Adaptive clustering based group target data association method and system

Abstract

The application provides a kind of array group target data association method and system based on adaptive clustering, comprising: accumulating same load data;Each load data is clustered, and initial array group division is obtained;Single cluster target set is screened;According to array group determination condition, remove non-array group target;The characteristics of array group target of two time points before and after are described respectively, and the characteristics of formation and formation are modeled;According to the array group target association matching model of the characteristics of formation and formation, the characteristics of formation and formation, the corresponding basic probability distribution function is calculated respectively;Evidence synthesis is carried out, and the comprehensive basic probability distribution function is established;Based on two-dimensional distribution, the comprehensive basic probability distribution function is associated with decision, and the array group target matching result is obtained.The application can effectively reduce the influence of array group target motion state change, sensor's missed detection and observation error, and significantly improve the correct association rate of target.

Description

Technical Field

[0001] This invention relates to the field of data association technology, and more specifically, to a method and system for associating cluster target data based on adaptive clustering. Background Technology

[0002] Due to the limitations of their orbital cycles, satellite visits to the monitored area are intermittent, with long intervals between visits, typically on the order of hours. Therefore, the revisit rate of satellite information is low. Furthermore, the acquired target information is subject to uncertainty and incompleteness due to changes in the motion state of targets within the satellite array, sensor misses, and observation errors. Under these conditions, it is difficult to achieve stable and effective identification and data correlation for every target within the monitored area.

[0003] Patent document CN117310642A (application number: CN202311239149.0) discloses a method for associating the tracks of multiple radar dense cluster targets based on multi-scale clustering. The method comprises the following steps: Step 1: Acquire radar sampling data and preprocess the data; Step 2: Estimate the track velocity based on the preprocessed data to obtain the velocity of the track point at each processing time; Step 3: Obtain the predicted landing point based on the target position and track point velocity, project the track onto a latitude-longitude network, and perform coarse clustering based on connected component clustering to obtain target clusters; Step 4: For each target cluster, determine the track similarity within the cluster using a weighted Hausdorff distance metric to obtain a fine clustering result, i.e., the preliminary association relationship between the two tracks; Step 5: Based on the preliminary association relationship in Step 4, select the path with the minimum weighted distance between the two targets as the optimal association relationship.

[0004] Patent document CN108665478A (application number: CN201810450301.2) discloses a multi-target data association method in complex scenes, characterized by the following steps: Step 1: Obtain the coordinate information of all target points in the binary image after detection of the current frame image, and count the number of target points; Step 2: Perform data point clustering on the coordinate information of the target points to convert it into multiple target blocks; Step 3: Then, after data registration of the target blocks, perform elimination judgment on the information of all target blocks; Step 4: Number the target blocks that meet the requirements; Step 5: Select the number of the target block in the image for tracking, that is, use the position information of the target block for tracking.

[0005] Traditional target recognition and tracking algorithms have significant limitations when dealing with situations where target location information is sparse and location accuracy errors are large, making it difficult to achieve stable and effective identification and data association for each target in the monitored area. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for associating cluster target data based on adaptive clustering.

[0007] The adaptive clustering-based target data association method provided by the present invention includes:

[0008] Step S1: Divide the region, obtain the grid division boundary threshold, and accumulate the target location information obtained by the same load within the region;

[0009] Step S2: Cluster the payload data in the order they enter the field of view to obtain the initial array group division;

[0010] Step S3: Filter the individual clustering target set and remove target points whose similarity does not meet the preset requirements;

[0011] Step S4: Remove non-array targets based on the array group determination criteria;

[0012] Step S5: Describe the target composition features of the array group at the two time points before and after the current time.

[0013] Step S6: Describe the formation characteristics of the target group at the two time points before and after the current time.

[0014] Step S7: Model the feature vectors of the target group composition features and formation features at the two time points before and after the current time;

[0015] Step S8: Calculate the corresponding basic probability allocation function based on the target association matching model of the array group with compiled features;

[0016] Step S9: Calculate the corresponding basic probability allocation function based on the target association matching model of the formation features;

[0017] Step S10: Calculate the corresponding basic probability allocation function based on the target association matching model of the array group according to the external topological structure features;

[0018] Step S11: Using DS evidence theory, evidence is synthesized from the basic probability allocation function based on the correlation of composition, formation and external topology features to establish a comprehensive basic probability allocation function;

[0019] Step S12: Based on the two-dimensional allocation, perform association decision on the basic probability allocation function of the synthesis to obtain the target matching result of the array group.

[0020] Preferably, step S2 includes:

[0021] Step S2.1: Calculate the distance between every two targets and construct a distance matrix;

[0022] Step S2.2: Find the threshold for density calculation based on the distance matrix, where the average number of points around each point whose distance is less than the density threshold accounts for 1%-4% of the total number of points;

[0023] Step S2.3: Calculate the local density of each target point and construct the local density vector;

[0024] Step S2.4: Calculate the nearest neighbor distance for each target point and construct the nearest neighbor distance vector;

[0025] Step S2.5: Adaptively obtain cluster centers;

[0026] Step S2.6: Classify the non-cluster center data points;

[0027] Step S2.2 includes:

[0028] Step S2.2.1: Parameter settings, including the upper and lower limits of the number percentage threshold, and the distance range threshold;

[0029] Step S2.2.2: Set the initial density threshold to the average of the maximum distance and the minimum distance;

[0030] Step S2.2.3: Calculate the percentage of points with a density less than the threshold;

[0031] Step S2.2.4: If the proportion is within the set range, output the current density threshold; otherwise, update the current maximum distance, minimum distance, and density threshold.

