Radar target interruption track point filling method based on weighted smooth spline

Through the weighted smoothing spline method, the sliding window prediction and sliding window smoothing of the radar target are combined to solve the problem of insufficient prior information for the interrupt completion of the radar target track, and improve the track Continuity and completion accuracy.

CN118939640AActive Publication Date: 2024-11-12XIDIAN UNIV
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411061361.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-12
Estimated Expiration
2044-08-05

AI Technical Summary

Technical Problem

When the prior information of the target motion is unknown or the prior information has a large deviation from the actual motion situation, it is difficult for existing radar target tracks to accurately complete the track, resulting in track interruptions and low extrapolation accuracy, affecting the reliability of tracking.

Method used

The weighted smoothing spline method is used to predict the sliding window and smooth the sliding window sections of the front and rear tracks in the disconnection area, and combine the prediction and smooth point traces to obtain the complete point traces and reduce the deviation between the complete point traces and the real state.

Benefits of technology

Without relying on the prior information of the target motion, the target's motion trend is accurately fitted, the track continuity is improved, and the error between the complete point trace and the real state is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939640A_ABST
    Figure CN118939640A_ABST
Patent Text Reader

Abstract

The invention discloses a radar target interrupted track point supplementing method based on a weighted smooth spline. The problem that the radar tracking reliability is reduced due to track interruption of a tracked target in the prior art is mainly solved. According to the implementation scheme, a weighted smoothing spline is used, and sliding window prediction and sliding window smoothing are carried out starting from track segments in front of and behind a target navigation cutoff area respectively; calculating the motion trend difference between the predicted trace point and the actual target trace point as well as between the smooth trace point and the actual target trace point, and obtaining a predicted trace point fusion weight and a smooth trace point fusion weight in the flight cutoff area; and according to the predicted trace point fusion weight and the smooth trace point fusion weight, fusing trace points located in the track interruption area in the predicted trace point and the smooth trace point to obtain a complemented trace point. The method does not need to utilize the prior information of the target motion, can effectively fit the real motion trend of the target, reduces the deviation between the complementation trace point of the tracked target in the flight cutoff area and the real state of the target, improves the track continuity of the tracked target, and can be used for radar target tracking.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of radar technology, in particular to a method for filling in points on a radar target interrupted track, which can be used for radar target tracking. Technical Background

[0002] Target tracking plays an important role in radar technology. The core of target tracking technology lies in real-time and accurate determination of the current position and complete motion trajectory of the target. With the increasing complexity of radar tracking environment, due to factors such as target obstruction or strong mobility of the target, it is easy to cause no corresponding radar measurement generation for the target at certain moments, or it is difficult to correctly associate the radar measurement corresponding to the target with the target during the data association process, resulting in track interruption. Target track interruption will have an adverse effect on the study of the motion characteristics of the tracked target and the situational awareness of the relevant personnel of the tracking scene, thereby affecting the reliability of radar tracking. Therefore, how to use the two tracks before and after the interruption area of ​​the target where the track is interrupted to complete the point track of the track interruption area, and the completed point track is as close as possible to the actual motion of the tracked target in the interruption area, so as to obtain the complete track of the target is an urgent problem to be solved.

[0003] The patent document with application number 202110381779.6 discloses a method for associating interrupted tracks based on Gaussian processes, which uses Gaussian processes to learn unknown functions of target motion, and predicts and reversely predicts the new and old track segments before and after the interrupted area respectively; then the Gaussian process is used to fuse the interrupted track prediction values ​​to obtain a complete track. Since this method needs to set the corresponding kernel function based on the prior motion information of the target, it is easy to fail to correctly obtain the target's true motion trend in scenarios where the prior information of the target motion is unknown or the known prior information is significantly different from the actual motion of the target, thereby affecting the track continuity of the tracked target.

[0004] The patent document with application number 201910495262.2 discloses a method for extrapolating tracks based on timestamps, which uses the constructed state transfer matrix and system noise information, combined with the prediction process of the probability hypothesis density filter to predict and extrapolate the probability hypothesis density of the target's reliable track, thereby obtaining the extrapolated track information. Since this method needs to set the state transfer matrix and system noise based on the prior motion information of the target, in scenarios where the prior information of the target motion is unknown or the known prior information is significantly different from the actual motion of the target, it is easy for the extrapolated track to deviate significantly from the actual motion of the tracked target, thereby affecting the accuracy of track extrapolation. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose a radar target interruption track filling method based on weighted smoothing splines, so as to fit the real motion trend of the target in the interruption area without using the prior information of the target motion, effectively reduce the deviation between the completed point track of the tracked target in the track interruption area and the real state of the target, and improve the track continuity of the tracked target.

