Image plane segment trajectory association method combining forward and backward prediction with velocity weighting
By using a combination of front-back prediction and velocity weighting in the observation process of optical load satellites, the problem of difficulty in fragment trajectory association is solved, and more accurate trajectory association and target state recognition are achieved.
Patent Information
- Application Number
- CN202110509764.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-05-11
- Publication Date
- 2025-02-18
- Estimated Expiration
- 2041-05-11
AI Technical Summary
During the observation process, the fragment trajectory generated by optical load satellites due to factors such as sensor scheduling or cloud occlusion, which leads to errors in the estimation of the target number or errors in the overall state cognition.
The image plane fragment trajectory association method combined with front-back prediction and velocity weighting is adopted. By forward and backward filtering of the old trajectory set, the new trajectory set is backward and forward filtering, and the time window is selected using the velocity weighting idea. , construct the test quantity and perform hypothesis test to determine the trajectory correlation.
It effectively avoids further expansion of prediction errors, improves the accuracy of trajectory correlation, and can perform data correlation without establishing a target three-dimensional motion model, overcoming the limitation that optical loads can only measure angles.
Smart Images
Figure CN113158127B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of aerospace data technology, and in particular to an image plane segment trajectory association method combining forward and backward prediction with velocity weighting. Background Art
[0002] Since humans sent the first satellite into space in the 1950s, humans have never stopped exploring space; and with the further advancement of science and technology in the 21st century, there has been a boom in space exploration. Compared with radar payloads, optical payloads have unique observation bands and can provide more intuitive observation results, but they can only provide angle measurement information and cannot provide the three-dimensional position of the target. And because the observation field of optical payloads is limited, it is easy for fragmented trajectories to appear due to sensor scheduling or cloud cover. If these fragmented trajectories cannot be correctly associated, in the case of a single target, it will lead to an error in estimating the number of targets; in the case of multiple targets, it will lead to an error in the overall state. For this reason, the present invention provides an image plane fragment trajectory association method that combines forward prediction, backward prediction and velocity weighting to solve the above problems. Summary of the invention
[0003] The purpose of the present invention is to provide an image plane fragment trajectory association method combining forward and backward prediction with velocity weighting, so as to solve the problems raised in the above background technology regarding the fragment trajectories generated by factors such as sensor scheduling and cloud occlusion during the observation process of optical payload satellites, as well as the shortcomings of the current algorithm.
[0004] To achieve the above object, the present invention provides the following technical solution: a method for associating image plane fragment trajectories by combining forward and backward prediction with velocity weighting, the method comprising the following steps:
[0005] S1: Preprocess the trajectories in the scene to obtain a set of trajectories that have no intersection in time and space. At this time, the trajectory sets can be associated in chronological order from front to back. Take two trajectories as an example for discussion. The one in front is called the old trajectory, and the one in the back is called the new trajectory.
[0006] S2: Forward prediction of the old trajectory set and backward prediction of the new trajectory set:
[0007] S2.1: Forward filtering is performed on the trajectories in the old trajectory set, and backward filtering is performed on the trajectories in the new trajectory set. At this time, the filtering uses a two-dimensional uniform acceleration model. For the two-dimensional uniform acceleration model, it is assumed that the target is performing uniform acceleration motion in the two-dimensional state space. It is assumed that the state vector at this time is:
[0008] X CA (k)=[x(k),y(k),v x (k),v y (k),a x(k),a y (k)] T
[0009] where x(k),y(k),v x (k),v y (k),a x (k),a y (k) represents the position, velocity and acceleration of the target in the x and y directions at time k respectively. The state equation of the target is:
[0010]
[0011] where q CA (k) represents zero mean and variance is Gaussian white noise, which essentially represents the disturbance term of acceleration;
[0012] In the filtering process, the CA model corresponds to the observation model:
[0013]
[0014] Where Z(k) represents the quantity measured at time k, r p (k) represents the pixel error in the observation process, which is assumed to have zero mean and variance Gaussian distribution of
[0015] S2.2: Backward prediction is performed on the trajectories in the old trajectory set, and forward prediction is performed on the trajectories in the new trajectory set. At this time, the filtering uses a two-dimensional uniform velocity model. For the two-dimensional uniform velocity model, it is assumed that the target moves at a uniform speed in the two-dimensional state space. Assume that the state vector at this time is:
[0016] X CV (k)=[x(k),y(k),v x (k),v y (k)] T
[0017] where x(k),y(k),v x (k),v y (k) represents the position and velocity of the target in the x and y directions at time k, respectively. The state equation of the target is:
[0018]
[0019] Where T represents the sampling time, q CV (k) represents zero mean and variance is Gaussian white noise, which essentially represents the disturbance term of acceleration;
[0020] S3: Select the time window for the old trajectory set and the new trajectory set using the speed weighting idea to construct the test volume:
[0021] S3.1: For the i-th old trajectory and the j-th new trajectory, use the speed weighting idea to select the midpoint position of the inspection time window:
[0022]
[0023] in represents the midpoint position of the time window between the i-th old trajectory and the j-th new trajectory, represents the initial position of the time window between the i-th old trajectory and the j-th new trajectory, n w represents the time window size, represents the time interval between the i-th old trajectory and the j-th new trajectory, Represents the backward estimation of the jth new trajectory at time speed, Represents the forward estimate of the i-th old trajectory at time speed, norm(x) represents the modulus value of vector x, and floor(x) represents the integer part of the logarithm value x;
[0024] S3.2: After selecting the test time window, define the statistic in have To obey x Degrees of freedom x 2 Distribution, where n x Represents the dimension of x(k), in this case n x Consider only the position component, that is, take n x =2, sum the statistics in the time window, and the correlation test quantity between track i and track j is:
[0025]
[0026] Since the statistics at each moment are independent of each other, we have Subject to the degree of freedom n x ×n w x 2 distributed;
[0027] S4: Perform hypothesis testing on the obtained correlation test quantity. Within the hypothesis testing threshold, if there are multiple new and old trajectories, construct a test matrix and use the two-dimensional assignment method to give the correlation between the new and old trajectories:
[0028] S4.1: Perform hypothesis testing on the associated test quantity, i.e.
[0029] H0: Trajectory i and trajectory j are the same target trajectory;
[0030] H1: Trajectory i and trajectory j are not the same target trajectory;
[0031] Under the assumption that H0 is true, the significance level α is obtained at this time The threshold value Th 12 , the judgment criteria are:
[0032]
[0033] right Conduct hypothesis testing;
[0034] S4.2: If there is only one set of new and old trajectories, a hypothesis test can be used to determine whether they are the same target. If there are multiple trajectories, a test matrix is constructed:
[0035]
[0036] By traversing and solving the two-dimensional assignment problem and obtaining the matching of the minimum value, the optimal association between the old trajectory set and the new trajectory set can be obtained.
[0037] Preferably, in step S1, only a set of trajectories without intersection in time and space is selected.
[0038] Preferably, the time window n in step S3.1 is w Select 5.
[0039] Preferably, the significance level α in step S4.1 is selected as 0.95.
[0040] Preferably, the threshold value Th in step S4.1 is 12 Select 3.94.
[0041] Compared with the prior art, the present invention has the following beneficial effects:
[0042] 1. During satellite observation, there is no need to establish a three-dimensional motion model of the target. Data association can be performed directly through its image plane trajectory, avoiding the disadvantage that optical payloads can only measure angles.
[0043] 2. When selecting the time window, the combination of forward prediction, backward prediction and speed weighting is used. Compared with the existing technology, it can effectively avoid the further expansion of the prediction error and thus obtain a better correlation effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a flow chart of the present invention;
[0045] Figure 2 It is a schematic diagram of the new and old trajectories of the present invention;
[0046] Figure 3 This is a schematic diagram of selecting the midpoint of the time window of the present invention;
[0047] Figure 4 This is a schematic diagram of a single target simulation scenario of the present invention;
[0048] Figure 5 This is a schematic diagram showing the variation of the single target association accuracy with the interruption time of the present invention;
[0049] Figure 6 This is a schematic diagram of a multi-target scenario of the present invention;
[0050] Figure 7 This is a schematic diagram of a second multi-target scenario of the present invention;
[0051] Figure 8 Schematic diagram of the multi-target scenario three of the present invention. DETAILED DESCRIPTION
[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0053] See also Figure 1-8 The present invention provides a technical solution: a method for associating image plane fragment trajectories by combining forward and backward prediction with velocity weighting, characterized in that the method comprises the following steps:
[0054] S1: Preprocess the trajectories in the scene to obtain a set of trajectories that have no intersection in time and space. At this time, the trajectory sets can be associated in chronological order from front to back. Take two trajectories as an example for discussion. The one in front is called the old trajectory, and the one in the back is called the new trajectory.
