Flight path disconnection repairing method based on adaptive Kalman prediction

Abstract

The invention discloses a track disconnection repairing method based on adaptive Kalman prediction, and the method comprises the implementation steps: carrying out the preprocessing of a ship track data set, and building a constant-speed linear motion CV model of a target ship; backtracking a target ship trace point for tracking, and updating a covariance matrix; updating the target state equation by using the tracking covariance matrix to realize target trace point prediction; a nearest neighbor association NN algorithm is adopted, new plots appearing after plot prediction are detected through a rectangular wave door, track batch number association is carried out on the new plots, and self-adaptive track prediction is achieved. According to the method, the ship trace point is predicted in real time through the self-adaptive Kalman prediction algorithm at the broken position of the track, the trace point with the missing track in the prior art is repaired, the complexity of the algorithm is reduced, the continuity of the track is improved, the quality of the predicted trace point is improved, the prediction speed is increased, and the practicability and the application range are expanded.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and further relates to a track disconnection repair method based on adaptive Kalman prediction in the field of target tracking technology. The present invention can be used to repair the track of a target ship with interruption phenomenon monitored by a shore-based sea radar in real time. Background Art

[0002] At present, the ship tracks are mainly derived from radar, AIS and other sensor monitoring. Different monitoring methods differ in monitoring range and update frequency. For example, satellite monitoring has a large range but is not very timely. Radar can continuously monitor the target but has a limited monitoring range. In the sea radar monitoring scenario, affected by terrain, clutter and interference, when the target ship sails under the bridge or passes through the offshore wind farm area, the radar will not receive the echo of the target ship, resulting in track disconnection. In this case, a large number of disconnected tracks will cause data redundancy in the tracking system, reduce the continuity of the track, and the management and maintenance of the track will consume a lot of computing resources, affecting the tracking performance. In addition, the tracks originating from the same target are identified as different targets, which will lead to incorrect estimates of the target status and target number, which will have a negative impact on the situational awareness and assessment, resource coordination and allocation, and information interaction and fusion of the tracking system.

[0003] Harbin Institute of Technology disclosed a method for tracking maneuvering targets at sea on a non-stationary platform in its patent document "A method for tracking maneuvering targets at sea on a non-stationary platform" (application number CN 202311521216.8 application publication number CN 117233745 A). The implementation steps of this method are: (1) collecting state measurement values ​​of the maneuvering target at different time points; (2) performing track filtering based on target features, scoring and pruning the track hypothesis tree according to the target features, and using interactive multi-models to align the complex motion state of the target in real time; (3) In order to solve the problem of track loss, the point filling method is used to maintain the track to achieve the purpose of protecting the long track. In order to solve the problem of track interruption, the track segment association method is used to train and learn the data before and after the track interruption, and the interrupted track is predicted based on the training model to achieve track continuation. However, this method still has the following three shortcomings: first, training and learning on the data before and after the track interruption will lead to inaccurate actual predictions; second, the algorithm is only applicable to the case where the track is disconnected once, and is not applicable to the case where the track is disconnected multiple times; third, the algorithm used in this method is to patch between the new and old tracks. After the track is disconnected, the disconnected track cannot be patched in real time.

