A method and system for fusion of multi-radar target detection data

Abstract

The present invention belongs to the field of radar data processing technology, and specifically relates to a method and system based on multi-radar target detection data fusion, wherein the method comprises respectively acquiring N initial trajectories of at least one target through N radars, and clustering the initial trajectories by introducing point track density and density sensitive distance according to the initial trajectories, so as to generate a unique and reliable target motion trajectory; the present invention can accurately evaluate the similarity and difference between the target point track and the track under the condition of small calculation amount, effectively avoid the problem of inaccurate fusion of scattered point tracks, thereby improving the target tracking accuracy and robustness.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar data processing, and in particular relates to a method and system for fusion of multi-radar target detection data. Background Art

[0002] Radar is a core sensor that uses electromagnetic waves to detect, sense and acquire information about targets. It is widely used in electronic countermeasures, low-altitude penetration and other fields. However, with the complex changes in the application environment, a single radar has low target tracking accuracy, a narrow effective detection range and a long system response time, which cannot meet the needs. Therefore, parallel multi-radar came into being.

[0003] During the multi-radar data collection process, point data fusion technology can be used to acquire and process information on demand to reduce information loss and obtain more accurate track status. Among them, data fusion technology is a process of association, estimation and fusion by obtaining accurate estimated status and attribute judgment.

[0004] For example, the Chinese invention patent with publication number CN106680806A discloses a multi-radar point track fusion method. This method provides a high-performance point track fusion processing algorithm for the input points of different radars based on preprocessing such as spatiotemporal processing and comprehensive correlation, and further provides reasonable track association and filtering algorithms to obtain the estimated result of the target track.

[0005] For example, the Chinese invention patent with publication number CN117805805A discloses a millimeter-wave radar multi-vehicle tracking method based on traffic scenarios. This method combines the characteristics of radar measurement distribution and motion patterns of vehicle targets to achieve robust tracking of multiple vehicle targets.

[0006] However, how to improve the accuracy of scattered point fusion and thus improve target tracking accuracy and robustness is still a problem that needs to be solved urgently. Summary of the invention

[0007] The purpose of the present invention is to provide a method and system for fusion of multi-radar target detection data to partially solve or alleviate the above-mentioned problems and improve the target tracking accuracy and robustness without affecting the radar data fusion efficiency.

[0008] In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions:

[0009] A method for fusion of multi-radar target detection data comprises the following steps:

[0010] S101, obtaining N initial tracks of at least one target through N radars respectively;

[0011] S102, defining the N initial trajectories as a plurality of point trace sets, and calculating the number of cluster ranges according to the plurality of point trace sets; wherein the point trace set includes: a point trace corresponding to the target detected by the radar at a plurality of time instants, and the point trace is the spatial coordinate of the target;

[0012] S103, selecting a plurality of cluster numbers from the cluster range number, wherein the cluster numbers are positive integers;

[0013] S104, clustering the point trace set according to the plurality of cluster numbers respectively, so as to obtain a plurality of cluster centroid lists accordingly; wherein S104 includes:

[0014] S41, calculating the average density parameter of the plurality of point trace sets under the current number of clusters;

[0015] S42, calculating the regional node density of each point trace in the point trace set according to the average density parameter, and selecting the point traces that meet the first rule to form a high-density point trace set; wherein the first rule requires that the regional node density of the point trace is greater than or equal to the average density parameter;

[0016] S43, calculating at least two initial centroids according to the high-density point trace set, and forming an initial centroid set; wherein S43 includes:

[0017] Selecting the centroid with the largest node density in the region in the high-density point trace set as the first initial centroid, and selecting at least one initial centroid in the high-density point trace set in sequence according to the first initial centroid using the second rule, wherein the total number of the initial centroids is the same as the number of clusters;

[0018] Wherein, the second rule requires that the selection of the at least one initial centroid needs to meet the maximum distance condition;

[0019] S44, calculating the density-sensitive distance between each of the point traces in the point trace set and each of the initial centroids, and selecting the initial centroid with the shortest density-sensitive distance as the attribution result of the current point trace, so as to form a plurality of cluster sets;

