Track interruption continuing method based on multi-model weighted prediction and time-varying fuzzy logic

Abstract

The invention discloses a track interruption continuing method based on multi-model weighted prediction and time-varying fuzzy logic. The method comprises the following steps: a rough correlation screening module quickly rejects obviously mismatched tracks by using a gate dynamically adjusted along with interruption time; the multi-model weighted prediction module rearranges tracks to be batched in an inverted order, obtains the weight of each model by establishing CV, CA and CT parallel filter banks, and obtains weighted prediction points through calculation; the similarity calculation module constructs a fuzzy membership model containing a nonlinear time function, and calculates the matching similarity between a new track to be batched and an interrupted old track; and the continuing processing module completes track continuing according to a similarity maximum principle. According to the method, through a strategy of coarse screening and fine judgment, the problem of misassociation caused by inaccurate track prediction in a target maneuvering scene and poor adaptability of traditional fuzzy similarity judgment under different interruption durations is effectively solved, and the accuracy and robustness of track continuation of a radar in a complex environment are improved.

Description

Technical Field

[0001] This invention relates to the field of radar data processing and target tracking technology, specifically to a method for resuming track interruptions in scenarios involving maneuvering targets and long or short-term interruptions. Background Technology

[0002] During radar target tracking, tracks are often interrupted due to clutter interference, missed detections, target maneuvers, or blind spot obstruction. Without effective reconnection, the track for the same target will be split into multiple tracks, severely impacting situational awareness. Existing track reconnection algorithms mainly include nearest neighbor association, weighted track association, and statistical dual-threshold algorithms. Among these, the nearest neighbor association method is simple to implement but prone to false associations; the weighted track association method relies on the covariance matrix and prediction accuracy of the Kalman filter output, and performs poorly in maneuvering target scenarios; while the statistical dual-threshold interruption track association algorithm has high accuracy, its engineering implementation is complex and highly dependent on the accuracy of system measurements and parameter settings, easily leading to missed associations when settings are unreasonable. In summary, addressing the problems of existing algorithms, particularly the inaccurate prediction of maneuvering targets due to a single motion model, and the poor adaptability of traditional fuzzy similarity discrimination under different interruption durations, an improved track interruption reconnection method is urgently needed. Summary of the Invention

[0003] This invention proposes a method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic. The technical solution to achieve the purpose of this invention is as follows:

[0004] Improvements in the prediction stage: To address the problem of inaccurate prediction of maneuvering targets due to reliance on a single motion model in traditional methods, an adaptive weighted inverse prediction method based on innovation and Gaussian likelihood is proposed. A parallel filter bank incorporating three motion models—CV (uniform velocity), CA (uniform acceleration), and CT (cooperative turning)—is established to perform inverse rearrangement and weighted prediction of the initial batch of trajectories to cope with target maneuvers.

[0005] Improvements in the screening phase: A coarse correlation gate that dynamically adjusts with the interruption time was designed to quickly and accurately eliminate mismatched track pairs and optimize computational efficiency.

[0006] Improvements in the decision-making stage: A fuzzy similarity model was designed that dynamically adjusts the membership function with the interruption time. The parameter in the membership function that measures error tolerance is designed as a nonlinear function with respect to the interruption time to match the growth of prediction uncertainty and achieve robust evaluation of the degree of association. Attached Figure Description

[0007] Figure 1 This is an overall flowchart of a trajectory interruption resuming method based on multi-model weighted prediction and time-varying fuzzy logic according to the present invention.

[0008] Figure 2 This is a schematic diagram illustrating the information of all flight paths in the simulation experiment scenario of this invention embodiment;

[0009] Figure 3 In the embodiments of the present invention, T it Diagram of the interrupted flight path when =8 (interruption time is 8 cycles);

[0010] Figure 4 In the embodiments of the present invention, T it Comparison of the correlation matrices for different algorithms when the value is 8;

[0011] Figure 5 In the embodiments of the present invention, T it Comparison of track continuity results for different algorithms at 8 o'clock. Detailed Implementation

[0012] To facilitate understanding of the present invention, a more comprehensive description will be given below with reference to the accompanying drawings. The present invention can be implemented in different forms and is not limited to the embodiments described in the text. Rather, embodiments are provided to make the disclosure of the present invention more thorough and complete. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0013] I. Overall Algorithm Flow

[0014] See Figure 1 The basic scheme of this invention adopts the idea of ​​"coarse screening - fine judgment" and proposes a track interruption continuation method based on multi-model weighted prediction and time-varying fuzzy logic to address the track interruption problem. This method mainly includes four steps: coarse correlation screening, multi-model weighted prediction, similarity calculation, and continuation processing.

