]

[0077] in, These are the representative velocity, acceleration, and rotation rate estimates for the model within the window; λ represents the estimated mean of horizontal wind offset and gust intensity, respectively; v , λ a , λ ω , λ b , λ g The weights are non-negative. max a max ω max This refers to the upper limit of the performance of the target drone.

[0078] Define the overall score of model m:

[0079]

[0080] The model with the lowest overall score is selected as the optimal model for the current time period. When the scores of multiple models are close, the model with lower complexity (CV < CA < CT) is selected first to reduce the risk of overfitting.

[0081] Use motion models to predict states and supplement the observation data.

[0082] The state propagation relationship is modeled as a nonlinear function:

[0083] X k =f m (X k-1 )+w k

[0084] Among them, f m(·) represents the selected m-motion model function, describing the state change from the previous time step to the current time step; w k This represents process noise, reflecting external disturbances and system modeling errors. It is assumed to follow a zero-mean Gaussian distribution: w k ~N(0,Q) k ).

[0085] To enhance the model's response to micro-maneuvers and sudden changes, this application introduces process noise modeling based on acceleration perturbations and designs an adaptive adjustment mechanism using observation residuals. The process noise covariance matrix is ​​set as follows:

[0086]

[0087] in, I0 represents the intensity of the acceleration disturbance at the current moment, used to reflect the motion uncertainty of the target during trajectory prediction; I3 is a 3×3 identity matrix. This matrix form is derived from the first-order approximate derivation of the Continuous White Noise Acceleration Model (CWNA) in the Extended Kalman Filter.

[0088] In some embodiments, the method further includes: taking the median of the three-dimensional vector of the difference between the two source locations as a robust center, calculating the centered Euclidean norm, and setting an adaptive threshold using the absolute deviation of the median as a scale; when the sample norm exceeds the adaptive threshold, it is judged as outlier data.

[0089] To achieve dynamic adjustment, this application further designs an observation-driven residual adjustment mechanism: for two consecutive observation points Z k Z k-1 The velocity residual index is defined as follows:

[0090]

[0091] Where ||·|| represents the Euclidean norm, measuring spatial distance. If Δv is significantly higher than the average sliding average of recent steps, the system automatically identifies a sudden velocity change and increases its speed accordingly. If the trajectory is continuous and smooth, the disturbance amplitude should be reduced to improve prediction accuracy.

[0092] S3. Based on the confidence interval of the difference between the two source locations, outlier data in the state prediction results are removed to obtain the data to be observed.

[0093] Outliers in both types of observation data are removed to prevent them from affecting subsequent fusion results. Using the difference between the two source locations as the criterion, and based on the confidence interval of the median of the location difference and the absolute deviation of the median, outlier samples exceeding the threshold are removed and recorded.

[0094] The positions of the two completed paths are recorded as follows:

[0095]

[0096] Position difference is A robust median μ is estimated for the location difference, and the dispersion around this center is measured to construct an adaptive threshold. μ is calculated using the median of each component, and the deviation of each sample is characterized by a three-dimensional centering norm.

[0097] r k =||ΔP k -μ||2

[0098] A robust metric σ = MAD({r_t) is constructed using the median absolute deviation (MAD). k} k＝1 Set a threshold τ to identify outliers:

[0099] τ=max(τ min ,ασ)

[0100] When r k If the value is greater than τ, it is considered abnormal data and will be removed.

[0101] S4 uses observation timestamps to perform threshold filtering based on time and height thresholds to limit the candidate observation set.

[0102] In some embodiments, the time gate threshold is adaptively tightened or relaxed based on the predicted covariance and chi-square statistic.

[0103] In some embodiments, in the height gate, the height gate threshold is used to apply a lenient range to TDOA to suppress vertical mismatches; the height gate threshold is also used to apply a normal range to Remote ID and is adaptively adjusted with innovation statistics.