[0004] The 20th Institute of China Electronics Technology Group Corporation disclosed a multi-radar measurement delayed disordered filtering method in its patent application "A multi-radar measurement delayed disordered filtering method" (application number CN 202210559004.8 application publication number CN 114895299 A). The implementation steps of this method are: (1) The real-time processing center in the multi-radar fusion receives the measurement data of the condensed point track. If there is no measurement data at the current moment, it is considered that the measurement data is missing and a point filling operation is performed, that is, prediction is performed based on the track information to obtain the predicted measurement data, and the predicted measurement data is considered to be the measurement data at this moment; if there is measurement data at the current moment, compare the timestamp of the measurement data with the time value at the current moment to see if they are consistent. If they are consistent, the measurement data is subjected to data association processing to obtain the measurement data after the same target track association; if there is measurement data at the current moment ... Data, if the timestamp of the measurement data is less than the current moment, it is considered that there is a delay in the measurement data, that is, there is a measurement lag disorder problem; the delayed measurement data is also subjected to data association processing to obtain the delayed measurement data after the same target track association, and the delayed measurement data replaces the measurement data of the point supplemented at that moment; (2) the target state at the current moment is re-filtered, and the specific steps of the filtering process are as follows: (2a) initial filtering: that is, least squares filtering, using the current measurement and the measurement of the previous moment for coarse filtering; (2b) using the Kalman filtering method for fine filtering. (2c) re-filtering update stage: finally estimate the covariance matrix of the state, the filter value and the overall covariance at time t, and then perform the next moment processing, iterative update, until the tracking is completed. Since the least squares filtering method can use the measurement of the current moment and the previous moment for fitting filtering, the delayed measurement can directly replace the measurement data of the supplemented point in the filtering process, which can effectively solve the problem of delayed disordered measurement, make full use of the effective measurement data that arrives late, improve the filtering accuracy and make the track smoother. However, this method still has the following three shortcomings: First, predictions based on track information may be inaccurate and unreliable; second, timestamps are generally in milliseconds, and the timestamp difference is used as the point-filling criterion, which is too harsh; third, the algorithm used in this method is too cumbersome, and frequent single-step predictions will lead to low program efficiency. In the case of long track interruptions and many points that need to be repaired, real-time performance is difficult to guarantee. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and to propose a track disconnection repair method based on adaptive Kalman prediction, aiming to solve the problem that the ship's track is difficult to predict in real time and the prediction is inaccurate.

[0006] The technical idea for achieving the purpose of the present invention is to first determine if a certain track has three points whose radar target state is not 1 in succession (radar target state 1 indicates tracking state, 2 indicates extrapolation, and 3 indicates loss), then start prediction, and the number of prediction steps depends on the next point or prediction threshold of the newly appearing radar target state 1. During the prediction process, take the first 10 points whose radar target state is 1, establish a CV model, and update it with the Kalman filter algorithm to obtain the covariance matrix. Use the covariance matrix to update the target state equation to achieve prediction. For each prediction step, the algorithm obtains a point trace and establishes a rectangular wave gate with a distance of 300 meters from the point. If a new point trace is detected in the rectangular wave gate (the point trace must meet the radar target state of 1), then stop prediction and perform nearest neighbor data association NN (Nearest Neighbor, NN) with the new point. The core idea of ​​NN is: if there is only one measurement value that falls into the relevant gate, then the measurement value can be directly used for track update; but if more than one measurement falls into the relevant gate of the target, the candidate measurement value with the smallest statistical distance should be used for track update. When terminating the prediction, one case is that the rectangular gate detects the newly appeared point trace. Another case is that no new point trace is detected, then the prediction threshold is set, and the threshold is reached and stopped. The long short-term memory gate network LSTM (Long Short-Term Memory, LSTM) shows high accuracy in single-dimensional time series prediction (such as stock price prediction), and can also show strong prediction ability in multi-input and multi-output scenarios. For example, in ship track prediction, many researchers use the longitude, latitude, speed and heading of the ship as the input of the LSTM network model, and output the longitude and latitude information of the ship through the LSTM network layer and the fully connected layer. The results show that LSTM can effectively learn and identify the characteristic values ​​of the ship's trajectory, such as longitude, latitude, speed and sailing direction, and then accurately predict the position of the ship at the future moment. However, this only learns the characteristics of one ship track, and this track is a complete track. When there is no disconnection, the prediction is made at the end of the track. If the track is disconnected somewhere in the middle, LSTM will not be able to accurately predict the track at the disconnection point, and there will be a phenomenon of model divergence. On the other hand, if there are multiple different ship tracks, that is, in the scenario of multiple inputs and multiple outputs under multiple tasks, the LSTM network model is not applicable. First of all, if a certain track is to be accurately predicted, the track can be trained separately. If there are multiple tracks, the distinction between tracks cannot be solved, resulting in the LSTM network model being unable to distinguish, causing the divergence of the training model and the inability to accurately predict the track. If the track batch number is input into the network as a feature, although the network can distinguish different tracks, the track batch number can only be used as a unique feature of a certain ship, resulting in the algorithm not being universally adaptable.

[0007] The steps of the present invention include the following:

[0008] Step 1, input track data;

[0009] Step 2, read the track points of each ship at each time that have not been read;

[0010] Step 3, determine whether there are three consecutive points in the track point track at the time of reading whose radar tracking target status is not 1. If so, it is considered that there is an interruption in the track and step 4 is executed. Otherwise, step 2 is executed.