[0032] Step S2.2.5: If the current distance range is less than the set threshold, output the current density threshold; otherwise, return to step S2.3 for iterative calculation.

[0033] Step S2.4 includes:

[0034] Step S2.4.1: Sort the local density vectors in descending order of density value;

[0035] Step S2.4.2: Calculate the distance between each point and other points with a higher density than it.

[0036] Step S2.4.3: Obtain the index of the nearest neighbor of each point and the distance between the two points;

[0037] Step S2.5 includes:

[0038] Step S2.5.1: Calculate the joint density vector based on the local density vector and the nearest neighbor distance vector;

[0039] Step S2.5.2: Calculate the mean and variance of the joint density vector;

[0040] Step S2.5.3: Traverse each data point and take the points that exceed the set threshold range as cluster centers;

[0041] Step S2.6 includes:

[0042] Step S2.6.1: Initialize the cluster number of each data point, and set the cluster number of non-cluster centers to -1;

[0043] Step S2.6.2: Sort the local density vectors in descending order of density value;

[0044] Step S2.6.3: Update the cluster number of the non-cluster center data point to be the same as the cluster number of the point that is closest to it and has a higher density.

[0045] Preferably, step S3 includes:

[0046] Step S3.1: Obtain the corresponding coordinates based on the cluster center index;

[0047] Step S3.2: Based on the cluster center and the target cluster number, obtain the position coordinates of each data point in the same cluster;

[0048] Step S3.3: Calculate the local density of each point in a single cluster;

[0049] Step S3.4: Calculate the similarity between each data point in the same cluster and the cluster center;

[0050] Step S3.5: Calculate the comprehensive similarity of each data point in the same cluster based on similarity and local density;

[0051] Step S3.6: Calculate the mean and variance of the combined similarity vector of data points in a single cluster;

[0052] Step S3.7: Add data points whose comprehensive similarity value exceeds the threshold range as new cluster centers and update their cluster numbers.

[0053] Preferably, step S4 includes:

[0054] Step S4.1: Obtain the target belonging to the same cluster based on the cluster number corresponding to each data point;

[0055] Step S4.2: Assemble data for individual clusters that meet the set cluster construction threshold;

[0056] Step S4.3: Data points that do not meet the conditions for array group formation are treated as isolated points and no further array group matching is performed.

[0057] Preferably, step S5 includes:

[0058] Step S5.1: Calculate the target type number for each target in each array group;

[0059] Step S5.2: Calculate the number of members of each target type in each array group.

[0060] Preferably, step S6 includes:

[0061] Step S6.1: Calculate the distance between every two targets in the array group;

[0062] Step S6.2: Construct the target formation feature vector for each array group.

[0063] Preferably, step S8 includes:

[0064] Step S8.1: Obtain the total number of target types in the two array groups and assign weights according to their importance;

[0065] Step S8.2: Normalize the weight set using the following formula:

[0066]

[0067] in, N is the weight set according to importance. ij Let x be the array group at time k-1. i With the array group y at time k j The total number of target types included;

[0068] Step S8.3: Preprocess the composition feature vectors of different array groups to make the composition feature vectors of array groups have the same length;

[0069] Step S8.4: Calculate the difference in eigenvectors between array groups;

[0070] Step S8.5: Calculate the weighted inter-group composition feature difference vector;

[0071] Step S8.6: Construct the basic probability assignment function based on the feature difference vector, the formula is:

[0072]

[0073] in, Let x be the array group at time k-1. i The compiled feature vector C i With the array group y at time k j The compiled feature vector C j The degree of difference between them.

[0074] Preferably, step S9 includes:

[0075] Step S9.1: Calculate the distance matrix between the characteristic elements of the array group formation at time k-1 and time k;

[0076] Step S9.2: Extract elements from the distance matrix to form a vector;

[0077] Step S9.3: Calculate the basic probability assignment function based on the formation feature vector;

[0078] Step S9.2 includes:

[0079] Step S9.2.1: Select the smaller value of the feature dimension of the two array groups as the number of iterations;

[0080] Step S9.2.2: Obtain the maximum and minimum values ​​in the distance matrix, and their corresponding positions;

[0081] Step S9.2.3: Add the minimum value in the distance matrix to the distance feature vector;

[0082] Step S9.2.4: Delete the row and column corresponding to the minimum value of the distance matrix, and update the distance matrix;

[0083] Step S9.2.5: Determine whether the number of iterations has been reached. If not, return to step 9.2.2 for iterative calculation; otherwise, output the formation feature vector.

[0084] Preferably, step S10 includes:

[0085] Step S10.1: Calculate the error matrix of the array group targets at time k-1 and time k;

[0086] Step S10.2: Construct the basic probability assignment function based on the error value, the formula is:

[0087]

[0088] in, Let x be the array group at time k-1. i With the array group y at time k j The topological error value;

[0089] Step S10.1 includes:

[0090] Step S10.1.1: Set the initial update point set;

[0091] Step S10.1.2: Use Euclidean distance to find the nearest pair of points in the corresponding point set;

[0092] Step S10.1.3: Calculate the centroid of the source point cloud and the centroid of the target point cloud;

[0093] Step S10.1.4: Calculate the covariance matrix between centroids;

[0094] Step S10.1.5: Perform singular value decomposition on the covariance matrix;

[0095] Step S10.1.6: Calculate the rotation matrix and centroid translation vector between the two point clouds, and update the transformation matrix;

[0096] Step S10.1.7: Calculate the update point set and error value based on the update transformation matrix.