[0006] The technical idea of ​​the present invention is: by using weighted smoothing spline curves, sliding window prediction and sliding window smoothing are performed on the track segments before and after the disconnected area, so as to fit the real movement trend of the target. By fusing the point tracks obtained by prediction in the disconnected area with the point tracks obtained by smoothing, the completed point tracks of the track interruption area are obtained, thereby effectively reducing the deviation between the completed point tracks of the tracked target and the real state of the target, and improving the track continuity of the tracked target.

[0007] According to the above ideas, the radar target interruption track filling method based on weighted smoothing spline of the present invention is characterized in that the implementation steps include the following:

[0008] (1) From the track segment t before the target disconnected area old Initially, the weighted smoothing spline method is used to sample l points pre The sliding window prediction is used to obtain the predicted state vector x pre ;

[0009] (2) Track segment t after the target disconnection area new Initially, the weighted smoothing spline method is used to sample l points smo Sliding window smoothing, get the smoothed state vector x smo ;

[0010] (3) Calculate the prediction point fusion weight w in the disconnected area old and smooth point fusion weight w new :

[0011] (3a) Calculate the predicted state vector x pre The point trace after the flight suspension area and t new The root mean square error RMSE of the corresponding point traces in nol ;

[0012] (3b) Calculate the smoothed state vector x smo The point trace before the flight suspension area and t old The root mean square error RMSE of the corresponding point traces in ool ;

[0013] (3c) Based on the results of step (3a) and step (3b), calculate the predicted point fusion weight w in the broken flight area oldand smooth point fusion weight w new :

[0014]

[0015] (4) Based on the predicted point fusion weight w old Fusion weight w with smooth traces new , for the predicted state vector x pre and the smoothed state vector x smo The points located in the track interruption area are fused to obtain the completed point x stc .

[0016] Compared with the prior art, the present invention has the following advantages

[0017] Firstly, since the present invention uses known point traces for sliding window prediction and sliding window smoothing, the sliding window prediction and sliding window smoothing processes do not need to rely on prior information of the target's motion, and thus the motion trend of the target can be fitted when the target motion information is missing.

[0018] Secondly, in the prediction and smoothing process, the present invention assigns corresponding weights to each point trace in the data window according to the time interval between each point trace and the point trace adjacent to the broken-flight area, so that the point traces in the data window close to the broken-flight area can be fully utilized, and the movement trend of the target in the broken-flight area can be accurately fitted.

[0019] Thirdly, the present invention effectively reduces the deviation between the completed point track of the tracked target and the actual state of the target by fusing the predicted point track with the smooth point track in the disconnected area, thereby improving the track continuity of the tracked target. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart for realizing the present invention;

[0021] Figure 2 A comparison of the interrupted track filling effect of the present invention and the existing method in the uniform moving target tracking scenario

[0022] Figure 3 Comparison of the interrupted track filling effect of the present invention and the existing method in the uniformly accelerated target tracking scenario DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work should all fall within the scope of protection of the present invention.

[0024] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the implementation scheme of the present invention to facilitate understanding, and the order of the step numbers is not limited.

[0025] Reference Figure 1 , the implementation steps of this example are as follows:

[0026] Step 1: From the track segment t before the target disconnected area old At the beginning, the number of sampling points is l pre The sliding window prediction is used to obtain the predicted state vector x pre .

[0027] The number of sampling points is l pre The sliding window means that the prediction data window input for each prediction is the current moment [t old x pre ] T l at the middle end pre That is, after the prediction is made, the prediction data window will slide to the right by one dot, so that the prediction data window input for the prediction covers the l points close to the prediction time. pre The predicted point trajectory can be close to the actual motion of the target at the predicted time.

[0028] The prediction refers to obtaining the fitting relationship between the point trace in the prediction data window and the corresponding time stamp by using a weighted smoothing spline curve, so as to obtain the next time point trace of the end point trace in the prediction data window. Through sliding window prediction, t old The estimated point trace after that is the predicted state vector x pre .