[0011] Step 4: Use the CV model to model the target ship with interrupted track, and obtain the target state equation and the observation equation of the radar system;

[0012] Step 5, starting from the interruption point of the track where the interruption phenomenon exists, retrace 10 points where the radar target state is 1 to perform Kalman tracking and update the covariance matrix;

[0013] Step 6, using the updated covariance matrix to update the target state equation to achieve single-step point trajectory prediction;

[0014] Step 7, determine whether the number of single-step point prediction steps is less than or equal to 120 steps. If so, execute step 8.

[0015] Otherwise, delete the predicted points and execute step 2;

[0016] Step 8, detecting whether the newly appeared point trace at the point trace of the single-step prediction falls within the rectangular wave gate, if so, executing step 9, otherwise, executing step 6;

[0017] Step 9: associate the single-step predicted point track with the newly appeared point track using the nearest neighbor NN data, update the track batch number, and implement track repair while adaptively predicting the track.

[0018] Compared with the prior art, the present invention has the following advantages:

[0019] First, the present invention establishes a target motion model by tracing back the point traces at the disconnection point, thereby overcoming the deficiency that the post-processing algorithm of the traditional track segment association algorithm TSA (Track Segment Association) is only applicable to the scenario where the target is disconnected once, so that the present invention improves the practicability of the algorithm.

[0020] Second, the present invention is applicable to multiple tracks with interruptions by individually modeling each track; in the target tracking process, the covariance rectangle is updated, and in the target prediction, the covariance matrix updated in the last step is used to update the target state equation, so as to achieve single-step prediction of ship tracks, and the track quality is better than LSTM. It overcomes the bottleneck of the difficulty in real-time prediction of ship tracks caused by the prior art LSTM algorithm, avoids the need for high hardware, and makes the algorithm of the present invention have low computational complexity, thereby effectively improving the practicality and scope of application.

[0021] Third, because the present invention predicts the ship's track in real time through an adaptive Kalman prediction algorithm at the point where the track is disconnected, the missing tracks of the prior art are repaired, and the predicted tracks are associated with the subsequent newly appearing tracks by nearest neighbors, so that the present invention improves the continuity of the track. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of an embodiment of the present invention;

[0023] Figure 2 Schematic diagram of interrupted track 1 and track 2 according to an embodiment of the present invention;

[0024] Figure 3 It is a repair result diagram of track 1 and track 2 predicted by adaptive Kalman in the simulation experiment of the present invention;

[0025] Figure 4 A track diagram in which there is no interruption phenomenon in the simulation experiment of the present invention;

[0026] Figure 5 It is a training loss curve diagram of the prior art LSTM track prediction method in the simulation experiment of the present invention;

[0027] Figure 6 It is a prediction diagram of the prior art LSTM track prediction method at the track disconnection point in the simulation experiment of the present invention;

[0028] Figure 7 It is a test graph of the prior art LSTM track prediction method in the simulation experiment of the present invention in the data set with all interrupted tracks. DETAILED DESCRIPTION

[0029] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0030] Reference Figure 1 , further describing the implementation steps of the embodiment of the present invention.

[0031] Step 1: Enter track data.

[0032] The point track data of each ship track at each moment received by the radar display and control device is entered into a hash table.

[0033] The point track data includes track batch number, tracking sequence number, timestamp, longitude, latitude, speed, heading, angle, and radar tracking target status.

[0034] In the embodiment of the present invention, track 1 and track 2 are used as measured track data.

[0035] Reference Figure 2 , the schematic diagrams of the interrupted track 1 and track 2 of the present invention are further described.

[0036] Figure 2 The horizontal axis in the figure represents longitude, and the vertical axis represents latitude. The red dots represent ship dots, and the background image uses the Satellite image that comes with MATLAB 2022b. Figure 2 (a) is a schematic diagram of the interrupted track 1. Figure 2 (b) is a schematic diagram of the interrupted track 2. Track 1's track batch number = 200108921, tracking sequence number = 2, timestamp = 193190464, longitude = 120.87564087°, latitude = 27.19651604°, speed = 0.583153 knots, heading = 302.637°, angle = 211.31°, radar tracking target status = 1. Track 2's track batch number = 200125341, tracking sequence number = 9, timestamp = 335852736, longitude = 120.78741455°, latitude = 27.44536209°, speed = 11.0799 knots, heading = 147.67°, angle = 264.09°, radar tracking target status = 1.