[0104] The global optimal match is solved by constructing a cost matrix using the horizontal residuals on candidate point pairs selected through time and height thresholds. Candidate point pairs include matched observation pairs and unmatched observation pairs.

[0105] like Figure 4 As shown, the matched data is processed by a conditional Kalman filter. When only RemoteID or only TDOA is reached, a corresponding single-source update is performed. When both sources are matched and available simultaneously in S4, a paired update is triggered. This invention performs conditional Kalman filtering on the two types of data separately, estimating the target state independently. Both the state result and the covariance result are continuous target state estimation sequences.

[0106] In some embodiments, when performing step S4, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes the following steps:

[0107] S401: Based on the three-dimensional position and velocity components of the matched trajectory pair at a certain moment, determine the target state vector and construct the state transition equation. Predict the current target state and covariance based on the target state at the previous moment.

[0108] The filter at the previous time t k-1 Give the posterior state Covariance P k-1 To compensate for the time difference caused by the inconsistency between the two types of measurement timestamps, we use the current time t... k The motion model is used to predict the target, thus obtaining the predicted state. and predicted location The process is as follows. The equation of motion for a continuous-time system is X. k =f(X) k-1 )+w k Discretize the continuous-time model with a time step Δt. k =t k -t k-1 The state transition matrix F can be obtained by using a discretization method based on matrix exponentiation. k process noise covariance Q c :

[0109]

[0110] Wherein, exp(AΔt) k ) represents the matrix index, The nth power of the time step These represent the variances of the corresponding acceleration noise. The first three columns of the first row are zero because position does not directly affect the position at the previous moment. Through the above discretization, the predicted state and the predicted covariance can be obtained:

[0111]

[0112] in, and P k-1|k-n These are the posterior state and covariance after the last update, respectively; x k|k-1 That is, time t k The predicted state, including the predicted location. With velocity components. This predicted position will be used as a reference coordinate.

[0113] After the prediction is obtained, threshold screening and matching are required based on the actual arrival of various observations.

[0114] Time threshold filtering:

[0115]

[0116] If Δt ij Less than the reference threshold Tth These may belong to the same moment, but to allow for prediction errors due to target motion, we add prediction bias compensation to the threshold:

[0117] T′ th =T th +K t δ ij ,

[0118] in, For the previous time t k-1 Time t obtained using the prediction model k The predicted location; δ ij The spatial deviation between the predicted location and the Remote ID measurement is measured. A large deviation indicates the target is moving quickly, and the time threshold should be appropriately relaxed. Only when Δt... ij ≤T th Only if the height threshold is met will subsequent height threshold checks continue; otherwise, it will be considered a mismatch.

[0119] Height threshold filtering:

[0120] Δh ij =|z i (T) -z j (D) |;

[0121] Since TDOA positioning accuracy is typically worse in the vertical direction than in the horizontal direction, a relatively lenient height threshold H needs to be set to avoid mismatches. th Only when Δh is satisfied... ij ≤H th If a TDOA measurement and a Remote ID measurement are considered to have a potential match, then they are considered to be mismatched.

[0122] The median m and median absolute deviation (MAD) of the height difference between TDOA and Remote ID measurements were calculated. Convert MAD to an approximate standard deviation. Simultaneously combine measurement and model uncertainties. Final height threshold:

[0123]

[0124] When Δh ij >H th The condition is determined to be a mismatch. It is marked as a height jump point and enters the jump point processing flow. This method can adaptively reflect the observation noise and historical fluctuation scale, thereby improving robustness to sudden jumps while ensuring matching accuracy.

[0125] S402, if in the same time window [t]k -T win , t k +T win If multiple TDOA-Remote ID candidate pairs satisfy the aforementioned dual thresholds exist within a given range, then a final pairing scheme must be determined through global optimization. Let...

