Multi-sensor multi-extended target tracking method and device based on improved ellipticity

Abstract

The present invention discloses a multi-sensor multi-extended target tracking method and device based on improved elliptic metrics. The present invention uses Log-Cholesky distance to improve the step of associating the same Bernoulli components under different sensors in the arithmetic mean fusion algorithm, which can not only provide a stable centroid distance metric, but also capture the slight differences in the shape matrices of different extended targets, and realize the effective association of the posterior densities of the same extended target under different sensors; at the same time, in the local sensor filtering stage, the present invention effectively solves the problem of missed detection of targets caused by the measurement distribution being closely divided into the same subset by compensating for the Bernoulli components corresponding to the lost targets in the adjacent area.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-sensor multi-extended target tracking, and in particular to a multi-sensor multi-extended target tracking method and device based on improved ellipticity. Background Art

[0002] With the continuous improvement of the resolution capabilities of modern radar, infrared and other advanced sensors, extended target tracking technology has become one of the core technologies in the national defense development strategies of various countries. This technology not only plays a key role in the military field, but also has a wide range of application scenarios in the civilian field. Unlike traditional point target tracking technology, extended target tracking technology needs to deal with targets that occupy multiple sensor resolution units, which generate multiple measurements at each time step. Extended target tracking requires not only estimating the center of mass state of the target, but also estimating the shape and contour of the target. Since the ellipse can concisely approximate the shape of the actual target and provide the size and direction information of the target, the spatial shape of the extended target is usually modeled as an ellipse. The researchers introduced a random matrix model to describe the elliptical shape, and based on the random matrix method, further modeled the target state as a Gamma Gaussian Inverse Wishart (GGIW) distribution, and combined the GGIW distribution with the Poisson multi-Bernoulli mixture (PMBM) filter to propose a GGIW-PMBM filter. This filter can not only handle complex target motion and shape changes, but also achieve a good balance between computational cost and tracking performance. The results show that the GGIW-PMBM filter outperforms other filters based on random finite sets (RFS) in terms of tracking performance and computational cost. This makes the algorithm promising in practical applications and can provide a reliable solution for multi-target tracking in various complex scenarios. With this improved filtering algorithm, the application of extended target tracking technology in defense and civilian fields will be further promoted and deepened.

[0003] In the field of multi-sensor multi-extended target tracking, the accuracy of target tracking depends largely on two key processes: local sensor filtering and multi-sensor information fusion. The main task of the local sensor filtering stage is to process the measurement data of a single sensor and make a preliminary estimate of the target state. The multi-sensor information fusion stage focuses on integrating the measurement data from different sensors to obtain a more accurate and comprehensive target state estimate.

[0004] However, in practical applications, when multiple extended targets are in adjacent or intersecting areas, the close distance between the centroids may easily lead to missed detection of the target, making tracking difficult. In the local sensor filtering stage, the measurement set at the current moment needs to be divided. If the distance between the measurement values ​​is less than the predetermined threshold, these measurement values ​​will be considered to be generated by the same extended target and divided into the same subset. However, due to the close distance between the centroids, the measurement distribution is relatively dense, resulting in the measurement values ​​from different targets being mistakenly divided into the same subset, making the estimated number of targets less than the actual number of targets, which seriously affects the target estimation accuracy.

[0005] In the multi-sensor information fusion stage, the choice of fusion mode has an important impact on the performance of multi-extended target tracking. After long-term exploration and research by experts in the field of target tracking, researchers have tried to study the multi-target multi-sensor density fusion algorithm based on random finite sets, and have achieved a series of research results. At present, multi-sensor density fusion is mainly divided into generalized covariance intersection fusion (GCI) and arithmetic average (AA). Among them, GCI fusion is more sensitive to missed detection, while AA fusion has stronger anti-missing detection ability and robustness. Researchers proposed a distributed multi-sensor AA fusion method to adapt to the multi-Bernoulli process. It is necessary to decompose the multi-Bernoulli AA fusion process into parallel Bernoulli fusion sub-processes, and then associate and fuse the Bernoulli components representing the same target under different sensors. Since the Bernoulli component contains the state information of the extended target, how to associate the same extended target under different sensors is crucial to the subsequent AA fusion process.