[0097] The adaptive clustering-based target data association system provided by the present invention, employing the aforementioned adaptive clustering-based target data association method, includes the following modules:

[0098] Module M1: Divide the region, obtain the grid division boundary threshold, and accumulate target location information obtained by the same load within the region;

[0099] Module M2: Clusters the payload data sequentially according to the order in which they enter the field of view to obtain the initial array group division;

[0100] Module M3: Filters a single clustering target set and removes target points whose similarity does not meet preset requirements;

[0101] Module M4: Removes non-array group targets based on array group determination conditions;

[0102] Module M5: Describes the target composition characteristics of the array group at the two consecutive time points;

[0103] Module M6: Describes the formation characteristics of the target group at the two consecutive time points;

[0104] Module M7: Models the feature vectors of the target group composition features and formation features at the two time points before and after the current time.

[0105] Module M8: Calculates the corresponding basic probability allocation function based on the target association matching model of the array group with compiled features;

[0106] Module M9: Calculates the corresponding basic probability allocation function based on the target association matching model of the formation features;

[0107] Module M10: Calculates the corresponding basic probability allocation function based on the target association matching model of the array group according to the external topological structure features;

[0108] Module M11: Using DS evidence theory, evidence is synthesized from the basic probability allocation function based on the correlation of composition, formation and external topology features, and a comprehensive basic probability allocation function is established.

[0109] Module M12: Based on two-dimensional allocation, it performs correlation decision on the basic probability allocation function of the synthesis to obtain the target matching result of the array group.

[0110] Compared with the prior art, the present invention has the following beneficial effects:

[0111] This invention divides a sparse target set into clusters and fully utilizes the relatively stable characteristics of the cluster targets' composition, formation, and external topological structure. It establishes a multi-feature mathematical model through evidence synthesis and constructs a comprehensive probability allocation function to perform target point association matching between clusters using a unified standard. This effectively reduces the impact of changes in the target motion state, sensor misses, and observation errors, and significantly improves the correct target association rate. Attached Figure Description

[0112] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0113] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0114] Figure 2 This is a schematic diagram of the process of array group partitioning based on clustering algorithm in this invention;

[0115] Figure 3 This is a schematic diagram of the process for filtering cluster target sets and removing target points with low similarity in this invention;

[0116] Figure 4 This is a flowchart illustrating the process of calculating the basic probability allocation function based on the composition features in this invention.

[0117] Figure 5 This is a flowchart illustrating the calculation of the basic probability allocation function based on formation features in this invention.

[0118] Figure 6 This is a flowchart illustrating the process of calculating the basic probability allocation function based on external topological features in this invention.

[0119] Figures 7a to 7i This is a schematic diagram of the spatial distribution of the payload acquisition targets for a set of data verified in this invention.

[0120] Figure 8 This is a schematic diagram of the spatial distribution of the target trajectory points in the verification data in Figure 7 of this invention;

[0121] Figures 9a to 9i Figure 7 is a schematic diagram of the spatial distribution of clustering results of the target data acquisition for each payload in the verification data in this invention;

[0122] Figure 10This is a schematic diagram of the trajectory association results of the verification data in Figure 7 of this invention. Detailed Implementation

[0123] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0124] Example 1

[0125] This invention provides a method for associating cluster target data based on adaptive clustering, such as... Figure 1 As shown, the method specifically includes the following:

[0126] Step S1: Perform preprocessing operations to obtain the mesh division boundary threshold and accumulate data under the same load.

[0127] In step S1, preprocessing operations are performed to obtain the mesh generation boundary threshold and accumulate data under the same load, including:

[0128] Step S1.1: Divide the region and obtain the grid division boundary threshold;

[0129] Step S1.2: Accumulate target location information acquired by the same load within this area;

[0130] Step S1.3: Process the load data in the order they enter the field of view.

[0131] Step S2: Perform clustering processing on each payload data to obtain the initial array group division.

[0132] In step S2, clustering is performed on each payload data to obtain the initial array group partitioning, such as... Figure 2 ,include:

[0133] Step S2.1: Calculate the distance between every two targets and construct a distance matrix;

[0134] Step S2.2: Find the threshold for density calculation based on the distance matrix, requiring that the number of points around each point whose distance is less than the density threshold accounts for 1%-4% of the total number of points;

[0135] Step S2.2 includes the following sub-steps:

[0136] Step S2.2.1: Parameter settings, including the upper and lower limits of the number percentage threshold and the distance range threshold;

[0137] Step S2.2.2: Set the initial density threshold as the average of the maximum distance and the minimum distance;

[0138] Step S2.2.3: Calculate the percentage of points with a density less than the threshold;

[0139] Step S2.2.4: If the proportion is within the set range, output the current density threshold; otherwise, update the current maximum distance, minimum distance, and density threshold.

[0140] Step S2.2.5: If the current distance range is less than the set threshold, output the current density threshold; otherwise, return to step S2.3 for iterative calculation.

[0141] Step S2.3: Calculate the local density of each target point and construct the local density vector;

[0142] Step S2.4: Calculate the nearest neighbor distance for each target point and construct the nearest neighbor distance vector;

[0143] Step S2.4 includes the following sub-steps:

[0144] Step S2.4.1: Sort the local density vectors in descending order of density value;

[0145] Step S2.4.2: Calculate the distance between each point and other points with a higher density than it.