[0015] Step 1: Coarse Association Filtering. To quickly remove obviously mismatched tracks and optimize computational efficiency, the coarse association filtering module employs a coarse association gate that dynamically adjusts with the interruption time. Let the tracks to be batched be Xj, and the old tracks in the eliminated track container be X. l Its dynamic threshold can be specifically expressed by the following formula:

[0016]

[0017] Among them, the location-based threshold value Γ pos_base =150m, basic speed threshold Γ vel_base =15m / s, s(t)=15t, v(t)=t, where t is the trajectory X to be launched. j The first point and the old track X i The last bit of time difference.

[0018] Step 2: Multi-model weighted prediction. The multi-model weighted prediction module processes the initial measurement data of the new trajectory by establishing a parallel filter bank containing three motion models: CV (uniform velocity), CA (uniform acceleration), and CT (uniform speed turning). This addresses the problem of inaccurate prediction of maneuvering targets by a single model.

[0019] 1. Reverse data order: Pending batch of flight tracks X j The sequence consisting of k original measurement points is Z. j ={z1, z2, ..., z k}, where z1 is the starting point of the new track. Rearranging this sequence in reverse order by timestamps yields a new sequence Z′. j ={z′1,z′2,...,z′ k}={z k , z k-1 , ..., z l Here, k is set to 5.

[0020] 2. Parallel filtering using multiple models, determining weights: using Z′ j The information v of the sequence at the last point in each model filtering process k and covariance S k The Gaussian likelihood value corresponding to the actual measurement is calculated. This likelihood value is used as a "score" to measure the model's fit; models with higher scores are assigned greater weight. Gaussian likelihood value L k The calculation is as follows:

[0021]

[0022] Model weights ω k Determined by the normalized Gaussian likelihood value:

[0023]

[0024] Where k = 1, 2, 3 correspond to the three motion models CV, CA, and CT, respectively.

[0025] 3. Calculate the number of prediction steps: Calculate the number of back-prediction steps m required for the new trajectory:

[0026]

[0027] Among them, t new_start This represents the time of the first point in the pending batch of track data, which is the time of the last point in the pending batch of track data after reverse ordering. old_end This indicates the last point in the old track where the old track was interrupted. T is the radar point update cycle, and round() means rounding to the nearest integer.

[0028] 4. Parallel prediction using multiple models to generate weighted prediction points: The filtered states of each model are calculated from t... new_start Predict the time step m to reach the interruption time t of the old track. old_end Finally, the predicted states of each model are aligned and weighted and fused to obtain the final predicted state.

[0029] X pred、final =ω cv X′ pred、cv +ω ca X′ pred、ca +ω ct X′ pred、ct

[0030] Note: Since this is a reverse prediction, the velocity value needs to be inverted.

[0031] Step 3: Calculate similarity. In the similarity calculation module, it is necessary to determine the fuzzy factors, membership functions, and weights of the fuzzy factors.

[0032] 1. The simulation here is in a two-dimensional plane, and the changes in position and velocity are used as fuzzy factors u1 and u2:

[0033]

[0034] Where, x i (l), y i (l) represents the position of the break point on the x and y axes of the old track i at time l, x j (l), y j (l) represents the weighted prediction point position of the x and y axes of the trajectory j to be launched at time l. Similarly, Equal to the corresponding speed.

[0035] 2. Nonlinear Time-Varying Membership Function: A membership function that dynamically adjusts with the interruption time was designed. The parameter measuring error tolerance is designed as a nonlinear function g(t) with respect to the interruption time, as shown in the following formula:

[0036]

[0037] Where, τ k For adjustment degree, u k Let σ be the k-th fuzzy factor in the fuzzy factor set. k This represents the extent of the corresponding fuzzy factor; g(t) is a nonlinear function of the interruption time t. Here, τ... k The value is 0.5; when k = 1, i.e., when the positional ambiguity factor is involved, σ1 = 3, g(t) = 8t. 1.5 When k = 1, i.e., when the velocity fuzzy factor is involved, σ² = 2, g(t) = 6t. 0.5 .

[0038] 3. Calculate the similarity and store the calculated similarity in a container: Since location ambiguity has the greatest impact on correlation, followed by speed, set the weights of the ambiguity factors to 0.6 and 0.4 respectively (i.e., a1 = 0.6, a2 = 0.4). Calculate the correlation between the new and old tracks using the following formula:

[0039]

[0040] Step 4: Continuation Processing. If the similarity storage container is empty, continuation is not possible, and the new track will be started from the next batch of tracks; otherwise, the old track with the highest similarity is selected. It is determined whether the highest similarity meets the set threshold. If it does, the tracks are merged and the batch numbers are unified; otherwise, continuation is considered impossible, and the new track will be started from the next batch of tracks.