[0006] However, when the centroids of different extended targets are close, their Bernoulli components can be easily associated into the same subset, resulting in multiple extended targets being miscalculated as one extended target. In this case, choosing an appropriate elliptical shape metric has an important impact on associating the same extended target under different sensors. Today, the most commonly used elliptical extended target metric is the Gaussian Wasserstein (GW) distance. However, the GW distance is defined as the distance between different Gaussian distributions, while the centroid of the elliptical extended target follows a Gaussian distribution, while the shape matrix follows an IW distribution. In addition, when measuring different shape matrices, using the GW distance may not be as good as using a metric specifically designed for the SPD shape matrix. Summary of the invention

[0007] Based on the problems raised by the above background technology, the purpose of the present invention is to provide a multi-sensor multi-extended target tracking method and equipment based on improved elliptical metrics. In the filtering stage, the probability of the existence of the Bernoulli component of the lost target in the adjacent area is compensated, thereby effectively alleviating the problem of target underestimation caused by the tight measurement distribution; in the multi-sensor information fusion stage, the Log-Cholesky distance is used as the elliptical extended target shape measurement indicator to better capture the differences between different shape matrices, which is suitable for processing shape matrices that obey the IW distribution. When the execution distances of different extended targets are close, the Bernoulli posterior density of the unified extended target under different sensors is effectively associated.

[0008] The present invention is achieved through the following technical solutions:

[0009] A first aspect of the present invention provides a multi-sensor multi-extended target tracking method based on improved ellipticity, comprising the following steps:

[0010] Step S1, obtaining elliptical extended targets of multiple sensors, using the GGIW-PMBM algorithm to predict and update the elliptical extended targets, and obtaining posterior density information of multiple sensors;

[0011] Step S2: using a distributed flooding algorithm to iteratively update the posterior density information of multiple sensors, and fusing the iteratively updated posterior density information to complete flooding communication;

[0012] Step S3, after completing the flooding communication, using Log-Cholesky metric to associate the Bernoulli components of the same elliptical extended target to form an associated Bernoulli subset;

[0013] Step S4: fuse the associated Bernoulli subsets to obtain a fused ellipse extended target, extract the state of the fused ellipse extended target, and return a state estimation value.

[0014] In the above technical scheme, in the filtering stage of the local sensor, the GGIW-PMBM algorithm is first used to predict and update the elliptical extended target of the local sensor to determine whether there is a lost target in the vicinity of the local sensor. If there is a lost target, it needs to be compensated, which effectively alleviates the problem of target underestimation caused by the close measurement distribution.

[0015] Secondly, in the multi-sensor information fusion stage, a distributed flooding algorithm is first used to propagate the posterior density information. In the sensor network composed of multiple sensors, each local sensor is equivalently regarded as a fusion center. The posterior density information from other sensors is fused through the distributed flooding algorithm, and the posterior density information of the local sensor is updated to obtain accurate target state estimation in a multi-sensor environment.

[0016] In the multi-sensor information fusion stage, accurately associating the same extended target under different sensors is the core of multi-sensor multi-extended target tracking. However, when the centroids of different extended targets are close, their Bernoulli components are easily associated with the same subset, resulting in multiple extended targets being misestimated as one extended target. Since the centroid of the elliptical extended target obeys the Gaussian distribution, but the shape matrix obeys the IW distribution, this method proposes an improved elliptical extended target shape metric, the Log-Cholesky distance, as an elliptical extended target metric. Compared with the currently used GW distance, the Log-Cholesky distance can better capture the differences between different shape matrices and is suitable for processing shape matrices that obey the IW distribution. When the target centroids are close, the Bernoulli posterior density information of the same extended target under different sensors is effectively associated.