[0037] Step 2, read the track points of each ship at each time that have not been read yet.

[0038] Step 3, determine whether there are three consecutive points in the track point track at the time of reading whose radar tracking target status is not 1. If so, it is considered that there is an interruption in the track and step 4 is executed. Otherwise, step 2 is executed.

[0039] Step 4: Use the CV model to model the target ship with interrupted track and obtain the target state equation and the observation equation of the radar system.

[0040] The CV model is as follows:

[0041] X(k)=[x(k)v x (k)y(k)v y (k)] T

[0042]

[0043] Where X(k) represents the radar tracking target state vector at the kth radar sampling moment, x(k), v x (k),y(k),y x (k) represents the kth sampling moment of the radar, the x-axis position and velocity of the radar tracking target, and the y-axis position and velocity. The superscript T represents the transposition operation, and T represents the sampling time of the radar. x(k-1), v x (k-1),y(k-1),y x (k-1) represents the k-1th radar sampling moment, the radar tracks the x-axis position and velocity, the y-axis position and velocity of the target, and w 2×1 (k) represents the zero mean and variance of the kth radar sampling moment. White noise, diag(·) represents a diagonal matrix, J = [0.5, 1, 0.5, 1].

[0044] The steps of modeling the target ship with interrupted track are as follows:

[0045] The first step is to construct the target state vector observed by the radar according to the following formula:

[0046] y(k)=s(k)+v noise (k)

[0047] Where y(k) represents the target state vector observed by the radar at the kth radar sampling moment, s(k) represents the true state vector of the radar tracking target at the kth radar sampling moment, and v noise (k) represents the radar positioning error at the kth radar sampling moment, which has zero mean and variance White noise, sqrtm(·) represents the matrix square root operation,

[0048] The second step is to calculate the true state vector of the radar tracking target and the speed of the radar tracking target according to the following formula:

[0049]

[0050] v(k+1)=v(k)+T 0 a(k)

[0051] a(k)=u(k)+w(k)

[0052] Where s(k+1) represents the true state vector of the radar tracking target at the k+1th radar sampling time, v(k) represents the speed of the radar tracking target at the kth radar sampling time, T0 =5s, a(k) represents the acceleration of the radar tracking target at the kth radar sampling moment, v(k+1) represents the speed of the radar tracking target at the k+1th radar sampling moment, u(k) represents the control signal of the ship power system, which is the known maneuvering signal output, and w(k) represents the random acceleration caused by sea breeze and waves, which is zero mean and variance Independent of v noise (k) white noise,

[0053] In the third step, the radar tracking target state vector is as follows:

[0054] x(k+1)=Φx(k)+Bu(k)+Γw(k)

[0055] y(k)=Hx(k)+v noise (k)

[0056] Where x(k+1) represents the radar tracking target state vector at the k+1th radar sampling time, Φ represents the state transfer matrix, x(k) represents the radar tracking target state vector at the kth radar sampling moment, B and Γ represent the noise driving matrix, H represents the observation matrix, H =

[10] .

[0057] Step 5, starting from the interruption point of the track where the interruption phenomenon exists, backtrack 10 points where the radar target state is 1 to perform Kalman tracking and update the covariance matrix.

[0058] The steps of Kalman tracking are as follows:

[0059] The first step is to initialize the initial conditions of the given filter: P 0 ,Q 0 ,R 0 ,in, represents the estimated value of the target state at the initial moment, P 0 represents the initial covariance matrix, Q 0 represents the initial target process noise matrix, R 0 represents the initial observation noise matrix;

[0060] In the second step, the state estimate and the estimation error covariance are predicted according to the following formula:

[0061]

[0062] in, Represents the estimated value predictions, represents the estimated value of the radar tracking target state at the k-1th radar sampling moment, Fk-1 represents the radar tracking target state transfer matrix at the k-1th radar sampling moment, P kk-1 Indicates P k-1 The predicted value, P k-1 is the covariance matrix of the k-1th radar sampling moment, Q k-1 is the process noise variance matrix at the k-1th radar sampling moment;

