A Group Target Separation and Detection Method Based on Ellipsoid Expansion Shape

Abstract

The present invention relates to a method for detecting and judging the separation of group targets, and particularly to a method for detecting and separating group targets based on an ellipsoidal extended shape, belonging to the field of radar technology. First, the measurements in the detection space are clustered and the centroid parameters are estimated. Secondly, based on the Bayesian recursive method, the Kalman filtering criterion and the Markov chain, the motion state and the extended state of the group targets are filtered. Furthermore, a three-dimensional ellipsoid model is established to describe the extended shape of the group targets, and the volume change rate parameter between frames of the group targets is used to judge the separation state of the group targets. Then, the clustering-related parameters are updated according to the extended shape of the separated group targets. Finally, the separation detection of the group targets and the tracking of the separated group targets are realized. The method of the present invention has a good detection effect on the separation state of group targets in three-dimensional space, can realize timely and effective detection and judgment of the separation of group targets and stable tracking of the separated group targets, and verifies the effectiveness of the method by using simulation data.

Description

Technical Field

[0001] The present invention relates to a method for separating, detecting and judging group targets, in particular to a method for separating and detecting group targets based on an ellipsoidal extended shape, belonging to the field of radar technology. Background Art

[0002] The separation detection of group targets is a difficult problem in the fields of aerospace and information fusion. Group targets are a common type of target in radar space detection, presenting a dense multi-target state, mainly manifested as a multi-target set with highly similar spatial positions, motion states, and a relatively stable spatial structure between targets. Due to the influence of the scattering point positions of member targets in the group target, member targets with similar spatial positions that meet certain distribution conditions often present an indistinguishable state. At this time, the group target as a whole can be tracked. However, during the tracking process, when some member targets within the group target show a maneuvering state different from that of the group target due to different tasks, routes, etc., it will lead to the separation phenomenon of the group target, such as the separation of the booster in the ascending section from the rocket body, the split maneuver of the formation target, etc. If the separation phenomenon of the group target cannot be detected in time, it will seriously affect the parameter estimation of the group centroid and group members, and further affect the tracking of the separated group targets. Therefore, in order to accurately and timely judge the separation state of the group target and achieve stable tracking of the separated group targets, it is of great practical significance to study an effective method for separating and detecting group targets.

[0003] Traditional methods for separating and detecting group targets mainly judge the separation of group targets based on the position relationship and motion state changes between separated group targets. This separation detection method that infers the separation state of group targets from the motion model of separated group targets first depends on the setting of the association threshold and clustering-related parameters for group targets. Secondly, since the separated group targets themselves present a maneuvering state different from that of the group target, the traditional separation detection method can no longer meet the requirements of separating and detecting group targets and tracking separated group targets in a dense multi-target scenario with high complexity. In recent years, how to reasonably model the separation state of group targets has become a key issue in the research on separating and detecting group targets, and the change state of the extended shape of group targets has gradually become an important index for judging whether a group target is separated. As a mainstream idea for solving the problem of modeling the extended shape of group targets in recent years, the random matrix method usually uses an ellipse to model the extended shape of group targets in a two-dimensional space. In addition, existing research also mostly focuses on how to finely model the extended shape of group targets in a two-dimensional space using various different modeling methods. However, in actual application scenarios, group targets are in a three-dimensional space, and the modeling method of the extended shape of group targets in a two-dimensional space has become invalid, and many refined modeling methods are also difficult to achieve dimensional expansion. At the same time, the extended shape information obtained by existing modeling methods has not been fully utilized during the tracking process. Therefore, it is necessary to propose a method suitable for modeling the extended shape of group targets in a three-dimensional space to perform timely and effective separation judgment of group targets and achieve stable tracking of separated group targets. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: overcoming the deficiencies of the prior art, and proposing a method for separating and detecting group targets based on an ellipsoid-expanded shape. This method is used to solve the problem of separating and detecting group targets in an actual three-dimensional group target tracking scenario. It clusters the measurements in the detection space and estimates the centroid parameters, and based on the Bayesian recursive method, the Kalman filter criterion, and the Markov chain, it realizes the filtering of the motion state and the expansion state of the group target, and then establishes a three-dimensional ellipsoid model to describe the expanded shape of the group target. It uses the volume change rate parameter between frames of the group target to judge the separation state of the group target, and updates the clustering-related parameters according to the expanded shape of the grouped target, so as to realize the separation detection of the group target and the tracking of the grouped target.