[0146] Step S2.4.3: Obtain the index of the nearest neighbor of each point and the distance between the two points.

[0147] Step S2.5: Adaptively obtain cluster centers;

[0148] Step S2.5 includes the following sub-steps:

[0149] Step S2.5.1: Calculate the joint density vector based on the local density vector and the nearest neighbor distance vector;

[0150] Step S2.5.2: Calculate the mean and variance of the joint density vector;

[0151] Step S2.5.3: Traverse each data point and take the points that exceed the set threshold range as cluster centers.

[0152] Step S2.6: Classify the non-cluster center data points.

[0153] Step S2.6 includes the following sub-steps:

[0154] Step S2.6.1: Initialize the cluster number of each data point, and set the cluster number of non-cluster centers to -1;

[0155] Step S2.6.2: Sort the local density vectors in descending order of density value;

[0156] Step S2.6.3: Update the cluster number of the non-cluster center data point to be the same as the cluster number of the point that is closest to it and has a higher density.

[0157] Step S3: Filter the individual cluster target set and remove target points with low similarity.

[0158] like Figure 3 Step S3 includes the following sub-steps:

[0159] Step S3.1: Obtain the corresponding coordinates based on the obtained cluster center indices;

[0160] Step S3.2: Based on the cluster center and the target cluster number, obtain the position coordinates of each data point in the same cluster;

[0161] Step S3.3: Calculate the local density of each point in a single cluster;

[0162] Step S3.4: Calculate the similarity between each data point in the same cluster and the cluster center;

[0163] Step S3.5: Calculate the comprehensive similarity of each data point in the same cluster based on similarity and local density;

[0164] Step S3.6: Calculate the mean and variance of the combined similarity vector of data points in a single cluster;

[0165] Step S3.7: Add data points whose comprehensive similarity value exceeds the threshold range as new cluster centers and update their cluster numbers.

[0166] Step S4: Remove non-array group targets based on the array group determination criteria.

[0167] Step S4 includes the following sub-steps:

[0168] Step S4.1: Obtain the target belonging to the same cluster based on the cluster number corresponding to each data point;

[0169] Step S4.2: Assemble data for individual clusters that meet the set cluster construction threshold;

[0170] Step S4.3: Data points that do not meet the conditions for array group formation are treated as isolated points and no further array group matching is performed.

[0171] Step S5: Describe the target composition features of the array group at the two time points before and after.

[0172] Step S5 includes the following sub-steps:

[0173] Step S5.1: Calculate the target type number for each target in each array group;

[0174] Step S5.2: Calculate the number of members of each target type in each array group, which is the target composition feature of the array group.

[0175] Step S6: Describe the formation characteristics of the target group at the two time points before and after.

[0176] The description of the formation characteristics of the target group at the two consecutive time points in step S6 includes the following sub-steps:

[0177] Step S6.1: Calculate the distance between every two targets in the array group;

[0178] Step S6.2: Construct the target formation feature vector for each array group.

[0179] Step S7: Model the feature vectors of the target group composition features and formation features at the two time points before and after.

[0180] Step S8: Calculate the corresponding basic probability allocation function based on the target association matching model of the array group with compiled features.

[0181] like Figure 4 Step S8, which calculates the corresponding basic probability allocation function based on the cluster target association matching model with compiled features, includes the following sub-steps:

[0182] Step S8.1: Obtain the total number of target types in the two array groups and assign weights according to their importance;

[0183] Step S8.2: Normalize the weight set according to the following formula:

[0184]

[0185] in, N is the weight set according to importance. ij Let x be the array group at time k-1. i With the array group y at time k j The total number of target types included.

[0186] Step S8.3: Preprocess the formation feature vectors of different array groups, fill the positions corresponding to the target type in the formation feature with 0, so that the formation feature vectors of the array groups have the same length;

[0187] Step S8.4: Calculate the difference in eigenvectors between array groups;

[0188] Step S8.5: Calculate the weighted inter-group composition feature difference vector;

[0189] Step S8.6: Construct the basic probability assignment function based on the feature difference vector, according to the following formula:

[0190]

[0191] in, Let x be the array group at time k-1. i The compiled feature vector C i With the array group y at time k j The compiled feature vector C j The degree of difference between them.

[0192] Step S9: Calculate the corresponding basic probability allocation function based on the target association matching model of the formation features.

[0193] like Figure 5 Step S9 includes the following sub-steps:

[0194] Step S9.1: Calculate the distance matrix between the characteristic elements of the array group formation at time k-1 and time k;

[0195] Step S9.2: Extract elements from the distance matrix to form a vector;

[0196] Step S9.2 includes the following sub-steps:

[0197] Step S9.2.1: Select the smaller value of the feature dimension of the two array groups as the number of iterations;

[0198] Step S9.2.2: Obtain the maximum and minimum values ​​in the distance matrix, and their corresponding positions;

[0199] Step S9.2.3: Add the minimum value in the distance matrix to the distance feature vector;

[0200] Step S9.2.4: Delete the row and column corresponding to the minimum value of the distance matrix, and update the distance matrix;

[0201] Step S9.2.5: Determine whether the number of iterations has been reached. If not, return to step 9.2.2 for iterative calculation; otherwise, output the formation feature vector.