[0020] S45, calculating the cluster centroid according to the cluster set using the following formula; wherein,

[0021] ;

[0022] in, For the cluster centroids, The range is 1- , For the The cluster set formed by the initial centroids The number of midpoint traces, The cluster set Midpoint The three-dimensional coordinates of

[0023] S46, judging whether the cluster centroid satisfies the iteration condition If satisfied, the cluster centroid is stored in the cluster centroid list; if not satisfied, the cluster centroid is used as the initial centroid and S44-S46 are repeatedly executed;

[0024] S105, respectively calculating the mean square errors of the plurality of cluster centroid lists, and selecting the cluster centroids in a group of cluster centroid lists with the smallest mean square error as the fused trajectory.

[0025] As an improvement, S101 includes:

[0026] S11, set the initial moment target Status , and through the state Obtaining an initial state covariance matrix of the target;

[0027] S12, using the initial state covariance matrix, using the Kalman filter algorithm to predict the target The first state covariance matrix at time ;

[0028] S13, The first state covariance matrix predicted at the moment is The difference of the second state covariance matrix actually measured at the moment is multiplied by the target measurement noise covariance matrix to obtain the similarity of the target , and set the tracking threshold , if satisfied ,but , that is, The measurement falls into the tracking gate of the target. The value of forms a confirmation matrix; if it does not meet ,but , that is, A measurement does not fall into the tracking gate of the target, and the data does not belong to the current target;

[0029] in, Indicates a binary variable measuring the number of tracking gates that fall into the target;

[0030] S14, definition a joint correlation event, and according to the joint correlation event, the confirmation matrix is ​​split by using the third rule and the fourth rule to obtain a joint event correlation matrix;

[0031] The third rule means that each target can only produce one valid measurement result, and the fourth rule means that each measurement result can only come from a certain target or clutter;

[0032] S15, calculating the joint associated events and the joint event association matrix to obtain the conditional probability of the associated events;

[0033] S16, calculated from the conditional probability of the associated event The moment The probability of association of a measurement with the target;

[0034] S17, associating the measurement with the largest association probability value with the target to form a local tracking trajectory of the target, and adding the measurement to a trajectory list of the target;

[0035] S19, repeatedly executing S12-S17 to form a local tracking trajectory at the next moment, until a complete initial trajectory is formed.

[0036] As an improvement, in S41, the calculation formula of the average density parameter is:

[0037] ;

[0038] in, is a set of points, is the average density parameter, is the number of clusters.

[0039] As an improvement, in S42, the calculation formula of the regional node density function is:

[0040] ;

[0041] in, For the The regional node density function of the points is The range is 1- , Respectively represent the points and dot trace location.

[0042] As an improvement, before S19, the following steps are also included:

[0043] S18, generating a new tracking trajectory for the unassociated measurement and updating the current trajectory list state.

[0044] As an improvement, the first state covariance matrix in S12 is:

[0045] ;

[0046] in, For the goal, is the first state covariance matrix, is the state transfer matrix, for Time target The state covariance matrix of , T is the transpose of the matrix, for The goal of the moment The process noise matrix.

[0047] As an improvement, in S15, the calculation formula of the conditional probability of the associated event is:

[0048] ;

[0049] in: is the conditional probability of the associated event, is the normalization constant; is the impurity density; is the number of measurements that are not associated with the target in the joint association event; is the prior Gaussian probability distribution of the measurement; for Moment Measurements, Joint related events Measurements and Targets Interconnection instructions; It is a target detection indication for a joint correlation event; and The values ​​are all 1 or 0; For the goal The detection probability of is the total number of measurements; is the total number of tracked targets.

[0050] As an improvement, in S16, The moment Measurements and Targets The probability of association The calculation is done using the following formula:

[0051] ;

[0052] in, is the joint event correlation matrix.