[0005] The technical solution of the present invention is as follows:

[0006] A method for separating and detecting group targets based on an ellipsoid-expanded shape, the steps of which include:

[0007] Step S1, obtaining the three-dimensional space multi-scattering point group target measurements in the wideband radar tracking scenario frame by frame;

[0008] Step S2, using the density-based spatial clustering of applications with noise method to cluster the group target measurements obtained in Step 1 according to the clustering parameters, and estimating the centroid state parameters and group member parameters of the group target after clustering based on the mean clustering method;

[0009] Step S3, establishing a group target motion state model, and predicting and updating the group target motion state in the established group target motion state model based on the Bayesian recursive method, the Kalman filter criterion, and the Markov chain using the centroid state parameters of the group target after clustering estimated in Step S2, so as to obtain an updated group target motion state model;

[0010] Step S4, establishing a group target expansion state model, and predicting and updating the group target expansion state in the established group target expansion state model based on the Bayesian recursive method, the Kalman filter criterion, and the Markov chain using the group member parameters of the group target after clustering estimated in Step S2 and the updated group target motion state model in Step S3, so as to obtain an updated group target expansion state model;

[0011] Step S5, using the group target motion state in the updated group target motion state model obtained in Step S3 and the group target expansion state in the updated group target expansion state model in Step S4 to construct a group target expansion shape model based on the three-dimensional ellipsoid model, and obtaining the lengths of the three-dimensional axes of the ellipsoid of the established group target expansion shape model;

[0012] Step S6: Calculate the volume of the ellipsoidal extended shape of the group target using the lengths of the three-dimensional axes of the ellipsoid in the group target extended shape model constructed in Step S5, and calculate the volume change rate between two consecutive frames as an index for judging the separation state of the group target, thus completing the detection of group target separation based on the ellipsoidal extended shape;

[0013] Step S7: After completing the detection of group target separation based on the ellipsoidal extended shape, continue the tracking. During the tracking process, when the lengths of the three-dimensional axes of the ellipsoid in the group target extended shape model tend to be stable, update the clustering parameters used in clustering in Step S2 with the lengths of the three-dimensional axes of the ellipsoid in the stabilized group target extended shape model, and use the updated clustering parameters for clustering to estimate the centroid state parameters and group member parameters of the group target after clustering, and then perform Step S3 until the tracking task is completed.

[0014] In the aforementioned Step S3, when predicting the motion state of the group target in the current frame, the initial motion state is estimated based on the centroid state parameters of the group target associated with multiple consecutive frames during the track initiation process. After the track initiation is completed, for the group target that meets the track initiation conditions, the motion state of the group target in the current frame is predicted according to the group target motion state model using the updated motion state of the group target in the previous frame.

[0015] When updating the motion state of the group target in the current frame, update the motion state of the group target using the predicted motion state of the group target in the current frame and the centroid state parameters of the clustered group target associated with the current frame.

[0016] In the aforementioned Step S4, the group member parameters include the number of scatter points of the group member and the scatter point structure of the group member.

[0017] When predicting the extended state of the group target in the current frame, the initial extended state is estimated based on the motion state of the group target and the group member parameters associated with multiple consecutive frames during the track initiation process. After the track initiation is completed, for the group target that meets the track initiation conditions, the extended state of the group target in the current frame is predicted according to the group target extended state model using the updated extended state of the group target in the previous frame.

[0018] When updating the extended state of the group target in the current frame, update the extended state of the group target using the updated motion state of the group target in the current frame, the predicted extended state of the group target in the current frame, and the group member parameters in the current frame.