[0063] In the third step, the measurement prediction, innovation covariance matrix and Kalman gain are calculated according to the following formula:

[0064]

[0065] in, Indicates estimated value predictions, represents the estimated value of the observed state of the target, H k represents the observation matrix at the kth radar sampling moment, S k represents the new information covariance matrix at the kth radar sampling moment, H k-1 represents the observation matrix at the k-1th radar sampling moment, R k represents the observation noise matrix at the kth radar sampling moment, K k represents the Kalman gain at the kth radar sampling moment;

[0066] The fourth step is to calculate the state estimate and the estimation error covariance according to the following formula:

[0067]

[0068] P k =P kk-1 -K k H k P kk-1 .

[0069] in, represents the estimated value of the radar tracking target state at the kth radar sampling moment, z k Represents the observed state value of the target.

[0070] Step 6: Use the updated covariance matrix to update the target state equation to achieve single-step point trajectory prediction.

[0071] The single-step point prediction is obtained by the following formula:

[0072] x k+1 =F k x k +P

[0073] Among them, x k+1represents the radar tracking target state value at the k+1th radar sampling moment, F k represents the radar tracking target state transfer matrix at the kth radar sampling moment, and P represents the covariance matrix updated for the last time during the tracking filtering process.

[0074] Step 7, determine whether the number of steps for single-step point trace prediction is less than or equal to 120 steps. If so, execute step 8. Otherwise, delete the predicted point trace and execute step 2.

[0075] Step 8, detect whether the newly appeared point trace at the single-step predicted point trace falls within the rectangular wave gate, if so, execute step 9, otherwise, execute step 6.

[0076] The rectangular wave gate is obtained by the following formula:

[0077]

[0078] Among them, v k,i represents the i-th component of the residual, |·| represents the absolute value operation, z k,i represents the i-th component of the measurement, z kk-1,i represents the i-th component of the measured predicted value, K G represents the gate constant, K G ≥3.5, i=1,2,...,n z , n z Indicates the measurement number.

[0079] Step 9: associate the single-step predicted point track with the newly appeared point track using the nearest neighbor NN data, update the track batch number, and implement track repair while adaptively predicting the track.

[0080] The nearest neighbor NN data association refers to using the nearest neighbor NN algorithm and, according to a discrimination criterion, taking the measurement that falls within the wave gate and is statistically closest to the target prediction position as a candidate measurement.

[0081] The discrimination criteria refer to any of the following situations:

[0082] Condition 1: If there is only one measurement in the gate of a target, then the target is associated with this measurement;

[0083] Condition 2: If a measurement falls into the gate of a target, then the measurement is associated with this target;

[0084] Condition 3: When a target's gate contains multiple measurements, the target is associated with the most recent measurement;

[0085] Condition 4: When a measurement falls within the gates of multiple targets, the measurement is associated with the closest target.

[0086] The technical effects of the present invention are further described below in conjunction with simulation experiments.

[0087] 1. Simulation experiment conditions:

[0088] The hardware platform of the simulation experiment of the present invention is: the processor is Inter(R)Core(TM)i7-12800HX CPU, the main frequency is 2.00GHz, the memory is 32GB, and the graphics card is RTX3070 Ti.

[0089] The software platforms for the simulation experiment of the present invention are: Windows 11 operating system, MATLAB 2022b and Pycharm2024.

[0090] The input data used in the simulation experiment of the present invention is Figure 2 Two tracks.

[0091] 2. Simulation content and result analysis:

[0092] The simulation experiment of the present invention uses the present invention and the prior art (long short-term memory gate network LSTM) to perform track repair on the input track data respectively, and the results are as follows: Figure 3 shown.

[0093] In the simulation experiment, an existing technology used is:

[0094] The prior art long short-term memory gate network LSTM method refers to the LSTM ship track prediction method proposed by Li Yong et al. in their published paper “LSTM Ship Track Prediction Model” (Computer Technology and Development vol.39, no.9, pp.149–154, 2021), referred to as the LSTM track prediction method.

[0095] Combine the following Figure 3 The simulation diagram of the present invention is further described.