[0041] II. Experimental Verification and Analysis

[0042] To further verify the effectiveness and robustness of the improved algorithm proposed in this invention, this embodiment constructs a simulation scenario involving the interruption of a maneuvering target's trajectory and conducts comparative experiments with the nearest neighbor association continuation algorithm, the weighted trajectory association continuation algorithm, and the statistical dual-threshold interruption trajectory association algorithm.

[0043] 1. Simulation Scene Construction

[0044] Assuming the target is flying in a two-dimensional plane, radar interference causes partial loss of measurement data, resulting in track interruption. The radar measurement noise follows a Gaussian distribution with zero mean, and its ranging error standard deviation σ... ρ =3m, azimuth error standard deviation σ θ =0.3°, update period T = 2.4s, tracking period number is 100. The simulation scene contains 4 targets, such as Figure 2 As shown, during cycles 1 to 30, all targets maintained uniform linear motion. Maneuvers occurred during cycles 30 to 59, after which uniform linear motion resumed. The interruption occurred during cycle 35. T it Define the number of radar update measurements during the interruption time. The motion state of each target is set as follows: (1) The initial position of target 1 is (500m, 300m), the initial velocity is (10m / s, 0), and it always maintains a uniform linear motion state; (2) The initial position of target 2 is (600m, 400m), the initial velocity is (9m / s, 0), and it is in a uniform acceleration motion state during the maneuver time, with an acceleration of (0.2m / s). 2 -0.3m / s 2 (3) The initial position of target 3 is (700m, 200m), the initial velocity is (8m / s, 0), and it is in a state of uniform acceleration during the maneuver time, with an acceleration of (0.2m / s²). 2 0.5m / s 2(4) The initial position of target 4 is (300m, 100m), the initial velocity is (12m / s, 0), and it is in a uniform turning motion state during the maneuver time, with an angular velocity of ω=π / 100(rad / s). Compare all the trajectory information in the simulation experiment as follows Figure 1 As shown.

[0045] 2. Evaluation Indicators

[0046] To quantify algorithm performance, the following evaluation metrics are defined:

[0047] (1) Average correct association rate

[0048]

[0049] (2) Average error association rate

[0050]

[0051] (3) Average missing association rate

[0052]

[0053] (4) Average running time

[0054]

[0055] Where n represents the total number of objectives in each experiment, and N represents the number of experiments. t n represents the number of targets correctly associated in each experiment. f n represents the number of incorrectly associated targets in each experiment. o Let n represent the number of targets missed in each experiment, and satisfy n = ... t +n f +n o =n, T0 represents the algorithm's running time in each experiment.

[0056] 3. Comparative Experimental Analysis

[0057] Set different T it One hundred Monte Carlo experiments were conducted, and the simulation correlation results of different algorithms are shown in Table 1. Meanwhile, T was selected. it The experimental results of =8 are used as an example in Figure 3 , Figure 4 and Figure 5 The information is displayed in the middle. Among them, Figure 3 Measurement information for four targets before and after the interruption is given. Figure 4 The correlation matrices for the four objectives are given under different algorithms (white represents correlation, black represents non-correlated). Figure 5The association and continuation results for the four targets are given under different algorithms (red dashed lines represent unsuccessful associations).

[0058] Table 1 shows the correlation results of the comparative simulation experiments.

[0059]

[0060] In short-term interruption scenarios (T) it =4): The correct correlation rate of all four algorithms is 100%, but the average running time of the improved algorithm of this invention is better than that of the statistical dual-threshold algorithm. In long-term interruption and maneuver scenarios (T it =8): As the interruption time increases, the performance of the nearest neighbor and weighted track association algorithms drops sharply, with correct association rates decreasing to 57.25% and 71.75%, respectively. This is because they rely on a single uniform velocity model, which fails due to model mismatch when the target maneuvers. In contrast, the improved algorithm of this invention maintains a correct association rate of 83.75%, outperforming all other comparative algorithms and demonstrating its advantage in scenarios with long interruptions and maneuvers. In summary, this invention significantly improves the track continuity performance of maneuvering targets in complex environments by combining multi-model weighted prediction with dynamic fuzzy logic.