[0019] In the aforementioned Step S5, the method for constructing the group target extended shape model based on the three-dimensional ellipsoid model is as follows:

[0020] Step S51: Calculate the azimuth angle and elevation angle of the centroid position of the group target according to the updated motion state of the group target, and solve the rotation matrix of the ellipsoidal extended shape based on the calculated azimuth angle and elevation angle of the centroid position of the group target.

[0021] Step S52: Diagonalize and decompose the updated group target extended state according to the ellipsoid extended shape rotation matrix obtained in Step S51 to obtain the ellipsoid extended shape feature matrix;

[0022] Step S53: Perform eigenvalue decomposition on the ellipsoid extended shape feature matrix obtained in Step S52 to obtain the ellipsoid feature diagonal matrix, and obtain the three-dimensional axis lengths of the ellipsoid model according to the ellipsoid feature diagonal matrix;

[0023] In the said Step S6, the method for judging the separation state of the group target is as follows:

[0024] Step S61: Set the group target detection threshold. When the volume change rate is greater than or equal to the detection threshold, it is judged that the separation of the group target is detected, and go to Step S62; otherwise, it is considered that the group target has not entered the separation state, and the judgment of the current frame ends;

[0025] Step S62: Set the protection time for the group target separation state. Within the protection time range, set the group target separation threshold. When the volume change rate is less than or equal to the separation threshold, it is judged that sub-targets have separated from the original group target, and go to Step S63; otherwise (when the volume change rate is greater than the separation threshold), it is considered that no sub-targets have separated from the original group target, and the judgment of the current frame ends. If the protection time range is exceeded, the judgment of the current frame ends;

[0026] Step S63: After detecting the separation of the group target, establish an association gate according to the predicted motion state of the group target in the current frame of the group target and the maximum volume of the ellipsoid extended shape of the group target during the separation process, and regard the sub-targets within the association gate as the grouped targets to achieve track association;

[0027] In the said Step S7, the method for judging that the three-dimensional axis lengths of the ellipsoid of the group target extended shape model tend to be stable is as follows:

[0028] Step S71: Set the stable threshold of the group target extended shape and the continuous stable time of the extended shape. Within the set continuous stable time range, if the volume change rate of the group target is always lower than the stable threshold, it is judged that the group target extended shape tends to be stable, and go to Step S72; otherwise (when there is a moment when the volume change rate of the group target is higher than the stable threshold), it is considered that the group target extended shape still fluctuates, and the judgment of the current frame ends;

[0029] Step S72: When it is considered that the group target extended shape tends to be stable, update the neighborhood radius in the clustering parameter according to the three-dimensional axis lengths of the ellipsoid of the group target extended shape model; otherwise, continue to use the initialization parameters to achieve target clustering.

[0030] In the aforesaid step S2, the clustering parameters include the neighborhood radius Eps and the density threshold MinPts. The specific clustering method is as follows: Classify the measurement points of the group target according to the neighborhood radius Eps and the density threshold MinPts. First, conduct a neighborhood search for each measurement point, and divide the measurement points into reachable points, core points or noise points, and then complete the target clustering according to the algorithm criteria; Estimate the centroid state parameters after clustering, and perform multi-scattering point fusion based on the mean clustering method. The centroid state parameters after fusion are as follows:

[0031] In the aforesaid steps S3 and S4, establish the group target motion state x model and the extended state X model, and predict the group target motion state as follows: Update the group target motion state as follows:

[0032] Predict the group target extended state as follows:

[0033] Update the group target extended state as follows:

[0034] In the aforesaid step S5, calculate the azimuth angle θ azi and the elevation angle θ ele of the centroid position of the group target, and then obtain the ellipsoid extended shape rotation matrix A. The rotation matrix is as follows:

[0035]