[0096] Figure 3 It is the repair result diagram of track 1 and track 2 predicted by the adaptive Kalman method of the present invention, wherein: Figure 3 (a) is the patching result of track 1. Figure 3 (b) is the repair result of track 2. Figure 3 The horizontal axis represents longitude and the vertical axis represents latitude. Figure 3 The magenta point indicates the trace with the radar target state of 1, which is the true value; the black point indicates the trace with the radar target state of 2, which is the abnormal value; the orange point indicates the trace with the radar target state of 3, which is the termination of the track; the green point indicates the repaired track.

[0097] Figure 3 There are two partial enlarged pictures of A and B. Figure 3 (a) At point A in track 1, the algorithm can adaptively predict the target track and successfully detect the newly appeared point track. Figure 3 (a) At point B of track 1, the track is interrupted for a long time and there is only one point with a radar target state of 1 near the end of point B, which is somewhat challenging. However, the algorithm can accurately detect this point, further verifying the accuracy of the prediction of the algorithm of the present invention. Figure 3 (b) At point A of track 2, there is also a long track interruption interval. The algorithm predicts that it does not exceed 10 minutes, which is about 116 points. Figure 3 (b) At B in track 2, due to the short interval of interruption, the algorithm can adaptively predict and correctly associate it with the subsequent track. Figure 3 It can be found that the Kalman prediction algorithm of the present invention can well repair the disconnected track, which proves the effectiveness of the algorithm of the present invention.

[0098] Figure 4 This is a track diagram without interruption phenomenon in the present invention. Figure 4 The horizontal axis represents longitude and the vertical axis represents latitude. Figure 4 The red dots in the middle represent ship tracks, which are the original data. These ship tracks are obtained by Figure 3 (b) Track 2 is a complete track generated by artificially filling in points A and B, and there is no track interruption. In addition, the existing LSTM track prediction method can only learn the characteristics of complete tracks, but cannot learn the characteristics of interrupted tracks. The first 80% is used as the training set, and the last 20% is used as the test set. The data in the training set and the test set are normalized to ensure that the data is in the same range.

[0099] Use the Keras library to build a multi-input multi-output LSTM model, which includes an input layer, an LSTM layer, two fully connected layers, and two output layers. The output layer predicts lat and lon respectively. Compile the model, specify the loss function as MSE and the optimizer as Adam. The structure and parameters of the LSTM network are shown in Table 1, where the total number of trainable parameters is 80034.

[0100] Table 1 LSTM network layer structure and parameter table

[0101] Layer Output Shape Param InputLayer (None,10,7) 0 LSTM (None,128) 69632 Dense (None,64) 8256 Dense (None,32) 2080 lat(Dense) (None,1) 33 lon(Dense) (None,1) 33

[0102] Use the fit() method to train the model, pass in the training set and the test set, set epochs = 100 and batch_size = 512. In the simulation experiment of the present invention, you can choose to input about 10 ship track features, that is, the input layer step length is 10 steps, and predict 1 from 10 historical points. Different step lengths will have different degrees of influence on the accuracy of the network, the number of hidden layer neurons, and the accuracy of the experiment.

[0103] Figure 5 Training loss curve of the existing LSTM track prediction method. Figure 5 The horizontal axis in the graph represents the number of training rounds, and the vertical axis represents the loss value. The blue line represents the loss of the training set, and the green line represents the loss of the test set. Save the model after training.

[0104] Load the trained model, define the single-step prediction function, iterate the model prediction results, and examine the LSTM algorithm's prediction of the ship's track.

[0105] Figure 6 This is the prediction graph of the existing LSTM track prediction method at the track disconnection point. Figure 6 The horizontal axis in represents longitude, and the vertical axis represents latitude. The blue dots represent real traces, and the red dots represent predicted traces. At the point where the track is disconnected, LSTM prediction can predict the traces well, but the end of the prediction is far away from the adjacent real traces. Figure 6 It just shows the best result among many training and prediction experiments.

[0106] Figure 7 This is a test graph of the existing LSTM track prediction method in a dataset with all interrupted tracks. Figure 7 The horizontal axis in is longitude, and the vertical axis is latitude. In most cases, the LSTM network predicts that the interval between points will be large. Figure 7 The complete data set of the track is used for training, and there is no track disconnection. After the model is trained, it is loaded into the data set with interrupted tracks. Figure 7 The blue in the figure represents the real track, and the khaki represents the predicted track. At the point where the track is disconnected, the predicted khaki track has a large interval. The reason for this phenomenon is that the model has not fully learned the characteristic information such as ship speed.