[0053] The present invention also provides a multi-radar target detection data fusion system, comprising:

[0054] A data acquisition module: configured to acquire N initial trajectories of at least one target through N radars respectively;

[0055] A clustering module: configured to define the N initial trajectories as point trace sets respectively, and calculate the number of cluster ranges; select multiple cluster numbers from the cluster ranges, and the cluster numbers are positive integers; and cluster the point trace sets according to the multiple cluster numbers respectively, so as to obtain multiple cluster centroid lists accordingly;

[0056] The trajectory fusion module is configured to respectively calculate the mean square errors of the plurality of cluster centroid lists, and select a group of cluster centroids in the cluster centroid list with the smallest mean square error as the fused trajectory.

[0057] As an improvement, when obtaining the cluster centroid list, the clustering module is configured to calculate the average density parameter of the point trace set under the current number of clusters;

[0058] Calculate the regional node density of each point trace in the N point trace sets according to the average density parameter, and select the point traces that meet the first rule to form a high-density point trace set;

[0059] At least two initial centroids are calculated based on the high-density point trace set to form an initial centroid set; the density-sensitive distance between each of the point traces in the point trace set and each of the initial centroids is calculated; and the initial centroid with the shortest density-sensitive distance is selected as the attribution result of the point trace to form a cluster set;

[0060] The cluster centroid is calculated according to the cluster set; and it is determined whether the cluster centroid meets the iteration condition. If so, the cluster centroid is stored in the cluster centroid list, otherwise the cluster centroid is recalculated.

[0061] The principle and beneficial technical effects of the present invention are:

[0062] The present invention provides a method for quickly performing traversal clustering on multiple cluster numbers based on a high-density point trace set, and simultaneously uses density-sensitive distance to locally calculate the point traces and key data (initial centroids obtained based on the high-density point trace set), thereby reducing the amount of calculation while effectively ensuring the accuracy of the fusion.

[0063] For the fusion of tracks with scattered points, for example, when dealing with the problem of drone track fusion, the flight path of the drone often covers a wide area, which will cause its trajectory points to be sparsely distributed. This dispersion makes the integration of track data more complicated.

[0064] Firstly, the present invention forms a high-density point trace set in the evacuated point traces, and then determines the initial value (i.e., selects the initial centroid) in the high-density point trace set with a smaller range, so as to obtain the initial centroid with higher accuracy; and by selecting multiple cluster numbers, respectively calculates and selects the optimal cluster centroid list, so that even if multiple cluster numbers are used for calculation, the amount of calculation will not increase. On the contrary, through the above method, the efficiency and accuracy of selecting the initial centroid can be improved to a certain extent.

[0065] Furthermore, in the specific clustering process, this scheme introduces density-sensitive distance to associate the points with the initial centroid. That is to say, the points are calculated and classified with the finite data (initial centroid) to obtain the clustering set, which can further reduce the amount of calculation and improve the efficiency and accuracy of data fusion.

[0066] In summary, the present invention forms a high-density point track set, selects multiple cluster numbers for calculation respectively, and introduces multiple means such as density sensitive distance to coordinate, which can accurately evaluate the similarities and differences between target point tracks and tracks with a small amount of calculation, effectively avoid the problem of inaccurate fusion of scattered point tracks, thereby improving target tracking accuracy and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, each element or part is not necessarily drawn according to the actual scale. Obviously, the drawings described below are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without paying creative labor.

[0068] Figure 1 This is a flow chart of a method for fusion of multi-radar target detection data in an embodiment of the present invention;

[0069] Figure 2 It is a schematic diagram of the modular structure of a multi-radar target detection data fusion system in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0071] Herein, the suffixes such as "module", "component" or "unit" used to represent elements are used only to facilitate the description of the present invention and have no specific meaning in themselves. Therefore, "module", "component" or "unit" can be used in a mixed manner. Herein, the orientation or positional relationship indicated by the terms "upper", "lower", "inner", "outer", "front", "back", "one end", "the other end", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0072] In this document, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In this document, "plurality" means two or more, that is, it includes two, three, four, five, etc.