[0055] S2: Forward prediction is performed on the old trajectory set, and backward prediction is performed on the new trajectory set, where the number of old trajectories is N1, the number of new trajectories is N2, and the duration of each old trajectory is assumed to be The duration of each new trajectory is where i∈{1,2,…,N1}, j∈{1,2,…,N2}, and have Since filtering is a discrete process, for the convenience of discussion, the time is discretized, and the duration of each old trajectory is: The duration of each new trajectory is and have
[0056] S2.1: Forward filtering is performed on the trajectories in the old trajectory set, and backward filtering is performed on the trajectories in the new trajectory set. At this time, the filtering uses a two-dimensional uniform acceleration model. For the two-dimensional uniform acceleration model, it is assumed that the target is performing uniform acceleration motion in the two-dimensional state space. It is assumed that the state vector at this time is:
[0057] X CA (k)=[x(k),y(k),v x (k),v y (k),a x (k),a y (k)] T
[0058] where x(k),y(k),v x (k),v y (k),a x (k),a y (k) represents the position, velocity and acceleration of the target in the x and y directions at time k respectively. The state equation of the target is:
[0059]
[0060] where q CA (k) represents zero mean and variance is Gaussian white noise, which essentially represents the disturbance term of acceleration;
[0061] In the filtering process, the CA model corresponds to the observation model:
[0062]
[0063] Where Z(k) represents the quantity measured at time k, r p (k) represents the pixel error in the observation process, which is assumed to have zero mean and variance Gaussian distribution of
[0064] S2.2: Backward prediction is performed on the trajectories in the old trajectory set, and forward prediction is performed on the trajectories in the new trajectory set. At this time, the filtering uses a two-dimensional uniform velocity model. For the two-dimensional uniform velocity model, it is assumed that the target moves at a uniform speed in the two-dimensional state space. Assume that the state vector at this time is:
[0065] X CV (k)=[x(k),y(k),v x (k),v y (k)] T
[0066] where x(k),y(k),v x (k),v y(k) represents the position and velocity of the target in the x and y directions at time k, respectively. The state equation of the target is:
[0067]
[0068] Where T represents the sampling time, q CV (k) represents zero mean and variance is Gaussian white noise, which essentially represents the disturbance term of acceleration;
[0069] S3: Select the time window for the old trajectory set and the new trajectory set using the speed weighting idea to construct the test volume:
[0070] S3.1: For the i-th old trajectory and the j-th new trajectory, use the speed weighting idea to select the midpoint position of the inspection time window:
[0071]
[0072] in represents the midpoint position of the time window between the i-th old trajectory and the j-th new trajectory, represents the initial position of the time window between the i-th old trajectory and the j-th new trajectory, n w represents the time window size, represents the time interval between the i-th old trajectory and the j-th new trajectory, Represents the backward estimation of the jth new trajectory at time speed, Represents the forward estimate of the i-th old trajectory at time speed, norm(x) represents the modulus value of vector x, and floor(x) represents the integer part of the logarithm value x;
[0073] S3.2: After selecting the test time window, define the statistic in have To obey x Degrees of freedom x 2 Distribution, where n x Represents the dimension of x(k), in this case n x Consider only the position component, that is, take n x =2, sum the statistics in the time window, and the correlation test quantity between track i and track j is:
[0074]
[0075] Since the statistics at each moment are independent of each other, we have Subject to the degree of freedom n x ×n w x 2 distributed;
[0076] S4: Perform hypothesis testing on the obtained correlation test quantity. Within the hypothesis testing threshold, if there are multiple new and old trajectories, construct a test matrix and use the two-dimensional assignment method to give the correlation between the new and old trajectories:
[0077] S4.1: Perform hypothesis testing on the associated test quantity, i.e.