[0029] The specific implementation of this step is as follows:

[0030] (1.1) Select the prediction data window with the current prediction sliding number m

[0031]

[0032] In the formula T s is the radar scanning period, which refers to the time interval at which the beam repeatedly appears in the same azimuth. is a set of integers, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L pop The sliding window prediction and t new The number of points that overlap, x oi1 Yesold l at the middle end pre The column vector of points, x oim Yes old l at the middle end pre -(m-1) points composed of a column vector, x pre is a column vector used to store the predicted extrapolated traces, x pi is x pre l at the middle end pre The column vector composed of points;

[0033] (1.2) Set the weight w of each point in the prediction data window pre :

[0034]

[0035] In the formula, β pre is the predicted fading factor, that is, the points in the predicted data window with a larger time interval from the end of the window are given a lower weight, β pre ∈(0,1); It means β pre The bth power, are the timestamps of each point in the prediction data window, T represents the transpose of the matrix, k pn is the timestamp of the end point in the current prediction data window;

[0036] (1.3) Get the next point trace at the end of the data window:

[0037] (1.3.1) Set the objective function f1:

[0038]

[0039] in, Yes pre The i-th element in yes The i-th point in the m,pre (k) is the desired fitting relationship, λ is the smoothing penalty parameter, and f m ″ ,pre (k) is f m,pre The second derivative of (k);

[0040] (1.3.2) By minimizing the objective function f1, we can obtain the fitting relationship between the point trace in the current input prediction data window and the corresponding timestamp. m,pre (k):

[0041] f m,pre (k) = minf1;

[0042] (1.3.3) According to the fitting relationship fm,pre (k) Calculate k oe +m·T s Predicted target point trace at time

[0043]

[0044] (1.4) Predict the target point trace Add to the predicted state vector x pre The tail end of

[0045] (1.5) Repeat steps (1.1) to (1.4) for a total of n p times, to obtain a length of n p The predicted state vector x pre :

[0046]

[0047] Among them, p v is the predicted target point obtained during the v-th extrapolation,

[0048] n p =(k nbg -k oe ) / T s +L pop -1.

[0049] Step 2: The track segment t after the target disconnection area new At the beginning, the number of sampling points is l smo Sliding window smoothing, get the smoothed state vector x smo .

[0050] The number of sampling points is l smo The sliding window means that the smoothing data window input for each smoothing is the current moment [x smo t new ] T The frontmost l smo That is, after smoothing, the smoothing data window will slide to the left by one point, so that the smoothing data window input for smoothing covers the l points near the smoothing moment. smo The known point traces can make the smooth point trace close to the actual motion of the target at the smoothing moment.

[0051] The smoothing refers to obtaining the fitting relationship between the point trace in the smooth data window and the corresponding time stamp by using a weighted smoothing spline curve, thereby obtaining the point trace at the previous moment of the front-end point trace in the smooth data window. By sliding window smoothing, t new The previous estimated point trace, that is, the smoothed state vector x smo .

[0052] The specific implementation of this step is as follows:

[0053] (2.1) Select the smoothing data window with the current smoothing sliding times n

[0054]

[0055] In the formula T s is the radar scanning period, that is, the time interval at which the beam repeatedly appears in the same azimuth. is a set of integers, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L sop The sliding window smoothing and t old The number of points that overlap, x ni1 Yes new Middle front end l smo The column vector of points, x smo is a column vector used to store the points obtained by smoothing extrapolation, x nin Yes new Middle front end l smo -(n-1) points composed of a column vector, x si is x smo The frontmost l smo The column vector composed of points;

[0056] (2.2) Set the weight w of each point in the smooth data window smo :

[0057]

[0058] In the formula, β smo is the smooth fading factor, that is, the points in the smooth data window with a larger time interval with the front end of the window are given a lower weight, β smo ∈(0,1); Refers to β smo The jth power, are the timestamps of each point in the smoothed data window, T represents the transpose of the matrix, k sn It is the timestamp of the front-end trace data in the current smooth data window;

[0059] (2.3) Get the point trace of the previous moment of the front point trace of the data window:

[0060] (2.3.1) Set the objective function f2:

[0061]

[0062] in, Yes smo The zth element in is the zth point in the smoothed data window, f n,smo (k) is the desired fitting relationship, λ is the smoothing penalty parameter, and f n ' , ' smo (k) is f n,smo The second derivative of (k);