[0202] Step S9.3: Calculate the basic probability assignment function based on the formation feature vector. The calculation formula is as follows:

[0203]

[0204] in, L ijLet x be the array group at time k-1. i With the array group y at time k j The distance feature vector, N min This represents the number of iterations.

[0205] Step S10: Calculate the corresponding basic probability allocation function based on the target association matching model of the array group according to the external topological structure features.

[0206] like Figure 6 Step S10 includes the following sub-steps:

[0207] Step S10.1: Calculate the error matrix of the array group targets at time k-1 and time k;

[0208] Step S10.1 includes the following sub-steps:

[0209] Step S10.1.1: Set the initial update point set;

[0210] Step S10.1.2: Use Euclidean distance to find the nearest pair of points in the corresponding point set;

[0211] Step S10.1.3: Calculate the centroid of the source point cloud and the centroid of the target point cloud;

[0212] Step S10.1.4: Calculate the covariance matrix between centroids;

[0213] Step S10.1.5: Perform singular value decomposition on the covariance matrix;

[0214] Step S10.1.6: Calculate the rotation matrix and centroid translation vector between the two point clouds, and update the transformation matrix;

[0215] Step S10.1.7: Calculate the update point set and error value based on the update transformation matrix.

[0216] Step S10.2: Construct the basic probability assignment function based on the error value, and the calculation formula is as follows:

[0217]

[0218] in, Let x be the array group at time k-1. i With the array group y at time k j The topological error value.

[0219] Step S11: Using DS evidence theory, evidence is synthesized from the basic probability allocation function based on the correlation of composition, formation and external topology features to establish a comprehensive basic probability allocation function;

[0220] Step S12: Based on the two-dimensional allocation, perform association decision on the comprehensive basic probability allocation function to obtain the target matching result of the array group, according to the following formula:

[0221]

[0222] The constraints are satisfied:

[0223] m ij >m_ref

[0224]

[0225] The optimal matching solution is

[0226] Where, m ij Represents the matching metric matrix The (i, j)th element; f ij Representing the allocation matrix The (i, j)th element, which is a binary variable; N k-1 N represents the number of group targets at time (k-1); k Indicates the number of group targets at time (k); m_ref is the set confidence threshold; (z 1i , z 2j () indicates a successfully matched target pair.

[0227] Step S13: The target information of the array group will be processed in the next time step.

[0228] The effects of this invention can be further illustrated by the following simulation results.

[0229] 1. Simulation data:

[0230] The sequence of maritime target location information obtained from multi-load observations within a specified area using STK simulation was processed using the PyCharm 2017 platform.

[0231] 2. Simulation content and results:

[0232] To verify the effectiveness of the adaptive clustering-based target data association method provided in this invention, experiments were conducted on target datasets with unknown array group information. The following section describes the experimental data and processing results of a sequence of position information of 30 moving maritime targets observed by 9 payloads over a period of 23 minutes.

[0233] Figure 7 is a schematic diagram of the spatial distribution of 30 moving targets at sea observed by 9 payloads over a simulated period of 23 minutes.

[0234] Figure 8 Figure 7 shows a schematic diagram of the spatial distribution of target trajectory points in the scenario described.

[0235] Figure 9 is a schematic diagram of the spatial distribution of clustering results of the targets acquired by each payload in the scenario described in Figure 7;

[0236] Figure 10 This is a schematic diagram of the trajectory association results in the scenario described in Figure 7.

[0237] Example 2

[0238] This invention also provides a cluster target data association system based on adaptive clustering. The cluster target data association based on adaptive clustering can be implemented by executing the process steps of the cluster target data association method based on adaptive clustering. That is, those skilled in the art can understand the cluster target data association method based on adaptive clustering as a preferred embodiment of the cluster target data association system based on adaptive clustering. The system specifically includes the following:

[0239] Module M1: Performs preprocessing operations to obtain the mesh division boundary threshold and accumulates data under the same load.

[0240] Module M1 includes:

[0241] Module M1.1: Divide the region and obtain the grid division boundary threshold;

[0242] Module M1.2: Accumulates target location information acquired by the same load within this area;

[0243] Module M1.3: Processes each load data sequentially according to the order in which they enter the field of view.

[0244] Module M2: Performs clustering processing on each payload data to obtain the initial array group division.

[0245] Module M2 includes:

[0246] Module M2.1: Calculates the distance between any two targets and constructs a distance matrix;

[0247] Module M2.2: Find the threshold for density calculation based on the distance matrix, requiring that the number of points around each point whose distance is less than the density threshold accounts for 1%-4% of the total number of points;

[0248] Module M2.2 includes:

[0249] Module M2.2.1: Parameter settings, including upper and lower limits for the percentage threshold and distance range threshold;

[0250] Module M2.2.2: Sets the initial density threshold, which is the average of the maximum distance and the minimum distance;

[0251] Module M2.2.3: Calculates the percentage of points with a density less than a threshold;

[0252] Module M2.2.4: If the proportion is within the set range, output the current density threshold; otherwise, update the current maximum distance, minimum distance, and density threshold.

[0253] Module M2.2.5: If the current distance range is less than the set threshold, output the current density threshold; otherwise, return to module M2.3 for iterative calculation.

[0254] Module M2.3: Calculates the local density of each target point and constructs the local density vector;

[0255] Module M2.4: Calculates the nearest neighbor distance for each target point and constructs the nearest neighbor distance vector;

[0256] Module M2.4 includes:

[0257] Module M2.4.1: Sort local density vectors in descending order of density value;

[0258] Module M2.4.2: Calculates the distance between each point and other points with a higher density than it.