[0073] Embodiment 1

[0074] See also Figure 1 The present invention mainly aims at the problem of inaccurate data fusion when using multiple radars to simultaneously measure the trajectory of one or more targets and fuse the trajectories of the same target obtained by multiple radars into one trajectory, and provides a method for fusion of multi-radar target detection data, including the following steps:

[0075] S101, obtaining N initial tracks of at least one target through N radars respectively;

[0076] In some embodiments, S101 specifically includes:

[0077] S11, Settings The goal of the moment Status , and through the state Get Target The initial state covariance matrix ;

[0078] In some specific embodiments, a single-point initialization method can be used to obtain the target The initial state covariance matrix ;

[0079] Among them, the state Also known as a state vector, or an initial / starting state vector; the single-point initialization method is also known as a monocular initialization method, which aims to obtain the initial relative pose between frames, fix the scale, and restore the initial map points through triangulation for subsequent tracking and map construction.

[0080] S12, according to the initial state covariance matrix , predict the target through the Kalman filter algorithm exist The first state covariance matrix at time , filter gain matrix and the innovation covariance matrix ;in,

[0081] ;

[0082] ;

[0083] ;

[0084] in: is the state transfer matrix, For the goal exist The state covariance matrix at time , is the process noise matrix, For the goal exist The process noise matrix at time , is the measurement noise matrix, For the goal The measurement noise matrix is is the observation matrix, For the goal The observation matrix, T is the transpose of the matrix, as the goal.

[0085] In this paper, Kalman filtering is an algorithm that uses the linear system state equation to optimally estimate the system state through system input and output observation data.

[0086] S13, The first state covariance matrix predicted at the moment is The difference of the second state covariance matrix actually measured at the moment is multiplied by the measured target measurement noise covariance matrix to obtain the similarity of the current target , and set the tracking threshold , if satisfied ,but , that is, A measurement (also called a measurement ) falls into the target In the tracking gate, The values ​​of form confirmation matrix If not satisfied ,but , that is, Measurements not on target In the tracking gate, the data does not belong to the current target;

[0087] in, Indicates measurement Falling into the target Binary variable for the tracking gate.

[0088] Specifically, if the tracking gates of multiple targets being tracked do not intersect or no echo is located in the intersecting area, the multi-target tracking problem can be simplified to PDA for processing. When there are tracking gates belonging to multiple tracks intersecting and the echo is located in the intersection of the tracking gates, Bar-Shalom introduces a confirmation matrix to obtain the full arrangement of valid echoes and clutter and all joint events.

[0089] For example, an example of a confirmation matrix is ​​given below:

[0090]

[0091] In this confirmation matrix, =l means the ( =1,2,…, ) measurements fall within the target ( =0,1,…, ) within the tracking gate; =0 means Measurements not on target Inside the tracking door. Indicates noise and falsehood, The first column element is always 1, which means that all measurement results may come from false alarms or clutter.

[0092] S14, definition The joint associated events are , using the third and fourth rules to transform the confirmation matrix Split to get the joint event correlation matrix ; Among them, the third rule means that each target can only produce one valid measurement result, and the fourth rule means that each measurement result can only come from a certain target or clutter.

[0093] S15, the joint association event and the joint event correlation matrix Calculate and get the conditional probability of the associated event ;in, For the Time to The current target is measured at all times, and the accumulated measurement results are a collection of multiple measurements;

[0094] Among them, the conditional probability of the associated event The calculation formula is:

[0095] ;

[0096] in: is the normalization constant; is the impurity density; is the number of measurements that are not associated with the target in the joint association event; is the prior Gaussian probability distribution of the measurement; for The moment Measurements; For joint correlation events With the goal Interconnection instructions; It is a target detection indication for a joint correlation event; and The values ​​are all 1 or 0; For the goal The detection probability of is the total number of measurements; is the total number of tracked targets.

[0097] S16, based on the conditional probability of the associated event Calculated Time target The state equation , error covariance matrix and Measurements and Targets The probability of association ;in,

[0098] ;

[0099] ;

[0100] ;

[0101] in, for Time target One-step prediction of the state, for Time target The one-step forecast error covariance matrix of for Time target The probability that all valid measurements of come from false measurements, For Moment Valid measurements come from the target The estimated value of the state, For the Measurement target The combined new information of measurement prediction.