[0061] The embodiments described above are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic, characterized in that, Includes the following steps: Step 1: Perform coarse correlation screening on the pending start-up tracks, traverse the eliminated track container. If the traversal is not completed, use the coarse correlation gate that is dynamically adjusted with the interruption time to filter out possible matching interrupted old tracks and execute Step 2; otherwise, execute Step 6. Step 2: Reverse the order of the initial measurement data of the tracks to be started, and output the rearranged track data to Step 3; Step 3: Filter the rearranged flight path data to be launched using a multi-model parallel filter bank, calculate the Gaussian likelihood value of each motion model based on the innovation and covariance, and determine the adaptive weight of each model accordingly, then proceed to step 4. Step 4: Based on the rearranged batch of flight tracks and the matching interrupted old flight track data, calculate the number of prediction steps for the rearranged batch of flight tracks. Combine the model weights obtained in Step 3 to perform multi-model parallel prediction on the rearranged batch of flight tracks and output the weighted prediction points of the batch of flight tracks to Step 5. Step 5: Based on the weighted prediction points of the to-beginning flight track and the interruption points of the old flight track, use the fuzzy membership function that dynamically adjusts with the interruption time to calculate the similarity between the prediction points of the to-beginning flight track and the interruption points of the old flight track and store it in the container, then proceed to step 6. Step 6: Traverse the similarity storage container. If the container is empty, the pending batch of tracks cannot be continued and will be used as the starting point for a new track. Conversely, the interrupted old track with the highest similarity value that meets the set threshold will be selected for continuation processing.

2. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 1, characterized in that, In step 1, the dynamic threshold of the coarse correlation gate, which is dynamically adjusted according to the interruption time, is expressed as follows: Among them, Γ pos and Γ vel These are the position and velocity thresholds, respectively. pos_base ,Γ vel_base s(r) and v(t) are the basic threshold values ​​for position and velocity, respectively. s(r) and v(t) are the direct proportional functions of position and velocity with respect to the interruption time t, respectively. That is, the longer the interruption time, the larger the function value.

3. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 1, characterized in that, In step 2, the reverse order rearrangement is performed, assuming the starting track X is to be started. j The sequence formed by the k original measurement points is Z. j ={z1, z2, ..., z k }, where z1 is the starting point of the track to be initiated. Reversing this sequence by timestamp order yields a new sequence Z′. j ={z′1,z′2,...,z′ k }={z k , z k-1 ,…,z1}.

4. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 1, characterized in that, In step 3, a parallel filter bank is established that includes three motion models: CV (uniform velocity), CA (uniform acceleration), and CT (uniform speed turning); the reverse-ordered sequence of flight track data Z′ to be started is processed. j Perform Kalman filtering and calculate the Gaussian likelihood value L of the k-th model. k : Among them, v k This is the information of the k-th model, S k It is its new information covariance.

5. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 4, characterized in that, In step 3, the adaptive weights ω of each model k Determined by the normalized Gaussian likelihood value: Where k = 1, 2, and 3 correspond to the three motion models CV, CA, and CT, respectively.

6. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 1, characterized in that, In step 4, the number of steps for predicting the rearranged flight paths to be launched is m. Among them, t new_start This represents the time of the first point in the pending batch of track data, which is the time of the last point in the pending batch of track data after reverse ordering. old_end This indicates the last point in the old track where the old track was interrupted. T is the radar point update cycle, and round() means rounding to the nearest integer. The weighted prediction point state is X pred,final The calculation formula is as follows: X pred,final =ω cv X′ pred，cv +oh ca X pred，ca +oh ct X pred，ct Where, X′ pred，cv 、X′ pred,ca 、X′ pred,ct These are the predicted values ​​of the CV, CA, and CT models after state alignment in the Cartesian coordinate system, and the velocity values ​​need to be inverted.

7. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 1, characterized in that, In step 5, the position ambiguity factor u1 and the velocity ambiguity factor u2 are first defined: Where, x i (l), y i (l), z i (l) represents the position of the break point on the x, y, and z axes of the old track i at time l, x j (l), y j (l) This represents the weighted predicted position of the x, y, and z axes of the trajectory j to be launched at time l. Similarly, Equal to the corresponding speed.

8. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 7, characterized in that, In step 5, the fuzzy membership function that is dynamically adjusted according to the interruption time is: Where, τ k For adjustment degree, u k Let σ be the k-th fuzzy factor in the fuzzy factor set. k To represent the extent of the fuzzy factors; g(t) is a nonlinear function of the interruption time t, meaning that the longer the interruption, the greater the influence of g(t), which is used to match the increase in prediction uncertainty.

9. The method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic according to claim 8, characterized in that, In step 5, the similarity function f between the old and new tracks ij The calculation formula is: Where a1 and a2 are the weights of the position and velocity ambiguity factors, respectively.

10. A method for resuming interrupted flight paths based on multi-model weighted prediction and time-varying fuzzy logic, characterized in that, include: The coarse association filtering module is used to filter out old tracks that may match the tracks to be batched based on the gate that is dynamically adjusted with the interruption time; the multi-model weighted prediction module is used to reverse the order of the track data to be batched, calculate the model weights using parallel filtering of multiple models such as CV, CA, and CT, and generate weighted inverse prediction points; the similarity calculation module is used to calculate the similarity between the weighted prediction points and the interruption points of the old tracks using the fuzzy membership function that is dynamically adjusted with the interruption time; the continuation processing module is used to determine whether to merge tracks and unify batch numbers based on the similarity calculation results.

Images