[0078] H0: Trajectory i and trajectory j are the same target trajectory;
[0079] H1: Trajectory i and trajectory j are not the same target trajectory;
[0080] Under the assumption that H0 is true, the significance level α is obtained at this time The threshold value Th 12 , the judgment criteria are:
[0081]
[0082] right Conduct hypothesis testing;
[0083] S4.2: If there is only one set of new and old trajectories, a hypothesis test can be used to determine whether they are the same target. If there are multiple trajectories, a test matrix is constructed:
[0084]
[0085] By traversing and solving the two-dimensional assignment problem and obtaining the matching of the minimum value, the optimal association between the old trajectory set and the new trajectory set can be obtained.
[0086] In order to verify the effectiveness of the present invention, a single target scenario and a multi-target scenario are selected to compare the algorithm of the present invention with the reference [1] (reference [1] Sun Fuchao. Research on Passive Radar Target Tracking Technology [D]. Chengdu University of Electronic Science and Technology, 2017: 30-43, which gives a trajectory association algorithm based on forward estimation and backward prediction and a trajectory association algorithm based on forward prediction and backward estimation, respectively referred to as reference [1] Algorithm 1 and Algorithm 2 in subsequent simulations) Algorithm 1 and Algorithm 2 for comparative analysis. In the single target scenario, Figure 4 As shown in Figure 2, three scenarios were selected for the multi-target case. Scenario 1 is the case where two targets intersect; Figure 6 As shown, scenario 2 is the case where five targets are parallel, such as Figure 7 As shown in the figure; Scenario 3 is the situation where the five targets are both crossed and parallel, such as Figure 8 The number of Monte Carlo simulations is 100, and the association accuracy is used as the effect measurement standard. The association accuracy is defined as follows: where n totalRepresents the total number of trajectories in the scene, Represents the number of correctly associated tracks in the scene.
[0087] In a single-target scenario, the measurement noise of the X and Y coordinates in the scene is set to 10 (pixel) and the sampling frequency is 1 Hz. Figure 5 The result of the association accuracy changing with the interruption time can be seen from the figure that when the interruption time is short, the algorithm of the present invention and the algorithm 1 and algorithm 2 of the reference [1] can both guarantee a good association accuracy. When the interruption time becomes longer, the algorithm of the present invention is significantly better than the algorithm 1 and algorithm 2 of the reference [1], which proves the effectiveness of the algorithm of the present invention in the single target case.
[0088] In the multi-target scenario, the measurement noise of the X coordinate and Y coordinate in the scene is set to 10 (pixel), the sampling frequency is 1Hz, and the interruption time is 60s. The correlation results of scene 1, scene 2, and scene 3 are shown in the following table:
[0089]
[0090]
[0091] It can be seen that the proposed algorithm has certain advantages over Algorithm 1 and Algorithm 2 in the literature [1] in the three scenarios, which verifies the effectiveness of the proposed algorithm in multi-target scenarios.
[0092] The experimental results show that the algorithm of the present invention is not only applicable to single-target scenarios, but can also achieve good results in multi-target scenarios, indicating that the method of the present invention is effective.
[0093] Although the present invention has been described above with reference to the embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the various features in the embodiments disclosed by the present invention may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.