[0102] S17, select The maximum value target is associated with the current measurement to form a local tracking trajectory of the current target, and the measurement is attributed to the target List of trajectories.

[0103] S19, repeatedly executing S12-S17 to form a local tracking trajectory at the next moment, until a complete initial trajectory is formed.

[0104] In some embodiments, before S19, the step of:

[0105] S18, generating a new tracking track for the measurement that is not associated, and updating the current track list state;

[0106] Accordingly, S19 includes repeatedly executing S12-S18 to form a local tracking trajectory at the next moment, until a complete initial trajectory is formed.

[0107] S102, defining the N initial trajectories as point track sets respectively , and calculate the number of cluster ranges ; Wherein, the point track set includes: a point track corresponding to the target detected by the radar at multiple time moments, and the point track is the spatial coordinate of the target.

[0108] S103, selecting a plurality of cluster numbers from the cluster range number , the number of clusters Take a positive integer.

[0109] S104, respectively, according to the plurality of clustering numbers Clustering the point trace set to obtain a plurality of cluster centroid lists; wherein S104 includes:

[0110] S41, calculate the current number of clusters The average density parameter of the multiple point sets described below ; Wherein, the average density parameter The calculation formula is:

[0111] .

[0112] S42, according to the average density parameter Calculate the regional node density of each point in the N point sets respectively , select the points that meet the first rule to form a high-density point set ; Wherein, the first rule requires the regional node density of the point trace Greater than or equal to the average density parameter ;

[0113] Among them, The regional node density function of the points Defined as:

[0114] ;

[0115] Respectively represent the points and dot trace location, is the sequence number of the point trace, in this formula The range is 1- .

[0116] S43, based on the currently acquired high-density point trace set A plurality of initial centroids are calculated and an initial centroid set is formed; wherein S43 includes:

[0117] Select the regional node density in the current point set (i.e. the high-density point set) The largest centroid is used as the first initial centroid , use the second rule to select in the current point set An initial centroid;

[0118] Among them, the second rule requires Initial centroid The selection needs to meet the maximum distance condition:

[0119] ;

[0120] in, For the The initial centroid, For the The initial centroid, represents the serial number of the initial centroid. The range is 1- ;

[0121] That is to say, select the point with the largest distance from the previous initial centroid as the next initial centroid. For example, select the point with the largest distance from the current point set. The farthest point is used as the second initial centroid , select the current point concentration distance The farthest point is used as the third initial centroid . . . , and so on, select the distance in the current point trace concentration The farthest point is the Initial centroid .

[0122] S44, calculating the point trace set The density-sensitive distance between each of the points and each of the initial centroids in ; and select the initial centroid with the shortest density-sensitive distance as the attribution result of the current point trace to form multiple clustering sets ;

[0123] Specifically, in S44, the point trace sets are calculated respectively The density-sensitive distance between a single point trace and each initial centroid is obtained, and the initial centroid with the shortest distance is selected as the affiliation of the point trace; the above method is used to calculate and classify each point trace separately, and finally all the point traces are matched with the corresponding initial centroids to obtain multiple clustering sets.

[0124] In some specific embodiments, S44 specifically includes:

[0125] The cluster set The point traces and centroids in the graph are defined as a weighted undirected graph , Represented as a set of vertices (where the vertices include … ), Represented as an edge set, Connect points on the graph and The number of paths, is the number of vertices, edges , is the vertex number, , then the dot trace With the centroid Density sensitive distance between Defined as:

[0126] ;

[0127] The shortest geometric density sensitive distance for:

[0128] ;

[0129] in, To connect the traces With the centroid The set of all paths of For the Vertices With +1 vertex The Euclidean distance between is the scaling factor, , is a natural constant.

[0130] S45, according to the clustering set The cluster centroid is calculated using the following formula: ;

[0131] ;

[0132] in, is the sequence number of the point trace, in this formula The range is 1- , For the The cluster set formed by the initial centroids The number of midpoint traces, The cluster set Midpoint The three-dimensional coordinates of .