[0063] (2.3.2) By minimizing the objective function f2, we obtain the fitting relationship between the point trace in the current input smooth data window and the corresponding timestamp. n,smo (k):

[0064] f n,smo (k) = minf2;

[0065] (2.3.3) According to the fitting relationship f n,smo (k) Calculate k nbg -n·T s Smooth target trace at time

[0066]

[0067] (2.4) Smooth the target point trace Add to the smoothed state vector x smo The forefront of

[0068] (2.5) Repeat steps (2.1) to (2.4) for a total of n s times, to obtain a length of n s The smoothed state vector x smo :

[0069]

[0070] Among them, s c is the smooth target point trace obtained during the c-th extrapolation, n s =(k nbg -k oe ) / T s +L sop -1.

[0071] Step 3: Calculate the predicted state vector x pre The point trace after the flight suspension area and t new The root mean square error RMSE of the corresponding point traces in nol .

[0072] (3.1) Select the predicted state vector x pre The point x located after the flight suspension area pop :

[0073]

[0074] In the formula, p q is the estimated value of the predicted state obtained by the qth prediction, T represents the transpose of the matrix,

[0075] is a set of integers, T s is the radar scanning period, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L pop The sliding window prediction and t new The number of points that overlap;

[0076] (3.2) Select t new Zhong and x pop Points x with the same timestamp nol :

[0077]

[0078] Among them, xn w t new The w-th element in

[0079] (3.3) Calculate x pop With x nol The root mean square error between:

[0080]

[0081] in, is x pop The e-th element in is x nol The e-th element in .

[0082] Step 4: Calculate the smoothed state vector x smo The point before the disconnection area and t old The root mean square error RMSE of the corresponding point traces in ool .

[0083] (4.1) Select the smooth state vector x smo The point x before the disconnection area sop :

[0084]

[0085] In the formula, s a is the smoothed state estimate obtained by the ath smoothing, T represents the transpose of the matrix,

[0086] is a set of integers, T s is the radar scanning period, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L sop The sliding window smoothing and t old The number of points that overlap;

[0087] (4.2) Select t old Zhong and x sop Traces of traces with the same timestamp:

[0088]

[0089] Among them, xo y t old The yth element in

[0090] (4.3) Calculate x sop With x ool RMSE ool :

[0091]

[0092] in, is x sop The u-th element in is x ool The u-th element in .

[0093] Step 5: According to RMSE nol and RMSE ool , calculate the prediction point fusion weight w in the broken flight area old and smooth point fusion weight w new .

[0094] Since RMSE nol It can reflect the difference between the predicted target movement trend and the actual target movement trend, RMSE ool It can reflect the difference between the smooth target motion trend and the actual target motion trend, so by using RMSE nol and RMSE oolThe prediction point fusion weight w in the broken flight area can be calculated old and smooth point fusion weight w new , the formula is as follows:

[0095]

[0096] where w old +w new =1.

[0097] Step 6: According to the predicted point fusion weight w old and smooth point fusion weight w new , for the predicted state vector x pre and the smoothed state vector x smo The points located in the track interruption area are fused to obtain the completed point x stc .

[0098] (6.1) Select the predicted state vector x pre The point x in the broken flight area pbt :

[0099]

[0100] Among them, pr d is the forecast state estimate obtained from the d-th forecast, is a set of integers, T s is the radar scanning period, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the terminal trace data, T represents the transpose of the matrix;

[0101] (6.2) Select the smooth state vector x smo The point x in the broken flight area sbt :

[0102]

[0103] Among them, sm g is the smoothed state estimate obtained by the g-th smoothing,

[0104] (6.3) for x pbt With x sbt To merge, x pbt With x sbt The points with the same timestamp in the grid are weightedly added to obtain the completed points x after the disconnection area is fused. stc :

[0105]

[0106] in, is x pbt The hth element in is x sbt The hth element in

[0107] Because of w old and w new Satisfy w old +w new =1, and w old and w new It reflects the difference between the predicted target point trace, the smoothed target point trace and the actual target motion trend, so w is used old and w new x pre With x smo The points located in the track interruption area are fused, so that the completion point track process can make full use of the predicted target point tracks and smoothed target point tracks in the track interruption area, thereby effectively reducing the deviation between the completion point tracks of the tracked target and the actual state of the target, and improving the track continuity of the tracked target.