[0126]

[0127] Among them, t k This can be taken as the current reference time, such as the minimum of all measurement times or determined by uniform gridding; T wim Allowed time search window. Once determined. Construct cost matrix Its elements are defined as:

[0128]

[0129] Where α ∞ A maximum value is selected to prohibit illegal matches. This cost only considers horizontal plane distance, reflecting the idea of ​​ensuring horizontal pairing first, and then taking vertical constraints into account. The global optimal match is achieved by minimizing the total cost, using the cost matrix C as input.

[0130]

[0131] Where x ij =1 indicates that TDOA measurement i is associated with Remote ID measurement j, and 0 indicates otherwise. After pairing is complete, all measurements satisfying x ij Index pairs with a value of 1 are considered matched pairs and used for subsequent fusion updates. Unpaired measurements are then entered as single-source measurements into the filter's single-source update process.

[0132] In S403, another implementation of CEKF, the branch selection satisfies the following: only the Remote ID branch performs real-time adaptive expansion of the Remote ID measurement covariance based on the observation interval and altitude transition; and only the TDOA branch performs vertical consistency correction and shrinks the vertical measurement variance in the vertical dimension with the most recent Remote ID posterior altitude as a reference; and the Remote ID and TDOA matched observation pair branch jointly updates the two sources after passing the consistency criterion; each branch uses a Mahalanobis distance threshold to eliminate outliers. Urban low-altitude UAV trajectory fusion methods based on multi-source sensing data also include:

[0133] S40321, when only Remote ID is available, performs Remote ID-only updates and adaptively expands the measurement covariance based on arrival intervals and height jumps to stabilize the Z-axis estimation. When only TDOA is available, performs TDOA-only updates, preferentially absorbing its constraints on XY, while imposing consistency constraints and weight contraction on the Z-axis using the most recently reliable Remote ID height. When both sources are available and matched, performs Paired updates.

[0134] Only when TDOA is updated;

[0135] The Remote ID measurement failed to pair with the current TDOA measurement, or the current Remote ID measurement was lost. TDOA measurement Independent arrival. Since TDOA is prone to abrupt changes in the altitude direction, the most recent posterior altitude is first obtained from the Remote ID subsystem. For reference, the current TDOA height is corrected:

[0136]

[0137] in This is the altitude prediction from the TDOA subsystem at the previous time step. The corrected TDOA altitude deviates from the Remote ID reference altitude by no more than the jump threshold δ. z Therefore, TDOA is considered to have high reliability, and its noise variance is measured. reduce:

[0138]

[0139] In the horizontal direction, the original measurement covariance of TDOA is used:

[0140]

[0141] in, Indicates the most recent N T One TDOA location sample, These are the sample means of each component in the x and y dimensions, respectively; this estimate can dynamically reflect the error characteristics of the TDOA system under the current geometric layout and signal quality.

[0142] The measured covariance is:

[0143]

[0144] The measurement residuals and innovation covariance are:

[0145]

[0146] in, It is to predict the state By advancing through time The prior estimate after; This corresponds to the covariance.

[0147] Similarly, Mahalanobis distance is used for screening:

[0148]

[0149] Where γ T Take χ² with 3 degrees of freedom and a confidence level of 1-α. 2 Distribution quantiles are used to remove outliers. The Kalman gain is calculated when the following conditions are met:

[0150]

[0151] And update the state and covariance:

[0152]

[0153] When only the Remote ID is updated:

[0154] No TDOA measurement is paired with the current Remote ID measurement. Independent arrival. The noise covariance is measured first based on the time interval. With height jump Perform adaptive expansion:

[0155]

[0156] in, These are the calibration variances of the Remote ID measurement in three directions; when If α is increased (due to lower Remote ID update frequency or network latency), then... t，j An increase indicates a decrease in the timeliness of the Remote ID, necessitating an increase in the measurement variance; when If the value is large (Remote ID output height jump), then α h，j An increase indicates that the current Remote ID data may be abnormal, and variance needs to be expanded for measurement. Then, similar to the single-source TDOA update process, the residuals and innovation covariance are calculated, Mahalanobis distance is used for screening, and an update is triggered when the filter passes.