[0036] Perform diagonalization decomposition on the group target extended state X, and the decomposition result is as follows: X = ACA T . It can be known from the decomposition that the ellipsoid extended shape feature matrix C can be solved according to the extended state X and the ellipsoid rotation matrix A. The feature matrix is as follows: C = A T XA; Perform eigenvalue decomposition on the ellipsoid extended shape feature matrix C, and the decomposition result is as follows: C = BΛB T The ellipsoid extended shape eigenvector matrix B and the eigenvalue diagonal matrix Λ can be obtained through eigenvalue decomposition. The eigenvalue diagonal matrix Λ = diag{a 2 , b 2 , c 2}, and a, b, and c are the three-dimensional axis lengths of the established ellipsoid model;

[0037] In the aforesaid step S6, calculate the volume V of the ellipsoid extended shape of the group target, and calculate the volume change rate l between two consecutive frames as an index for judging the separation state of the group target. Specifically:

[0038] Set the group target detection threshold TrD. When the volume change rate l≥TrD, it is determined that the group target is detected to be separated. Otherwise, it is considered that the group target is not in a separated state. Set the protection time for the group target separation state to k p , for k s ≤k≤k s +k p Within the time range, set the group target separation threshold to TrS. When the volume change rate l≤TrS, it is determined that sub-targets have separated from the original group target. Otherwise, it is considered that no sub-targets have separated from the original group target. After detecting the separation of the group target, based on the predicted motion state of the group target in the current frame and the maximum volume V of the group target ellipsoid expansion shape during the separation process max Establish an association gate, and regard the sub-targets within the association gate as grouped targets to achieve track association;

[0039] In the step S7 described above, set the group target expansion shape stability threshold TrE and the group target expansion shape continuous stability time k e , when there is a volume change rate l of the group target satisfying l≤TrE within consecutive k e time, it is determined that the group target expansion shape tends to be stable. Otherwise, it is considered that the group target expansion shape still fluctuates. When it is considered that the group target expansion shape tends to be stable, update the neighborhood radius Eps in the clustering parameter according to the three-dimensional axis lengths of the ellipsoid of the group target expansion shape model. Otherwise, continue to use the initial parameters to achieve target clustering.

[0040] The most prominent feature and significant beneficial effect of the present invention are:

[0041] (1) Based on the measurement information of the group target in three-dimensional space and the estimation of the motion state and expansion state of the group target during the tracking process, a model of the ellipsoid expansion shape of the group target is established. This ellipsoid expansion shape can effectively describe the expansion shape of the group target based on the distribution of target scattering points during the motion of the group target;

[0042] (2) Based on the obtained ellipsoid expansion shape of the group target, during the period when the states and structures of the group members are relatively stable, a stable estimation of the expansion shape of the group target can be achieved, and then the clustering parameters of the group target corresponding to this target in the tracking link can be effectively updated to improve the tracking accuracy of the group target; in events that may affect the change of the group target expansion shape among the group member targets, the change state of the group target expansion shape can be detected, and then through analyzing the change process of the group target expansion shape, such events can be detected in a timely and effective manner;

[0043] (3) Based on the relevant parameters in the obtained group target ellipsoid extended shape model, the volume change rate of the group target extended shape between frames is used as an important parameter for judging the separation state of the group target during the tracking process. Through continuous multi-frame refined group target separation detection and group target separation state judgment, timely and accurate group target separation detection is achieved, effectively reducing the tracking error of each sub-group target in the group target separation link and improving the tracking accuracy of the sub-group target. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a schematic flow chart of the technical solution of the present invention;

[0045] Figure 2 is the group target separation scenario in the ascending section of the simulation of the present invention;

[0046] Figure 3 is the group target measurement diagram corresponding to the simulation scenario of the present invention;

[0047] Figure 4 is the method of the present invention for Figure 3 the tracking results of the group target measurement data therein, including the group target track and the extended shape;

[0048] Figure 5 is Figure 4 the volume change rate curve of target 1 during the tracking process;

[0049] Figure 6 is Figure 4 the volume change rate curve of target 2 during the tracking process. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0051] A method for detecting group target separation based on an ellipsoid extended shape, as Figure 1 shown, the steps of the method include:

[0052] Step S1, obtaining the three-dimensional space multi-scattering point group target measurement in the broadband radar tracking scenario frame by frame;

[0053] Step S2, clustering the group target measurements obtained in step S1, and estimating the centroid state parameters and group member parameters of the clustered group target;

[0054] Step S3, establishing a group target motion state model, and using the centroid state parameters of the clustered group target estimated in step S2 to predict and update the group target motion state in the established group target motion state model, so as to obtain an updated group target motion state model;

[0055] Step S4: Establish a group target extended state model. Use the group member parameters of the clustered group targets estimated in Step S2 and the updated group target motion state model in Step S3 to predict and update the group target extended state in the established group target extended state model, and obtain an updated group target extended state model;

[0056] Step S5: Construct a group target extended shape model based on a three-dimensional ellipsoid model using the group target motion state in the updated group target motion state model obtained in Step S3 and the group target extended state in the updated group target extended state model in Step S4, and obtain the lengths of the three-dimensional axes of the ellipsoid of the established group target extended shape model;

[0057] Step S6: Calculate the volume of the group target ellipsoidal extended shape using the lengths of the three-dimensional axes of the ellipsoid in the group target extended shape model constructed in Step S5, and calculate the volume change rate between two consecutive frames as an index for judging the separation state of the group target, thus completing the detection of group target separation based on the ellipsoidal extended shape.

[0058] The following gives an example of detecting group target separation based on the ellipsoidal extended shape in a simulation scenario using the above method.

[0059] Embodiment

[0060] Simulate the separation scenario of space targets in the ascending stage. Assume that the total detection time of group targets in space is 100 s. The group targets move at a constant speed from 0 to 60 s. Starting from 60 s, the group targets separate. Target 1 breaks away from the original group target and moves with a constant acceleration, while Target 2 still maintains the constant speed motion state of the original group target. The simulation scenario is as Figure 2 shown. The measurement distribution corresponding to the motion trajectories of the group targets is as Figure 3 shown. Use the method proposed in the present invention to track the Figure 3 measurement data in, and the tracking results of the group targets with extended shapes are as Figure 4 shown. It can be observed that during the separation process of the group targets, there is a sudden increase in the extended shape of the group targets during the separation stage. After the separation is completed, the extended shapes of the separated group targets will change according to the different distributions of target scattering points. Through Figure 5 and Figure 6 the volume change rate curves of Target 1 and Target 2 in, it can be clearly seen that at 60 s, the volume change rate of the group targets exceeds the detection threshold, and the separation event of the group targets is detected. Subsequently, the volume change rate is lower than the separation threshold, indicating that the group targets have separated. Then, track the separated group targets respectively. Finally, the volume change rates of each separated group target tend to be stable, which also corresponds to the Figure 4 stable tracking and extended shape estimation states of the separated group targets shown.

[0061] In summary, the method proposed by the present invention can effectively detect group target separation events and stably track grouped targets.