[0259] Module M2.4.3: Get the index of the nearest neighbor of each point and the distance between the two points.

[0260] Module M2.5: Adaptive acquisition of cluster centers;

[0261] Module M2.5 includes:

[0262] Module M2.5.1: Calculates the joint density vector based on the local density vector and the nearest neighbor distance vector;

[0263] Module M2.5.2: Calculate the mean and variance of the joint density vector;

[0264] Module M2.5.3: Iterates through each data point and uses points that exceed the set threshold range as cluster centers.

[0265] Module M2.6: Classifies non-cluster center data points.

[0266] Module M2.6 includes:

[0267] Module M2.6.1: Initialize the cluster number for each data point, and set the cluster number for non-cluster centers to -1;

[0268] Module M2.6.2: Sort local density vectors in descending order of density value;

[0269] Module M2.6.3: Updates the cluster number of non-cluster center data points to be the same as the cluster number of the nearest point with a higher density.

[0270] Module M3: Filters a single cluster target set and removes target points with low similarity.

[0271] Module M3 includes:

[0272] Module M3.1: Obtain the corresponding coordinates based on the acquired cluster center indices;

[0273] Module M3.2: Based on the cluster center and the target cluster number, obtain the position coordinates of each data point in the same cluster;

[0274] Module M3.3: Calculates the local density of each point in a single cluster;

[0275] Module M3.4: Calculates the similarity between each data point and the cluster center within the same cluster;

[0276] Module M3.5: Calculates the overall similarity of each data point in the same cluster based on similarity and local density;

[0277] Module M3.6: Calculates the mean and variance of the combined similarity vector of data points in a single cluster;

[0278] Module M3.7: Adds data points whose comprehensive similarity value exceeds the threshold range as new cluster centers and updates their cluster numbers.

[0279] Module M4: Removes non-array group targets based on array group determination conditions.

[0280] Module M4 includes:

[0281] Module M4.1: Obtain the target belonging to the same cluster based on the cluster number corresponding to each data point;

[0282] Module M4.2: Assembles data from individual clusters that meet the set cluster construction threshold;

[0283] Module M4.3: Data points that do not meet the conditions for array group formation are treated as isolated points and are not subject to subsequent array group matching.

[0284] Module M5: Describes the target composition characteristics of the array group at the two time points before and after.

[0285] Module M5 includes:

[0286] Module M5.1: Calculates the target type number for each target in each array group;

[0287] Module M5.2: Calculates the number of members of each target type in each array group, which is the target composition feature of the array group.

[0288] Module M6: Describes the formation characteristics of the target group at the two time points before and after.

[0289] Module M6 describes the formation characteristics of the target group at two different time points, including:

[0290] Module M6.1: Calculates the distance between any two targets in the array group;

[0291] Module M6.2: Constructs the target formation feature vector for each array group.

[0292] Module M7: Models the feature vectors of the target group composition features and formation features at the two time points before and after the current time.

[0293] Module M8: Calculates the corresponding basic probability allocation function based on the target association matching model of the array group with the compiled features.

[0294] In module M8, the basic probability allocation function is calculated based on the target association matching model of the array group with compiled features, including:

[0295] Module M8.1: Obtains the total number of target types in two array groups and assigns weights according to their importance;

[0296] Module M8.2: Normalizes the weight set according to the following formula:

[0297]

[0298] in, N is the weight set according to importance. ij Let x be the array group at time k-1. i With the array group y at time k j The total number of target types included.

[0299] Module M8.3: Preprocesses the formation feature vectors of different array groups, filling the positions corresponding to the non-existent target types in the formation features with 0s, so that the formation feature vectors of the array groups have consistent lengths;

[0300] Module M8.4: Calculates the difference in eigenvectors between array groups;

[0301] Module M8.5: Calculates the weighted inter-group compositional feature difference vector;

[0302] Module M8.6: Constructs a basic probability allocation function based on feature difference vectors, according to the following formula:

[0303]

[0304] in, Let x be the array group at time k-1. i The compiled feature vector C i With the array group y at time k j The compiled feature vector C j The degree of difference between them.

[0305] Module M9: Calculates the corresponding basic probability allocation function based on the target association matching model of the formation characteristics.

[0306] Module M9 includes:

[0307] Module M9.1: Calculate the distance matrix between the characteristic elements of the array group formation at time k-1 and time k;

[0308] Module M9.2: Extracts elements from the distance matrix to form a vector;

[0309] Module M9.2 includes:

[0310] Module M9.2.1: Select the smaller value of the feature dimension of the two array groups as the number of iterations;

[0311] Module M9.2.2: Obtain the maximum and minimum values ​​in the distance matrix, and their corresponding positions;

[0312] Module M9.2.3: Adds the minimum value in the distance matrix to the distance feature vector;

[0313] Module M9.2.4: Delete the row and column corresponding to the minimum value in the distance matrix and update the distance matrix;

[0314] Module M9.2.5: Determines if the required number of iterations has been reached. If not, returns to module M9.2.2 to continue the iteration count.

[0315] Calculate, otherwise output the formation feature vector.

[0316] Module M9.3: Calculates the basic probability assignment function based on the formation feature vector. The calculation formula is as follows:

[0317]

[0318] in, L ij Let x be the array group at time k-1. i With the array group y at time k j The distance feature vector, N min This represents the number of iterations.