[0157] S5 constructs a three-branch Conditional Extended Kalman Filter (CEKF) consisting of Remote ID, Time Difference of Arrival (TDOA), and matched observation pairs of Remote ID and TDOA. It adaptively selects the update path using the Mahalanobis distance threshold criterion and performs single-source updates on unmatched observation pairs to maintain trajectory continuity.

[0158] The filtering uses a motion model for prediction and updates based on observation availability: a single-source update is performed when only one source is available, and a paired update is performed when both TDOA and Remote ID are available and already matched. The entire process incorporates a chi-square threshold to eliminate outliers and a covariance-driven adaptive mechanism to enhance robustness to asynchrony and noise anisotropy.

[0159] S6 performs covariance-consistent fusion in both the horizontal and vertical subspaces, and adaptively weights the source data based on the information entropy of the covariance matrix.

[0160] The horizontal and vertical subspaces are processed separately, and adaptive weighted fusion is performed under consistency constraints to ensure the reliability of the results.

[0161] In some embodiments, the target fusion trajectory is the final target state estimate and its uncertainty.

[0162] like Figure 5 As shown, after obtaining the state estimates of the two trajectories, this invention employs a fusion algorithm driven by latitudinal decomposition fusion and covariance information entropy. This fully leverages the complementary advantages of Remote ID and TDOA observations in the horizontal and vertical dimensions, rather than simply using traditional CI fusion or fixed-weighting.

[0163] Since the two EKF paths operate independently, the output state estimates x1 and x2 and their covariances P1 and P2 may have unknown correlations. Directly using a common weighted average may lead to an underestimation of uncertainty or the introduction of bias. In some embodiments, when performing step S6, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes the following steps:

[0164] For the horizontal subspace, the horizontal covariance of TDOA is given in S5. and the horizontal covariance after adaptive dilation measured by Remote ID. The preliminary horizontal fusion covariance and the corresponding observation mean at the fusion level are calculated using the Bayesian optimal fusion formula:

[0165]

[0166] To avoid the subjectivity and lack of robustness inherent in static weighting, the covariance information entropy index is introduced:

[0167]

[0168] Covariance after joint fusion The corresponding entropy is:

[0169]

[0170] Define entropy gain and like This indicates that TDOA contributes more to the reduction of uncertainty in the horizontal subspace, and the weight of TDOA should be increased.

[0171] In some embodiments, the covariance gain is normalized for each information source data to obtain the fusion weights for the horizontal subspace; if the gain is below a threshold or the condition number of the measurement geometry exceeds a threshold, the fusion weight of that information source data is adaptively reduced. The fusion weights are constructed using the normalized entropy gain, and the final fusion formula is rewritten as follows:

[0172]

[0173] The above adaptive weights It will be dynamically updated as the level covariance of TDOA and Remote ID changes in real time, ensuring that the differences in data characteristics are fully utilized.

[0174] In some embodiments, the urban low-altitude UAV trajectory fusion method based on multi-source sensing data further includes: using Remote ID as the primary factor to fuse the vertical subspace; and introducing TDOA vertical information as an auxiliary factor to improve continuity, provided that the vertical consistency threshold and residual stability conditions are met.

[0175] For the vertical subspace, step S5 provides the vertical covariance of the Remote ID measurement. Because the Remote ID height has high accuracy, it is preferentially used as the fusion reference. If the TDOA height is at the current paired observation time... With Remote ID height The deviation between them satisfies Furthermore, if the recent horizontal residuals are stable, then the TDOA height is considered auxiliary information. Set the fusion weights:

[0176]

[0177] Otherwise Ensure high integration by primarily using Remote ID measurements.

[0178] Therefore, the fused height measurement and its corresponding covariance are constructed as follows:

[0179]

[0180] Horizontal fusion Vertical integration Combined into three-dimensional fusion observation Its fusion measurement covariance matrix is Finally, the observations will be merged. Covariance Incorporating EKF updates, computational innovation r k and innovation covariance S k :

[0181]

[0182] in, Let P(t) be the prior state obtained through prediction before this update. k |t k-1 H represents the predicted covariance; H is the uniform measurement matrix. Mahalanobis distance is used for screening, which meets the criteria. The Kalman gain, state, and covariance are calculated and updated in real time to obtain the final 3D fused trajectory.