Claims
1. A method for associating image plane fragment trajectories by combining forward and backward prediction with velocity weighting, characterized in that: The method comprises the following steps: S1: Preprocess the trajectories in the scene to obtain a set of trajectories that have no intersection in time and space. At this time, the trajectory sets can be associated in chronological order from front to back. Take two trajectories as an example for discussion. The one in front is called the old trajectory, and the one in the back is called the new trajectory. S2: Forward prediction of the old trajectory set and backward prediction of the new trajectory set: S2.1: Forward filtering is performed on the trajectories in the old trajectory set, and backward filtering is performed on the trajectories in the new trajectory set. At this time, the filtering uses a two-dimensional uniform acceleration model. For the two-dimensional uniform acceleration model, it is assumed that the target is performing uniform acceleration motion in the two-dimensional state space. Assume that the state vector at this time is: X CA (k)=[x(k),y(k),v x (k),v y (k),a x (k),a y (k)] T where x(k),y(k),v x (k),v y (k),a x (k),a y (k) represents the position, velocity and acceleration of the target in the x and y directions at time k respectively. The state equation of the target is: where q CA (k) represents zero mean and variance is Gaussian white noise, which essentially represents the disturbance term of acceleration; In the filtering process, the CA model corresponds to the observation model: Where Z(k) represents the quantity measured at time k, r p (k) represents the pixel error in the observation process, which is assumed to have zero mean and variance Gaussian distribution of S2.2: Backward prediction is performed on the trajectories in the old trajectory set, and forward prediction is performed on the trajectories in the new trajectory set. At this time, the filtering uses a two-dimensional uniform velocity model. For the two-dimensional uniform velocity model, it is assumed that the target moves at a uniform speed in the two-dimensional state space. Assume that the state vector at this time is: X CV (k)=[x(k),y(k),v x (k),v y (k)] T where x(k),y(k),v x (k),v y (k) represents the position and velocity of the target in the x and y directions at time k, respectively. The state equation of the target is: Where T represents the sampling time, q CV (k) represents zero mean and variance is Gaussian white noise, which essentially represents the disturbance term of acceleration; S3: Select the time window for the old trajectory set and the new trajectory set using the speed weighting idea to construct the test volume: S3.1: For the i-th old trajectory and the j-th new trajectory, use the speed weighting idea to select the midpoint position of the inspection time window: in represents the midpoint position of the time window between the i-th old trajectory and the j-th new trajectory, represents the initial position of the time window between the i-th old trajectory and the j-th new trajectory, n w represents the time window size, represents the time interval between the i-th old trajectory and the j-th new trajectory, Represents the backward estimation of the jth new trajectory at time speed, Represents the forward estimate of the i-th old trajectory at time speed, norm(x) represents the modulus value of vector x, and floor(x) represents the integer part of the logarithm value x; S3.2: After selecting the test time window, define the statistic in have To obey x Degrees of freedom x 2 Distribution, where n x Represents the dimension of x(k), in this case n x Consider only the position component, that is, take n x =2, sum the statistics in the time window, and the correlation test quantity between track i and track j is: Since the statistics at each moment are independent of each other, we have Subject to the degree of freedom n x ×n w x 2 distributed; S4: Perform hypothesis testing on the obtained correlation test quantity. Within the hypothesis testing threshold, if there are multiple new and old trajectories, construct a test matrix and use the two-dimensional assignment method to give the correlation between the new and old trajectories: S4.1: Perform hypothesis testing on the associated test quantity, i.e. H0: Trajectory i and trajectory j are the same target trajectory; H1: Trajectory i and trajectory j are not the same target trajectory; Under the assumption that H0 is true, the significance level α is obtained at this time The threshold value Th 12 , the judgment criteria are: right Conduct hypothesis testing; S4.2: If there is only one set of new and old trajectories, a hypothesis test can be used to determine whether they are the same target. If there are multiple trajectories, a test matrix is constructed: By traversing and solving the two-dimensional assignment problem and obtaining the matching of the minimum value, the optimal association between the old trajectory set and the new trajectory set can be obtained.
2. The image plane segment trajectory association method combining forward and backward prediction with velocity weighting according to claim 1, characterized in that: In the step S1, only a set of trajectories without intersection in time and space is selected.
3. The image plane segment trajectory association method combining forward and backward prediction with velocity weighting according to claim 1, characterized in that: The time window n in step S3.1 w Select 5.
4. The image plane segment trajectory association method combining forward and backward prediction with velocity weighting according to claim 1, characterized in that: In step S4.1, the significance level α is selected as 0.
95.
5. The image plane segment trajectory association method combining forward and backward prediction with velocity weighting according to claim 1, characterized in that: The threshold value Th in step S4.1 12 Select 3.94.