[0133] S46, determine the current cluster centroid Whether the preset iteration conditions are met , if satisfied, then the current cluster centroid Save the current cluster centroid list ; If not satisfied, then the cluster centroid is used as the initial centroid and S44-S46 are repeatedly executed; wherein the iteration condition It means the cluster centroid obtained by S44 and S45 Becoming stable.

[0134] S105, respectively calculating the mean square errors of the plurality of cluster centroid lists, and selecting a group of cluster centroids in the cluster centroid lists with the smallest mean square error as the fused trajectory.

[0135] In summary, the present invention provides a method for quickly performing traversal clustering on multiple cluster numbers based on a high-density point trace set, and simultaneously uses density-sensitive distance to locally calculate the points and key data (initial centroids obtained based on the high-density point trace set), thereby reducing the amount of calculation while effectively ensuring the accuracy of the fusion.

[0136] Specifically, by preprocessing the scattered point traces, a high-density point trace set with a more concentrated point trace distribution is obtained, and then the initial centroid is selected from it. Compared with selecting the centroid in the sparsely distributed point traces, the initial centroid obtained by this method is more representative, and the calculation speed is much faster than obtaining the centroid in the scattered point traces. That is to say, by introducing a high-density point trace set, the application provides a more accurate initialization data (initial centroid) for subsequent data processing while ensuring calculation efficiency; at the same time, the density-sensitive distance is introduced to associate the point traces with the initial centroid, that is, the point traces are respectively calculated and classified with the finite data (initial centroid) to obtain a cluster set, and a finite number of cluster centroids are calculated in the finite cluster set; based on the above method, the number of iterations can be reduced, the amount of calculation can be greatly reduced, and the data fusion efficiency and accuracy can be improved.

[0137] Furthermore, based on the above clustering method, the present application selects multiple cluster numbers to calculate the cluster centroid list respectively, and selects the optimal cluster centroid list to generate the trajectory. That is to say, this scheme is actually based on a more accurate initialization data and a simplified cluster centroid list acquisition method, and selects multiple cluster numbers to calculate the cluster centroid list under the cluster number respectively, and finally selects the cluster centroid in the cluster centroid list with the smallest mean square error as the fused trajectory. The fusion accuracy of the final trajectory is improved by improving the accuracy of cluster number selection, but at the same time it will not increase the workload.

[0138] Embodiment 2

[0139] See also Figure 2 The present invention also provides a multi-radar target detection data fusion system, comprising:

[0140] A data acquisition module: configured to acquire N initial trajectories of at least one target through N radars respectively;

[0141] In some specific embodiments, the data acquisition module is configured to:

[0142] set up The goal of the moment Status , and through the state Get Target The initial state covariance matrix ;

[0143] According to the initial state covariance matrix , predict the target through the Kalman filter algorithm exist The first state covariance matrix at time , filter gain matrix and the innovation covariance matrix ;

[0144] Will The first state covariance matrix predicted at time The difference of the second state covariance matrix at the moment is multiplied by the target measurement noise covariance matrix to obtain the similarity of the current target , and set the tracking threshold , if satisfied ,but , that is, Measurements fall within target In the tracking gate, The values ​​of form confirmation matrix If not satisfied ,but , that is, Measurements not on target In the tracking gate of , the data does not belong to the current target; Indicates measurement Falling into the target Binary variable of the tracking gate;

[0145] Definition The joint associated events are , using the third and fourth rules to transform the confirmation matrix Split to get the joint event correlation matrix ;

[0146] For the joint association event and the joint event correlation matrix Calculate and get the conditional probability of the associated event ;

[0147] The conditional probability of the associated event Calculated Measurement of time With the goal The probability of association ;

[0148] Select The target with the largest value is associated with the current measurement to form a local tracking track of the current target, and the location trace corresponding to the measurement is included in the track list of the current target;

[0149] Generate new tracking tracks for the points corresponding to the unassociated measurements and update the current track list status;

[0150] Repeat the above steps to form the local tracking trajectory at the next moment until a complete initial trajectory is formed.