[0183] In some embodiments, a cost matrix is ​​constructed using horizontal residuals on candidate point pairs selected by time gate and height gate, and the global optimal match is solved on this matrix; the candidate point pairs include matched observation pairs and unmatched observation pairs.

[0184] Through the above process, the present invention ultimately outputs the fused trajectory state sequence for each target. This trajectory integrates information from both TDOA and Remote ID, exhibiting high accuracy and completeness. In practical applications, the system can execute the aforementioned fusion algorithm on all matched target pairs, achieving simultaneous perception of multiple low-altitude UAV targets. The method provided by this invention has low hardware requirements, can post-process offline data, and can also be extended to real-time systems (provided that the timestamps of each sensor's data are available). Experiments show that for low-altitude, slow-moving targets such as UAVs, the positioning error is significantly reduced and trajectory breakpoints are decreased after fusion using this method, more reliably reflecting the target's true motion trajectory, demonstrating the effectiveness and practical value of this invention.

[0185] As shown in the above embodiments, this urban low-altitude UAV trajectory fusion method based on multi-source sensing data utilizes the complementarity of two types of sensor information to reduce the positioning error of a single data source, significantly improving the spatial positioning accuracy and temporal synchronization accuracy of the target trajectory. During the fusion process, when one track data is missing or the sampling interval is long, the data from the other track can fill the gap, resulting in a smoother and more continuous trajectory after fusion, avoiding target loss. Therefore, this method does not require strict synchronization of different data sources, does not make prior assumptions about the correlation of sensor data, and uses conditional Kalman filtering fusion to ensure the consistency of results. It is suitable for asynchronous, multi-source sensing scenarios, and its tracking effect is significantly improved, especially for low-altitude, slow-moving, small targets (such as UAVs). The multi-source asynchronous trajectory fusion method solves the problems of track matching difficulties and low fusion accuracy caused by asynchronous and unknown correlation of multi-source trajectory data. This method, through optimized trajectory matching and fusion algorithms, can improve the spatiotemporal accuracy and continuity of target trajectory positioning, and is particularly suitable for the perception and tracking of low-altitude, slow-moving targets such as UAVs. Through the above steps, effective fusion of asynchronously sampled multi-source trajectory data is achieved, improving the multi-source trajectory data fusion capability, and thus improving the accuracy and completeness of UAV target trajectory data.

[0186] Based on the same application concept, this application also provides an urban low-altitude UAV trajectory fusion system based on multi-source sensing data. The method corresponding to the urban low-altitude UAV trajectory fusion system based on multi-source sensing data can be the urban low-altitude UAV trajectory fusion method based on multi-source sensing data in the aforementioned embodiments, and its problem-solving principle is similar to that method. The urban low-altitude UAV trajectory fusion system based on multi-source sensing data provided in this application includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the methods and / or technical solutions of the aforementioned embodiments of this application.