[0319] Module M10: Calculates the corresponding basic probability allocation function based on the target association matching model of the array group according to the external topological structure features.

[0320] Module M10 includes:

[0321] Module M10.1: Calculates the error matrix of the target group at time k-1 and time k;

[0322] Module M10.1 includes:

[0323] Module M10.1.1: Sets the initial update point set;

[0324] Module M10.1.2: Find the nearest pair of points in a set of points using Euclidean distance;

[0325] Module M10.1.3: Calculates the centroid of the source point cloud and the centroid of the target point cloud;

[0326] Module M10.1.4: Calculate the covariance matrix between centroids;

[0327] Module M10.1.5: Performs singular value decomposition on the covariance matrix;

[0328] Module M10.1.6: Calculates the rotation matrix and centroid translation vector between two point clouds, and updates the transformation matrix;

[0329] Module M10.1.7: Calculates the updated point set and error value based on the update transformation matrix.

[0330] Module M10.2: Constructs a basic probability assignment function based on error values, calculated as follows:

[0331]

[0332] in, Let x be the array group at time k-1. i With the array group y at time k j The topological error value.

[0333] Module M11: Using DS evidence theory, evidence is synthesized from the basic probability allocation function based on the correlation of composition, formation and external topology features, and a comprehensive basic probability allocation function is established.

[0334] Module M12: Based on the two-dimensional allocation, it performs association decision-making on the comprehensive basic probability allocation function to obtain the target matching result of the array group, according to the following formula:

[0335]

[0336] The constraints are satisfied:

[0337] mij >m_ref

[0338]

[0339]

[0340] The optimal matching solution is

[0341] Where, m ij Represents the matching metric matrix The (i, j)th element; f ij Representing the allocation matrix The (i, j)th element, which is a binary variable; N k-1 N represents the number of group targets at time (k-1); k Indicates the number of group targets at time (k); m_ref is the set confidence threshold; (z 1i , z 2j () indicates a successfully matched target pair.

[0342] Module M13: Processing the target information of the array group in the next time step.

[0343] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0344] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for adaptive clustering based group target data association, comprising: The method comprises the following steps: Step S1: dividing a region, obtaining a grid division boundary threshold value, and accumulating target position information obtained by the same load within the region; Step S2: sequentially clustering each load data according to the order of entering the field of view to obtain initial array group division; Step S3: screening a single cluster target set and removing target points that do not meet preset requirements in similarity; Step S4: removing non-array group targets according to array group judgment conditions; Step S5: respectively describing formation features of array group targets at two time points; Step S6: respectively describing formation features of array group targets at two time points; Step S7: respectively modeling feature vectors of formation features and formation features of array group targets at two time points; Step S8: calculating corresponding basic probability assignment functions according to an array group target correlation matching model based on formation features; Step S9: calculating corresponding basic probability assignment functions according to an array group target correlation matching model based on formation features; Step S10: calculating corresponding basic probability assignment functions according to an array group target correlation matching model based on external topological structure features; Step S11: synthesizing evidences of basic probability assignment functions based on correlation of formation features, formation features and external topological structure features by D-S evidence theory to establish a comprehensive basic probability assignment function; Step S12: making a correlation decision on the comprehensive basic probability assignment function based on two-dimensional distribution to obtain an array group target matching result. The step S3 comprises: Step S3.1: obtaining corresponding coordinates according to a cluster center index; Step S3.2: obtaining position coordinates of each data point in the same cluster according to the cluster center and each target cluster number; Step S3.3: calculating local density of each point in a single cluster; Step S3.4: calculating similarity of each data point in the same cluster to the cluster center; Step S3.5: calculating comprehensive similarity of each data point in the same cluster based on similarity and local density; Step S3.6: calculating mean value and variance of a comprehensive similarity vector of data points in a single cluster; Step S3.7: adding a data point with a comprehensive similarity value beyond a threshold range as a new cluster center and updating the cluster number thereof.

2. The adaptive clustering based group target data association method of claim 1, wherein, The step S2 comprises: Step S2.1: calculating distance between each two targets to construct a distance matrix; Step S2.2: finding a density calculation threshold value according to the distance matrix, wherein the number of points with a distance less than the density threshold value around each point accounts for 1%-4% of the total number of points; Step S2.3: calculating local density of each target point and constructing a local density vector; Step S2.4: calculating nearest neighbor distance of each target point and constructing a nearest neighbor distance vector; Step S2.5: adaptively obtaining a cluster center; Step S2.6: classifying non-cluster center data points; The step S2.2 comprises: Step S2.2.1: parameter setting, including upper and lower limits of the number ratio threshold value and distance range threshold value; Step S2.2.2: setting an initial density threshold value as an average of the maximum distance and the minimum distance; Step S2.2.3: calculating the ratio of the number of points with a density less than the threshold value; Step S2.2.4: If the proportion is within the set range, output the current density threshold; otherwise, update the current maximum distance, minimum distance and density threshold; Step S2.2.5: If the current distance range is less than the set threshold, output the current density threshold; otherwise, return to step S2.3 for iterative calculation; The step S2.4 includes: Step S2.4.1: Sort the local density vector by density value from large to small; Step S2.4.2: Calculate the distance between each point and other points with greater density than the point; Step S2.4.3: Get the index corresponding to the nearest neighbor point of each point and the distance between the two points; The step S2.5 includes: Step S2.5.1: Calculate the joint density vector according to the local density vector and the nearest neighbor point distance vector; Step S2.5.2: Calculate the mean and variance of the joint density vector; Step S2.5.3: Traverse each data point, and take the points exceeding the set threshold range as the cluster center; The step S2.6 includes: Step S2.6.1: Initialize the cluster number of each data point, and set the cluster number of non-cluster center to -1 uniformly; Step S2.6.2: Sort the local density vector by density value from large to small; Step S2.6.3: Update the cluster number of non-cluster center data point to the same as the cluster number of the point with the closest distance and greater density.