[0151] Clustering module: configured to define the N initial trajectories as point trace sets , and calculate the number of cluster ranges ; Select multiple cluster numbers in the cluster range number , the number of clusters Take a positive integer; and respectively according to the plurality of clustering numbers Clustering the point trace set to obtain a plurality of cluster centroid lists;

[0152] In some specific embodiments, the clustering module is configured to calculate the current cluster number The average density parameter of the multiple point sets described below ;

[0153] According to the average density parameter Calculate the regional node density of each point in the N point sets respectively , select the points satisfying the first rule to form a high-density point set ;

[0154] According to the currently acquired high-density point set Calculate multiple initial centroids and form an initial centroid set; calculate the point trace set The density-sensitive distance between each of the points and each of the initial centroids in ; and select the initial centroid with the shortest density-sensitive distance as the attribution result of the current point trace to form a clustering set ;

[0155] According to the clustering set Calculate the cluster centroid ; and determine the current cluster centroid Whether the iteration condition is met , if satisfied, then the current cluster centroid Save the current cluster centroid list , otherwise recalculate the cluster centroid .

[0156] The trajectory fusion module is configured to respectively calculate the mean square errors of the plurality of cluster centroid lists, and select a group of cluster centroids in the cluster centroid list with the smallest mean square error as the fused trajectory.

[0157] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0158] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for fusion of multi-radar target detection data, characterized in that: Includes steps: S101, obtaining N initial tracks of at least one target through N radars respectively; S102, defining the N initial trajectories as a plurality of point trace sets, and calculating the number of cluster ranges according to the plurality of point trace sets; wherein the point trace set includes: a point trace corresponding to the target detected by the radar at a plurality of time instants, and the point trace is the spatial coordinate of the target; S103, selecting a plurality of cluster numbers from the cluster range number, wherein the cluster number is a positive integer; S104, clustering the point trace set according to the plurality of cluster numbers respectively, so as to obtain a plurality of cluster centroid lists accordingly; wherein S104 includes: S41, calculating the average density parameter of the plurality of point trace sets under the current number of clusters; S42, calculating the regional node density of each point trace in the point trace set according to the average density parameter, and selecting the point traces that meet the first rule to form a high-density point trace set; wherein the first rule requires that the regional node density of the point trace is greater than or equal to the average density parameter; S43, calculating at least two initial centroids according to the high-density point trace set, and forming an initial centroid set; wherein S43 includes: Selecting the centroid with the largest node density in the region in the high-density point trace set as the first initial centroid, and selecting at least one initial centroid in the high-density point trace set in turn using the second rule according to the first initial centroid, and the total number of initial centroids is the same as the number of clusters; The second rule requires that the selection of at least one initial centroid must satisfy a maximum distance condition, and the maximum distance condition requires that the point trace with the largest distance from the previous initial centroid be selected as the next initial centroid; S44, calculating the density-sensitive distance between each of the point traces in the point trace set and each of the initial centroids, and selecting the initial centroid with the shortest density-sensitive distance as the attribution result of the current point trace, so as to form a plurality of cluster sets; S45, calculating the cluster centroid according to the cluster set using the following formula; wherein, ； in, For the cluster centroids, The range is 1- , For the The cluster set formed by the initial centroids The number of midpoint traces, The cluster set Midpoint The three-dimensional coordinates of S46, judging whether the cluster centroid satisfies the iteration condition If satisfied, the cluster centroid is stored in the cluster centroid list; if not satisfied, the cluster centroid is used as the initial centroid and S44-S46 are repeatedly executed; S105, respectively calculating the mean square errors of the plurality of cluster centroid lists, and selecting the cluster centroids in a group of cluster centroid lists with the smallest mean square error as the fused trajectory.