[0062] The present invention may also have many other embodiments. The above are only the preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for separating and detecting group targets based on an ellipsoidal extended shape, characterized in that the steps of this method include: Step S1, obtain the three-dimensional space multi-scattering point group target measurement in the broadband radar tracking scene frame by frame; Step S2, cluster the group target measurements obtained in Step S1, and estimate the centroid state parameters and group member parameters of the group targets after clustering; Step S3, establish a group target motion state model, and use the centroid state parameters of the group targets after clustering estimated in Step S2 to predict and update the group target motion state in the established group target motion state model, obtaining an updated group target motion state model; Step S4, establish a group target extended state model, and use the group member parameters of the group targets after clustering estimated in Step S2 and the updated group target motion state model in Step S3 to predict and update the group target extended state in the established group target extended state model, obtaining an updated group target extended state model; Step S5, construct a group target extended shape model based on a three-dimensional ellipsoid model using the group target motion state in the updated group target motion state model obtained in Step S3 and the group target extended state in the updated group target extended state model in Step S4, and obtain the lengths of the three-dimensional axes of the ellipsoid of the established group target extended shape model; Step S6, calculate the volume of the ellipsoidal extended shape of the group target using the lengths of the three-dimensional axes of the ellipsoid in the group target extended shape model constructed in Step S5, and calculate the volume change rate between two consecutive frames as an index for judging the separation state of the group target, completing the separation detection of the group target based on the ellipsoidal extended shape; In the said Step S3, when predicting the group target motion state of the current frame, the initial motion state is estimated according to the centroid state parameters of the group targets associated with multiple consecutive frames during the track initiation process. After the track initiation is completed, for the group targets that meet the track initiation conditions, the group target motion state of the current frame is predicted according to the group target motion state model using the updated group target motion state of the previous frame; When updating the group target motion state of the current frame, the group target motion state is updated using the predicted group target motion state of the current frame and the centroid state parameters of the clustered group targets associated with the current frame; In the said Step S4, the group member parameters include the number of group member scattering points and the structure of group member scattering points; When predicting the group target extended state of the current frame, the initial extended state is estimated according to the group target motion state and group member parameters associated with multiple consecutive frames during the track initiation process. After the track initiation is completed, for the group targets that meet the track initiation conditions, the group target extended state of the current frame is predicted according to the group target extended state model using the updated group target extended state of the previous frame; When updating the group target extended state of the current frame, the group target extended state is updated using the updated group target motion state of the current frame, the predicted group target extended state of the current frame, and the group member parameters of the current frame; In the said Step S5, the method for constructing a group target extended shape model based on a three-dimensional ellipsoid model is: Step S51: Calculate the azimuth angle and pitch angle of the centroid position of the group target according to the updated group target motion state, and solve the ellipsoid extended shape rotation matrix based on the calculated azimuth angle and pitch angle of the centroid position of the group target. Step S52: Diagonalize and decompose the updated group target extended state according to the ellipsoid extended shape rotation matrix obtained in Step S51 to obtain the ellipsoid extended shape feature matrix. Step S53: Perform eigenvalue decomposition on the ellipsoid extended shape feature matrix obtained in Step S52 to obtain the ellipsoid feature diagonal matrix, and obtain the three-dimensional axis lengths of the ellipsoid model based on the ellipsoid feature diagonal matrix.

2. A method for detecting separation of group targets based on an ellipsoid extended shape according to claim 1, wherein: After completing the detection of separation of group targets based on the ellipsoid extended shape, continue with tracking. During the tracking process, when the three-dimensional axis lengths of the ellipsoid of the group target extended shape model tend to be stable, update the clustering parameters used in clustering in Step S2 with the three-dimensional axis lengths of the ellipsoid of the stable group target extended shape model, and use the updated clustering parameters to estimate the centroid state parameters and group member parameters of the group targets after clustering, and then perform Step S3 until the tracking task is completed.

3. A method for detecting separation of group targets based on an ellipsoid extended shape according to claim 1, wherein: In the said Step S6, the method for judging the separation state of the group target is: Step S61: Set the group target detection threshold. When the volume change rate is greater than or equal to the detection threshold, judge that the separation of the group target is detected, and enter Step S62; otherwise, consider that the group target has not separated, and end the judgment of the current frame. Step S62: Set the protection time for the separation state of the group target. Within the protection time range, set the group target separation threshold. When the volume change rate is less than or equal to the separation threshold, judge that there are sub-targets separated from the original group target, and enter Step S63; otherwise, consider that there are no sub-targets separated from the original group target, and end the judgment of the current frame. If the protection time range is exceeded, end the judgment of the current frame. Step S63: After detecting the separation of the group target, establish an association gate according to the predicted group target motion state of the group target in the current frame and the maximum volume of the ellipsoid extended shape of the group target during the separation process, and regard the sub-targets within the association gate as sub-group targets to achieve track association.