[0187] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0188] Specifically, this embodiment can employ any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0189] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0190] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0191] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0192] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0193] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0194] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or page components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0195] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0196] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0197] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0198] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0199] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a device claim may also be implemented by a single unit or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for fusion of urban low-altitude unmanned aerial vehicle (UAV) trajectories based on multi-source sensing data, characterized in that, include: Based on a unified time axis, a bidirectional completion operation based on a motion model is performed on the missing moments to obtain the completion result; The unified timeline is used to characterize the relationship between remote identity identifiers and arrival time differences; Based on at least one of the constant velocity motion model, constant acceleration motion model and constant steering motion model, a prediction scoring operation is performed on the completion result to obtain the state prediction result. Based on the confidence interval of the difference between the two source locations, outlier data in the state prediction results are removed to obtain the data to be observed. Based on the timestamps of the observation data, the candidate observation data are gating and filtering sequentially using time gates and height gates. A cost matrix is ​​constructed in the horizontal plane, and the global optimal matching is solved to obtain the matched and unmatched observation pairs. A three-branch Conditional Extended Kalman Filter (CEKF) is constructed, consisting of Remote ID, Time Difference of Arrival (TDOA), and matched observation pairs of Remote ID and TDOA. The update path is adaptively selected using the Mahalanobis distance threshold criterion, and single-source updates are performed on the unmatched observation pairs to maintain trajectory continuity. Covariance-consistent fusion is performed in both the horizontal and vertical subspaces, and the information source data is adaptively weighted based on the information entropy of the covariance matrix. In the vertical subspace, variance inflation is applied to TD0A based on the residuals to stabilize the vertical estimate.

2. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1, characterized in that, The method further includes: The optimal model is determined by constructing a score based on the robust scale and robust loss of the prediction residuals of each candidate model, and combining it with the prior penalty term of UAV dynamics; the optimal model is one of the constant velocity motion model, constant acceleration motion model and constant steering motion model. When multiple model scores meet a preset threshold, the model with lower complexity is selected as the optimal model.

3. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1 or 2, characterized in that, The method further includes: The median of the three-dimensional vector of the position difference between the two sources is taken as the robust center according to the components. The centered Euclidean norm is calculated, and an adaptive threshold is set with the absolute deviation of the median as the scale. When the sample norm exceeds the adaptive threshold, it is determined to be outlier data.

4. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1 or 2, characterized in that: Based on the predicted covariance and chi-square statistic, the time threshold is adaptively tightened or relaxed; The height threshold is used to apply a lenient range to TDOA in order to suppress vertical mismatches; the height threshold is also used to apply a normal range to Remote ID and is adaptively adjusted with innovation statistics; The global optimal match is solved by constructing a cost matrix using the horizontal residuals on the candidate point pairs obtained by filtering through the time threshold and the height threshold; the candidate point pairs include matched observation pairs and unmatched observation pairs.

5. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1 or 2, characterized in that, The branch selection of CEKF satisfies: Only the Remote ID branch performs real-time adaptive dilation of the Remote ID measurement covariance based on the observation interval and height transition; and, Only the TDOA branch performs vertical consistency correction and reduces vertical measurement variance by referencing the posterior height of the most recent Remote ID in the vertical dimension; and, Remote ID and TDOA matching observation pairs are jointly updated for both sources after passing the consistency criterion; each branch uses Mahalanobis distance threshold to eliminate outliers.

6. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1 or 2, characterized in that, The method further includes: The covariance gain is normalized according to the data from each information source to obtain the fusion weight of the horizontal subspace; If the gain is below a threshold or the number of conditions for the measurement geometry exceeds a threshold, the fusion weight of the information source data is adaptively reduced.

7. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1 or 2, characterized in that, The method further includes: The vertical subspace is integrated with Remote ID as the primary factor; Under the condition of satisfying the vertical consistency threshold and residual stability, TDOA vertical information is introduced as an aid to improve continuity.

8. The urban low-altitude UAV trajectory fusion method based on multi-source sensing data according to claim 1, characterized in that, The method further includes: The first target trajectory is generated based on the TDOA positioning system; Based on the Remote ID device parsing, the trajectory of the second target is obtained; The first target trajectory and the second target trajectory: The first target trajectory is the target trajectory generated by the time difference of arrival positioning system; the second target trajectory is the target trajectory parsed by the Remote ID device. Both types of data can be asynchronously sampled and have different timestamps.

9. A trajectory fusion system for urban low-altitude unmanned aerial vehicles based on multi-source sensing data, characterized in that: include: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 8.

10. A computer-readable medium having computer program instructions stored thereon, characterized in that: The computer program instructions can be executed by a processor to implement the method as described in any one of claims 1 to 8.

Images