2. The method for fusion of multi-radar target detection data according to claim 1, characterized in that: S101 includes: S11, set the initial moment target Status , and through the state Obtaining an initial state covariance matrix of the target; S12, using the initial state covariance matrix, using the Kalman filter algorithm to predict the target The first state covariance matrix at time ; S13, The first state covariance matrix predicted at the moment is The difference of the second state covariance matrix actually measured at the moment is multiplied by the target measurement noise covariance matrix to obtain the similarity of the target , and set the tracking threshold , if satisfied ,but , that is, The measurement falls into the tracking gate of the target. The value of forms a confirmation matrix; if it does not meet ,but , that is, A measurement does not fall into the tracking gate of the target, and the data does not belong to the current target; in, Indicates a binary variable measuring the number of tracking gates that fall into the target; S14, definition a joint correlation event, and according to the joint correlation event, the confirmation matrix is ​​split by using the third rule and the fourth rule to obtain a joint event correlation matrix; The third rule means that each target can only produce one valid measurement result, and the fourth rule means that each measurement result can only come from a certain target or clutter; S15, calculating the joint associated events and the joint event association matrix to obtain the conditional probability of the associated events; S16, calculated from the conditional probability of the associated event The moment The probability of association of a measurement with the target; S17, associating the measurement with the largest association probability value with the target to form a local tracking trajectory of the target, and adding the measurement to a trajectory list of the target; S19, repeatedly executing S12-S17 to form a local tracking trajectory at the next moment, until a complete initial trajectory is formed.

3. The method for fusion of multi-radar target detection data according to claim 1, characterized in that: In S41, the calculation formula of the average density parameter is: ； in, is a set of points, is the average density parameter, is the number of clusters.

4. The method for fusion of multi-radar target detection data according to claim 3 is characterized in that: In S42, the calculation formula of the regional node density function is: ； in, For the The regional node density function of the points is The range is 1- , Respectively represent the points and dot trace location.

5. The method for fusion of multi-radar target detection data according to claim 2, characterized in that: Before S19, it also includes the following steps: S18, generating a new tracking trajectory for the unassociated measurement and updating the current trajectory list state.

6. The method for fusion of multi-radar target detection data according to claim 2, characterized in that: The first state covariance matrix in S12 is: ； in, For the goal, is the first state covariance matrix, is the state transfer matrix, for Time target The state covariance matrix of , T is the transpose of the matrix, for The goal of the moment The process noise matrix.

7. The method for fusion of multi-radar target detection data according to claim 2, characterized in that: In S15, the calculation formula of the conditional probability of the associated event is: ； in: is the conditional probability of the associated event, is the normalization constant; is the impurity density; is the number of measurements that are not associated with the target in the joint association event; is the prior Gaussian probability distribution of the measurement; for The moment Measurements, For the joint association event Measurements and Targets Interconnection instructions; It is a target detection indication for a joint correlation event; and The values ​​are all 1 or 0; For the goal The detection probability of is the total number of measurements; is the total number of tracked targets.

8. The method for fusion of multi-radar target detection data according to claim 7, characterized in that: In S16, The moment Measurements and Targets The probability of association The calculation is done using the following formula: ； in, is the joint event correlation matrix.

9. A multi-radar target detection data fusion system, characterized in that: include: A data acquisition module: configured to acquire N initial trajectories of at least one target through N radars respectively; Clustering module: configured to define the N initial trajectories as point trace sets respectively, and calculate the number of cluster ranges; select multiple cluster numbers from the cluster ranges, and the cluster numbers are positive integers; and cluster the point trace sets according to the multiple cluster numbers respectively, so as to obtain multiple cluster centroid lists correspondingly; wherein, when obtaining the cluster centroid list, the clustering module is specifically configured to: Calculate the average density parameter of the plurality of point trace sets under the current number of clusters; Calculate the regional node density of each point trace in the N point trace sets according to the average density parameter, and select the point traces that meet the first rule to form a high-density point trace set; At least two initial centroids are calculated based on the high-density point trace set to form an initial centroid set; the density-sensitive distance between each of the point traces in the point trace set and each of the initial centroids is calculated; and the initial centroid with the shortest density-sensitive distance is selected as the attribution result of the point trace to form a cluster set; The cluster centroid is calculated according to the cluster set; and whether the cluster centroid satisfies the iteration condition is determined. If so, the cluster centroid is stored in the cluster centroid list; otherwise, the cluster centroid is recalculated; The trajectory fusion module is configured to respectively calculate the mean square errors of the plurality of cluster centroid lists, and select a group of cluster centroids in the cluster centroid list with the smallest mean square error as the fused trajectory.

Images