[0107] against Figure 3(a) Track 1, similarly, it is necessary to manually fill in the interrupted points, make a data set, and input it into the LSTM network to prepare for subsequent training and prediction. This will make the training process very cumbersome and not conducive to engineering implementation. On the other hand, the LSTM prediction network model has more than 80,000 parameters, which further increases the time for model training, has certain requirements for computer equipment, and is not conducive to the real-time prediction of the track. Compared with the algorithm proposed in the present invention, the algorithm proposed in the present invention can better predict the points at the disconnection, and the intervals of the predicted points are uniform, and the speed is faster than the LSTM network.

Claims

1. A track disconnection repair method based on adaptive Kalman prediction, characterized in that: A uniform linear motion model of the target ship is established, and the target state equation is updated using the updated covariance matrix to achieve target point track prediction. The nearest neighbor association NN algorithm is used to achieve adaptive track prediction. The steps of the patching method include the following: Step 1, input track data; Step 2, read the track points of each ship at each time that have not been read; Step 3, determine whether there are three consecutive points in the track point track at the time of reading whose radar tracking target status is not 1. If so, it is considered that the track is interrupted and step 4 is executed. Otherwise, step 2 is executed. Step 4: Use the CV model to model the target ship with interrupted track, and obtain the target state equation and the observation equation of the radar system; Step 5, starting from the interruption point of the track where the interruption phenomenon exists, retrace 10 points where the radar target state is 1 to perform Kalman tracking and update the covariance matrix; Step 6, using the updated covariance matrix to update the target state equation to achieve single-step point trajectory prediction; Step 7, determine whether the number of steps of the single-step point trace prediction is less than or equal to 120 steps, if so, execute step 8, otherwise, delete the predicted point trace and execute step 2; Step 8, detecting whether the newly appeared point trace at the point trace of the single-step prediction falls within the rectangular wave gate, if so, executing step 9, otherwise, executing step 6; Step 9: associate the single-step predicted point track with the newly appeared point track using the nearest neighbor NN data, update the track batch number, and implement track repair while adaptively predicting the track.

2. The track disconnection repair method based on adaptive Kalman prediction according to claim 1 is characterized in that: The track data entry in step 1 refers to entering the point track data of each ship track received by the radar display and control device at each moment into the hash table; the point track data includes the track batch number, tracking sequence number, timestamp, longitude, latitude, speed, heading, angle, and radar tracking target status.

3. The track disconnection repair method based on adaptive Kalman prediction according to claim 1, characterized in that: The CV model described in step 4 is as follows: X(k)=[x(k)v x (k)y(k)v y (k)] T Where X(k) represents the radar tracking target state vector at the kth radar sampling moment, x(k),v x (k),y(k),y x (k) respectively represent the x-axis position and velocity of the radar tracking target at the kth sampling moment, The position and velocity of the y-axis. The superscript T indicates the transposition operation. T indicates the sampling time of the radar. x(k-1),v x (k-1),y(k-1),y x (k-1) represents the k-1th radar sampling moment, the radar tracks the x-axis position and velocity, the y-axis position and velocity of the target, and w 2×1 (k) represents the zero mean and variance of the kth radar sampling moment. White noise, diag(·) represents a diagonal matrix, J = [0.5, 1, 0.5, 1].