[0108] The effect of the present invention can be further illustrated by the following simulation test:

[0109] 1. Simulation conditions

[0110] Simulation test scenario 1: Use radar to track a target in a uniform motion scenario. The simulation duration is t = 250s. The target track interruption period is (100s, 150s). The initial signal-to-noise ratio (S / N) of the tracked target is 0 = 23dB. The initial position of the target is (x0, y0) = (150000m, 100000m). The target movement speed (v x ,v y )=(250m / s,250m / s), the covariance matrix of the target motion noise is diag(3(m / s 2 ) 2 ,3(m / s 2 ) 2 ). Radar pulse width τ=8×10 -7 s, radar antenna beam half-power width θ0 = 4°, radar scanning period T s =5s.

[0111] Simulation test scenario 2: Use radar to track a uniformly accelerated target. The simulation duration is t = 250s. The target track interruption period is (100s, 150s). The initial signal-to-noise ratio (S / N) of the tracked target is 0 = 23dB. The initial position of the target is (x0, y0) = (150000m, 100000m). The initial speed of the target is (v x0 ,v y0 )=(100m / s,100m / s), the acceleration of the target motion (a x ,a y )=(1(m / s 2 ),1(m / s 2 )), the covariance matrix of the target motion noise is diag(3(m / s 2 ) 2 ,3(m / s 2 ) 2 ). Radar pulse width τ=8×10 -7 s, radar antenna beam half-power width θ0 = 3°, radar scanning period T s =5s.

[0112] In simulation test scenarios 1 and 2, the radar generates observation values ​​of distance and azimuth, and both the distance and azimuth observation values ​​contain errors caused by zero-mean Gaussian white noise.

[0113] The standard deviation of the distance measurement noise is:

[0114]

[0115] Where ρ is the measured distance; ρ0 is the distance at the initial tracking time; S / N is the signal-to-noise ratio; (S / N)0 is the signal-to-noise ratio at the initial tracking time.

[0116] The standard deviation of the azimuth measurement noise is:

[0117]

[0118] The parameters used in sliding window prediction and sliding window smoothing in simulation test scenario 1 and simulation test scenario 2 are: the number of sliding window prediction sampling points l pre =20, number of sliding window smoothing sampling points l smo = 20, the number of overlapping points between the sliding window prediction and the track segment after the disconnection area L pop = 10, the number of overlapping points of sliding window smoothing and track segments before the disconnection area L sop =10, predicted fading factor β pre =0.97, smooth fading factor β smo =0.97.

[0119] The formula for calculating the root mean square error between the point completion result and the real point trace of the target in the broken-off area is:

[0120]

[0121] Where, L bro is the length of the target track interruption area, is the real point trace of the urth target in the disconnected area, It is the urth point in the completed point track of the interrupted track filling method.

[0122] 2. Simulation content

[0123] Simulation 1: Under the conditions of the above simulation test scenario 1, the interrupted track point filling method of the radar target based on binomial polynomial is used to fill in the interrupted track points in the scenario where the real point track of the interrupted area of ​​the uniformly moving target is unknown. The results are as follows: Figure 2 As shown in the figure, the horizontal axis is the X-dimensional position, and the vertical axis is the Y-dimensional position.

[0124] from Figure 2 It can be seen that in the scenario, the completed trace result of the present invention is closer to the real trace of the uniformly moving target in the disconnected area. The root mean square error RMSE between the real trace of the target in the disconnected area and the completed trace result can be calculated. cv1 =251.70m, and the completion point trace result of the existing radar target interruption track filling method based on binomial polynomial deviates from the real point trace of the uniformly moving target in the interruption area. The root mean square error RMSE between the real point trace of the target in the interruption area is cv2 =792.04m. Compared with the existing method, the root mean square error of the present invention is reduced by 540.34m.

[0125] Simulation 2, under the conditions of the above simulation test scenario 2, the interrupted track point filling method of the radar target based on binomial polynomial is used to fill in the interrupted track point in the scenario where the real point track of the interrupted area of ​​the uniformly accelerated moving target is unknown. The results are as follows: Figure 3 shown.