4. A method for detecting separation of group targets based on an ellipsoid extended shape according to claim 2, wherein: The method for judging that the three-dimensional axis lengths of the ellipsoid of the group target extended shape model tend to be stable is: Step S71: Set the stable threshold of the group target extended shape and the continuous stable time of the extended shape. Within the set continuous stable time range, if the volume change rate of the group target is always lower than the stable threshold, judge that the group target extended shape tends to be stable, and enter Step S72; otherwise, consider that the group target extended shape still fluctuates, and end the judgment of the current frame. Step S72: When it is considered that the group target extended shape tends to be stable, update the neighborhood radius in the clustering parameters according to the three-dimensional axis lengths of the ellipsoid of the group target extended shape model; otherwise, continue to use the initialization parameters to achieve target clustering.

5. A method for separating and detecting group targets based on an ellipsoidal extended shape according to claim 4, characterized in that: In the said step S2, the clustering parameters include the neighborhood radius Eps and the density threshold MinPts. The specific clustering method is as follows: Classify the measurement points of the group target according to the neighborhood radius Eps and the density threshold MinPts. First, perform a neighborhood search on each measurement point, and divide the measurement points into reachable points, core points or noise points, and then complete the target clustering according to the algorithm criteria; Estimate the centroid state parameters after clustering, and perform multi-scattering point fusion based on the mean clustering method. The centroid state parameters after fusion are as follows:

6. A method for separating and detecting group targets based on an ellipsoidal extended shape according to claim 5, characterized in that: In the aforesaid steps S3 and S4, a group target motion state x model and an extended state X model are established, and the group target motion state is predicted as follows: The group target motion state is updated as follows: The predicted extended state of the group target is as follows: Update the group target extension status as follows: In the said step S5, the azimuth angle θ of the centroid position of the group target is calculated according to the motion state of the group target azi and the pitch angle θ ele , and then the rotation matrix A of the ellipsoidal extended shape is obtained. The rotation matrix is as follows: Diagonalize and decompose the extended state X of the group target, and the decomposition result is as follows: X = ACA T , after decomposition, it is known that the ellipsoid expansion shape feature matrix C is solved according to the extended state X and the ellipsoid rotation matrix a, and the feature matrix is as follows: C = A T XA; perform eigenvalue decomposition on the ellipsoid expansion shape feature matrix C, and the decomposition result is as follows: C = BΛB T Through eigenvalue decomposition, the ellipsoid expansion shape feature vector matrix B and the eigenvalue diagonal matrix Λ can be obtained. The eigenvalue diagonal matrix A = diag{a 2 , b 2 , c 2}, where a, b, and c are the lengths of the three-dimensional axes of the established ellipsoid model.

7. A method for separating and detecting group targets based on an ellipsoidal extended shape according to claim 6, characterized in that: In the step S6, the volume V of the ellipsoidal extended shape of the group target is calculated, and the volume change rate l between two consecutive frames is calculated as an index for judging the separation state of the group target. Specifically: A group target detection threshold ThrD is set. When the volume change rate l≥ThrD, it is judged that the separation of the group target is detected; otherwise, it is considered that the group target has not entered a separation state. Set the protection time of the group target separation state to k p , for k s ≤ k ≤ k s + k p Within the time range, set the group target separation threshold to ThrS. When the volume change rate l ≤ ThrS, it is determined that a sub-target has been separated from the original group target; otherwise, it is considered that no sub-target has been separated from the original group target. After detecting the group target separation, according to the predicted group target motion state of the current frame of the group target and the maximum group target ellipsoid expansion shape volume V during the separation process max Establish an association gate, and regard the sub-targets within the association gate as grouped targets to achieve track association; Set the group target extended shape stability threshold ThrE and the group target extended shape continuous stability time k e , when there is a group target volume change rate l satisfying l ≤ ThrE within consecutive k e time, it is judged that the group target extended shape tends to be stable, otherwise it is considered that the group target extended shape still fluctuates; when it is considered that the group target extended shape tends to be stable, the neighborhood radius Eps in the clustering parameter is updated according to the ellipsoid three-dimensional axis length of the group target extended shape model, otherwise the initialization parameter is continued to achieve target clustering.

Images