3. The adaptive clustering based group target data association method of claim 1, wherein, The step S4 includes: Step S4.1: Obtain the same cluster belonging target according to the cluster number corresponding to each data point; Step S4.2: Assemble data for a single cluster that meets the set array group construction threshold; Step S4.3: Take the data points that do not meet the array group formation condition as isolated points, and do not perform subsequent array group matching.

4. The adaptive clustering based group target data association method of claim 1, wherein, The step S5 includes: Step S5.1: Calculate the number of each target belonging to the target type in each array group; Step S5.2: Calculate the number of members of each target type contained in each array group.

5. The adaptive clustering based group target data association method of claim 1, wherein, The step S6 includes: Step S6.1: Calculate the distance between each two targets in the array group; Step S6.2: Construct the target formation feature vector of each array group.

6. The adaptive clustering based group target data association method of claim 1, wherein, The step S8 includes: Step S8.1: Obtain the total number of target types of the two array groups, and assign weights according to the importance; Step S8.2: Normalize the weight set, formula: wherein, is a weight set according to the importance, is an array group at k-1 time is an array group at k time total number of target types contained Step S8.3: Preprocess the formation feature vector of different array groups to make the length of the array group formation feature vector consistent; Step S8.4: Calculate the formation feature vector difference between array groups; Step S8.5: Calculate the formation feature difference degree vector between array groups based on the weight; Step S8.6: Construct the basic probability assignment function based on the feature difference degree vector, formula: wherein, the array group at k-1 time the composed eigenvector the array group at k time the composed eigenvector the difference degree between 7. The adaptive clustering based group target data association method of claim 1, wherein, The step S9 includes: Step S9.1: Calculate the distance matrix between the formation feature elements of the k-1 time and the k time array group; Step S9.2: Extract elements from the distance matrix to form a vector; Step S9.3: Calculate the basic probability assignment function based on the formation feature vector; The step S9.2 includes: Step S9.2.1: Select the smaller value of the two array group formation feature dimensions as the iteration number; Step S9.2.2: Obtain the maximum and minimum values in the distance matrix and the corresponding positions; Step S9.2.3: Add the minimum value in the distance matrix into the distance feature vector; Step S9.2.4: Delete the row and column corresponding to the minimum value in the distance matrix and update the distance matrix; Step S9.2.5: Determine whether the iteration number is reached, if not, return to step 9.2.2 for iteration calculation, otherwise output the formation feature vector.

8. The adaptive clustering based group target data association method of claim 1, wherein, The step S10 comprises: Step S10.1: Calculate the error matrix of the array group target at k-1 time and k time; Step S10.2: Construct the basic probability assignment function based on the error value, the formula is: wherein, topology error value of the array group at k-1 time topology error value of the array group at k time topology error value of the array group at k-1 time The step S10.1 comprises: Step S10.1.1: Set the initial update point set; Step S10.1.2: Find the nearest point pair of the corresponding point set by using the Euclidean distance; Step S10.1.3: Calculate the centroid of the source point cloud and the centroid of the target point cloud; Step S10.1.4: Calculate the covariance matrix between the centroids; Step S10.1.5: Perform singular value decomposition on the covariance matrix; Step S10.1.6: Calculate the rotation matrix and centroid translation vector between the two point clouds, and update the transformation matrix; Step S10.1.7: Calculate the update point set according to the update transformation matrix and calculate the error value.

9. An adaptive clustering based array group target data association system, characterized by, The adaptive clustering-based array group target data association method according to any one of claims 1 to 8 comprises the following modules: Module M1: Divide the region, obtain the grid division boundary threshold value, and accumulate the target position information obtained by the same load within the region; Module M2: Process the clustering of each load data in turn according to the order of entering the field of view, and obtain the initial array group division; Module M3: Screen the single clustering target set and remove the target points that do not meet the preset requirement of similarity; Module M4: Remove the non-array group target according to the array group determination condition; Module M5: Describe the formation characteristics of the array group targets at the front and rear two time points respectively; Module M6: Describe the formation characteristics of the array group targets at the front and rear two time points respectively; Module M7: Model the feature vectors of the formation characteristics and the formation characteristics of the array group targets at the front and rear two time points respectively; Module M8: Calculate the corresponding basic probability assignment function according to the array group target association matching model based on the formation characteristics; Module M9: Calculate the corresponding basic probability assignment function according to the array group target association matching model based on the formation characteristics; Module M10: Calculate the corresponding basic probability assignment function according to the array group target association matching model based on the external topological structure characteristics; Module M11: Perform evidence synthesis on the basic probability assignment functions based on the formation, formation and external topological structure characteristics by using the D-S evidence theory, and establish a comprehensive basic probability assignment function; Module M12: Make association decision based on the two-dimensional distribution of the comprehensive basic probability assignment function, and obtain the array group target matching result.

Images