[0126] from Figure 3 It can be seen that in the described scenario, the completed trace result of the present invention is closer to the real trace of the uniformly accelerated target in the disconnected area. The root mean square error RMSE between the real trace of the target in the disconnected area and the completed trace result can be calculated. CA1 =329.43m, and the completion point trace result of the existing radar target interruption track filling method based on binomial polynomial obviously deviates from the real point trace of the uniformly accelerated moving target in the interruption area. Through calculation, it can be known that the root mean square error RMSE between it and the real point trace of the target in the interruption area is CA2=2332.59m. Compared with the existing method, the root mean square error of the present invention is reduced by 2003.16m

[0127] In summary, in the simulation scenario, compared with the existing radar target interruption track point filling method based on binomial polynomial, the completed point track results of the present invention are closer to the real point track of the target in the interruption area, and the root mean square error with the real point track of the target in the interruption area is lower, indicating that the present invention can effectively fit the real motion trend of the target without using prior information on the target motion, reduce the deviation between the completed point track of the tracked target in the interruption area and the real point track of the target, and improve the track continuity of the tracked target.

Claims

1. A radar target interruption track filling method based on weighted smoothing splines, characterized in that: (1) From the track segment t before the target disconnected area old Initially, the weighted smoothing spline method is used to sample l points pre The sliding window prediction is used to obtain the predicted state vector x pre ; (2) Track segment t after the target disconnection area new Initially, the weighted smoothing spline method is used to sample l points smo Sliding window smoothing, get the smoothed state vector x smo ; (3) Calculate the prediction point fusion weight w in the disconnected area old and smooth point fusion weight w new : (3a) Calculate the predicted state vector x pre The point trace after the flight suspension area and t new The root mean square error RMSE of the corresponding point traces in nol ; (3b) Calculate the smoothed state vector x smo The point trace before the disconnection area and t old The root mean square error RMSE of the corresponding point traces in ool ; (3c) Based on the results of step (3a) and step (3b), calculate the predicted point fusion weight w in the broken flight area old and smooth point fusion weight w new : (4) Based on the predicted point fusion weight w old and smooth point fusion weight w new , for the predicted state vector x pre and the smoothed state vector x smo The points located in the track interruption area are fused to obtain the completed point x stc .

2. The method according to claim 1, characterized in that: Step (1) From the track segment t before the target disconnected area old Initially, the weighted smoothing spline method is used to sample l points pre The sliding window prediction is implemented as follows: (1a) Select the prediction data window with the current prediction sliding number m In the formula T s is the radar scanning period, is a set of integers, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L pop The sliding window prediction and t new The number of points that overlap, x oi1 Yes old l at the middle end pre The column vector of points, x oim Yes old l at the middle end pre -(m-1) points composed of a column vector, x pre is a column vector used to store the predicted extrapolated traces, x pi is x pre l at the middle end pre The column vector composed of points; (1b) Set the weight w of each point in the prediction data window pre : In the formula, β pre is the predicted decay factor, Refers to β pre The bth power, T represents the transpose of the matrix, are the timestamps of each point in the prediction data window, k pn is the timestamp of the end point in the current prediction data window; (1c) Get the next point trace at the end of the data window: (1c1) Set the objective function f1: in Yes pre The i-th element in yes The i-th point in the m,pre (k) is the desired fitting relationship, λ is the smoothing penalty parameter, and f″ m,pre (k) is f m,pre The second derivative of (k); (1c2) By minimizing the objective function f1, the fitting relationship between the point trace in the current input prediction data window and the corresponding timestamp is obtained. m,pre (k): f m,pre (k)=min f1; (1c3) According to the fitting relationship f m,pre (k) Calculate k oe +m·T s Predicted target point trace at time (1d) Predict the target point trace Add to the predicted state vector x pre The tail end of (1e) Repeat steps (1a) to (1d)n p times, to obtain a length of n p The predicted state vector x pre : where p v is the predicted target point obtained during the v-th extrapolation, n p =(k nbg -k oe ) / T s +L pop -1.