4. The track disconnection repair method based on adaptive Kalman prediction according to claim 1, characterized in that: The steps for modeling the target ship with interrupted track described in step 4 are as follows: The first step is to construct the target state vector observed by the radar according to the following formula: y(k)=s(k)+v noise (k) Where y(k) represents the target state vector observed by the radar at the kth radar sampling moment, s(k) represents the true state vector of the radar tracking target at the kth radar sampling moment, and v noise (k) represents the radar positioning error at the kth radar sampling moment, which has zero mean and variance White noise, sqrtm(·) represents the matrix square root operation, The second step is to calculate the true state vector of the radar tracking target and the speed of the radar tracking target according to the following formula: v(k+1)=v(k)+T0a(k) a(k)=u(k)+w(k) Where s(k+1) represents the true state vector of the radar tracking target at the k+1th radar sampling time, v(k) represents the speed of the radar tracking target at the kth radar sampling time, T0=5s, a(k) represents the acceleration of the radar tracking target at the kth radar sampling time, v(k+1) represents the speed of the radar tracking target at the k+1th radar sampling time, u(k) represents the control signal of the ship power system, which is the known maneuvering signal output, and w(k) represents the random acceleration caused by sea breeze and waves, which is zero mean and variance Independent of v noise (k) white noise, In the third step, the radar tracking target state vector is as follows: x(k+1)=Φx(k)+Bu(k)+Γw(k) y(k)=Hx(k)+v noise (k) Where x(k+1) represents the radar tracking target state vector at the k+1th radar sampling time, Φ represents the state transfer matrix, x(k) represents the radar tracking target state vector at the kth radar sampling moment, B and Γ represent the noise driving matrix, H represents the observation matrix, H = [1 0].

5. The track disconnection repair method based on adaptive Kalman prediction according to claim 1, characterized in that: The steps of Kalman tracking described in step 5 are as follows: The first step is to initialize the initial conditions of the given filter: in, represents the target state estimate at the initial moment, P0 represents the initial covariance matrix, Q0 represents the initial target process noise matrix, and R0 represents the initial observation noise matrix; In the second step, the state estimate and the estimation error covariance are predicted according to the following formula: in, Represents the estimated value predictions, represents the estimated value of the radar tracking target state at the k-1th radar sampling moment, F k-1 represents the radar tracking target state transfer matrix at the k-1th radar sampling moment, P kk-1 Indicates P k-1 The predicted value, P k-1 is the covariance matrix of the k-1th radar sampling moment, Q k-1 is the process noise variance matrix at the k-1th radar sampling moment; In the third step, the measurement prediction, innovation covariance matrix and Kalman gain are calculated according to the following formula: in, Indicates estimated value predictions, represents the estimated value of the observed state of the target, H k represents the observation matrix at the kth radar sampling moment, S k represents the new information covariance matrix at the kth radar sampling moment, H k-1 represents the observation matrix at the k-1th radar sampling moment, R k represents the observation noise matrix at the kth radar sampling moment, K k represents the Kalman gain at the kth radar sampling moment; The fourth step is to calculate the state estimate and the estimation error covariance according to the following formula: P k =P kk-1 -K k H k P kk-1 in, represents the estimated value of the radar tracking target state at the kth radar sampling moment, z k Represents the observed state value of the target.

6. The track disconnection repair method based on adaptive Kalman prediction according to claim 1, characterized in that: The single-step point prediction described in step 6 is obtained by the following formula: x k+1 =F k x k +P Among them, x k+1 represents the radar tracking target state value at the k+1th radar sampling moment, F k represents the radar tracking target state transfer matrix at the kth radar sampling moment, and P represents the covariance matrix updated for the last time during the tracking filtering process.

7. The track disconnection repair method based on adaptive Kalman prediction according to claim 1, characterized in that: The rectangular wave gate in step 8 is obtained by the following formula: Among them, v k,i represents the i-th component of the residual, |·| represents the absolute value operation, z k,i represents the i-th component of the measurement, z k|k-1,i represents the i-th component of the measured predicted value, K G represents the gate constant, K G ≥3.5, i=1,2,...,n z , n z Indicates the measurement number.

8. The track disconnection repair method based on adaptive Kalman prediction according to claim 1, characterized in that: The nearest neighbor NN data association in step 8 refers to using the nearest neighbor NN algorithm and, according to a discrimination criterion, taking the measurement that falls within the wave gate and is statistically closest to the target predicted position as a candidate measurement.

9. The track disconnection repair method based on adaptive Kalman prediction according to claim 8, characterized in that: The discrimination criteria refer to any of the following situations: Condition 1: If there is only one measurement in the gate of a target, then the target is associated with this measurement; Condition 2: If a measurement falls into the gate of a target, then the measurement is associated with this target; Condition 3: When a target's gate contains multiple measurements, the target is associated with the most recent measurement; Condition 4: When a measurement falls within the gates of multiple targets, the measurement is associated with the closest target.

Images