3. The method according to claim 1, characterized in that The track segment t after the target disconnection area in step (2) new Initially, the weighted smoothing spline method is used to sample l points smo The sliding window smoothing is implemented as follows: (2a) Select the smoothing data window with the current smoothing sliding times n In the formula T s is the radar scanning period, is a set of integers, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L sop The sliding window smoothing and t old The number of points that overlap, x ni1 Yes new Middle front end l smo The column vector of points, x smo is a column vector used to store the points obtained by smoothing extrapolation, x nin Yes new Middle front end l smo -(n-1) points composed of a column vector, x si is x smo The frontmost l smo The column vector composed of points; (2b) Set the weight w of each point in the smoothing data window smo : In the formula, β smo is the smoothing fading factor, Refers to β smo The jth power, T represents the transpose of the matrix, are the timestamps of each point in the smoothed data window, k sn It is the timestamp of the front-end trace data in the current smooth data window; (2c) Get the trace of the previous moment of the front trace of the data window: (2c1) Set the objective function f2: in, Yes smo The zth element in is the zth point in the smoothed data window, f n,smo (k) is the desired fitting relationship, λ is the smoothing penalty parameter, and f″ n,smo (k) is f n,smo The second derivative of (k); (2c2) By minimizing the objective function f2, the fitting relationship between the point trace in the current input smooth data window and the corresponding timestamp is obtained. n,smo (k): <h2 style=";text-align:left;direction:ltr">f<h2 style=";text-align:left;direction:ltr"> n,smo <h2 style=";text-align:left;direction:ltr"> (k) = min f2 (2c3) According to the fitting relationship f n,smo (k) Calculate k nbg -n·T s Smooth target trace at time (2d) Smooth the target points Add to the smoothed state vector x smo The forefront of (2e) Repeat steps (2a) to (2d) for a total of n s times, to obtain a length of n s The smoothed state vector x smo : Among them, s c is the smooth target point trace obtained during the c-th extrapolation, n s =(k nbg -k oe ) / T s +L sop -1.

4. The method according to claim 1, characterized in that: In step (3a), the predicted state vector x is calculated pre The point trace after the flight suspension area and t new The root mean square error RMSE of the corresponding point traces in nol , implemented as follows: (3a1) Select the predicted state vector x pre The point x located after the flight suspension area pop : In the formula, T represents the transpose of the matrix, p q is the estimated value of the predicted state obtained by the qth prediction, is a set of integers, T s is the radar scanning period, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L pop The sliding window prediction and t new The number of points that overlap; (3a2) Select t new Zhong and x pop The corresponding point x nol : where xn w t new The w-th element in (3a3) Calculate x pop With x nol The root mean square error between: in, is x pop The e-th element in is x nol The e-th element in .

5. The method according to claim 1, characterized in that In step (3b), the smoothed state vector x is calculated smo The point trace before the flight suspension area and t old The root mean square error RMSE of the corresponding point traces in ool , implemented as follows: (3b1) Select the smooth state vector x smo The point x before the disconnection area sop : In the formula, T represents the transpose of the matrix, s a is the smoothed state estimate obtained by the a-th smoothing, is a set of integers, T s is the radar scanning period, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the end point data, L sop The sliding window smoothing and t old The number of points that overlap; (3b2) Select t old Zhong and x sop The corresponding traces: Among them, xo y t old The yth element in (3b3) Calculate x sop With x ool RMSE ool : In the formula, is x sop The u-th element in is x ool The u-th element in .

6. The method according to claim 1, characterized in that In step (4), the predicted point fusion weight w old and smooth point fusion weight w new , for the predicted state vector x pre and the smoothed state vector x smo The point tracks in the track interruption area are fused as follows: (4a) Select the predicted state vector x pre The point x in the broken flight area pbt : Among them, pr d is the forecast state estimate obtained from the d-th forecast, is a set of integers, T s is the radar scanning period, k nbg Yes new The timestamp of the frontmost trace in k oe Yes old The timestamp of the terminal trace data, T represents the transpose of the matrix; (4b) Select the smooth state vector x smo The point x in the broken flight area sbt : Among them, sm g is the smoothed state estimate obtained by the g-th smoothing, (4c) for x pbt With x sbt To merge, x pbt With x sbt The points with the same timestamp in the grid are weightedly added to obtain the completed points x after the disconnection area is fused. stc : in, and They are x pbt With x sbt The hth element in w old and w new They are respectively the predicted point trace fusion weight and the smoothed point trace fusion weight calculated in step (3c).

Citation Information

Patent Citations

  • Short-term flight path prediction method, device and system based on motion state of unmanned aerial vehicle

    CN115577511A

  • User positioning method

    CN116528166A

  • Highway vehicle trajectory reconstruction method and system

    CN117334051A