[0017] In an optional embodiment, the step S1 of using the GGIW-PMBM algorithm to predict and update the elliptical extended target includes the following steps:

[0018] Step S11, arbitrarily selecting two elliptical extended targets, and obtaining the extended target states of the two elliptical extended targets at time k; wherein the extended target states include Poisson measurement rate, motion state vector and shape matrix;

[0019] Step S12: calculating the center-of-mass Euclidean distance of the two elliptical extended objects based on the motion state vector; if the center-of-mass Euclidean distance is less than or equal to a distance threshold, the two elliptical extended objects are in a neighboring area;

[0020] Step S13, calculating the number of targets at time k, comparing the number of targets at time k with the number of targets at time k-1, if the number of targets decreases and the reduced elliptical expanded targets are located in the adjacent area, compensating for the lost targets in the adjacent area.

[0021] In an optional embodiment, the step S2 of using a distributed flooding algorithm to communicate the posterior density information of multiple sensors includes the following steps:

[0022] The sensor serving as the fusion center is denoted as sensor s;

[0023] Record the set of sensors adjacent to the sensor s as a neighbor sensor set, and select any sensor from the neighbor sensor set as a neighbor sensor i;

[0024] The iterative information received by sensor s from neighbor sensor i at the tth iteration is obtained, the iterative information of sensor s at the tth iteration is fused with the iterative information from neighbor sensor i, and the fused iterative information is used as the iterative information of sensor s at the t+1th iteration for the next iteration.

[0025] In an optional embodiment, fusing the iteration information of sensor s at the tth iteration with the iteration information of neighbor sensor i at the tth iteration includes:

[0026]

[0027] In the above formula, is the iteration information at sensor s at the tth iteration, x is the target motion state information, is the iterative information received by sensor s from neighbor sensor i at the tth iteration, S s is the set of neighbor sensors, p i (x) is the posterior density of neighbor sensor i.

[0028] In an optional embodiment, the fusing of the posterior density information after communication in step S2 includes:

[0029]

[0030] In the above formula, Represents the information set at sensor s after the iteration The set of sensors involved, p i (x) is the posterior density of sensor i, and x is the target motion state information.

[0031] In an optional embodiment, the step S3 of associating Bernoulli components of the same elliptical extended target using Log-Cholesky metric includes the following steps:

[0032] Obtain the shape matrices of the two elliptical extended targets, use the Cholesky algorithm to inversely decompose the shape matrices, and generate two inverse decomposition matrices;

[0033] The Log-Cholesky metric is used to calculate the Log-Cholesky distance of two inverse decomposition matrices in the triangular matrix space, and the Log-Cholesky distance of the shape matrix is ​​obtained;

[0034] Calculate the centroid Euclidean distance of the shape matrices of the two elliptical extended targets, and combine the centroid Euclidean distance and the shape matrix Log-Cholesky distance to obtain the distance between the two elliptical extended targets;

[0035] The distance between the two elliptical extended targets is compared with a threshold value. If the distance between the two elliptical extended targets is less than the threshold value, the Bernoulli components of the two elliptical extended targets are associated.

[0036] In an optional embodiment, calculating the Log-Cholesky distance of two inverse decomposition matrices in a triangular matrix space using the Log-Cholesky metric includes:

[0037]

[0038]

[0039] In the above formula, X and represents the shape matrix of two elliptical extension targets, P and Q represent the two inverse decomposition matrices of the shape matrix of two elliptical extension targets, Indicates L + The distance function of the P and Q matrices in space, Indicated in The distance function of the shape matrices of two ellipsoidally extended targets in space, It is the lower triangular part in the strict sense; represents the diagonal part of the matrix, ||·|| F Represents the 2-norm of the matrix.

[0040] A second aspect of the present invention provides a multi-sensor multi-extended target tracking system based on improved ellipticity, comprising:

[0041] A target detection module, wherein the target detection module is used to obtain elliptical extended targets of multiple sensors, and use a GGIW-PMBM algorithm to predict and update the elliptical extended targets to obtain posterior density information of multiple sensors;

[0042] A flooding communication module, wherein the flooding communication module is used to iteratively update the posterior density information of multiple sensors using a distributed flooding algorithm, and to fuse the iteratively updated posterior density information to complete flooding communication;

[0043] An association module, wherein the association module is used to associate Bernoulli components of the same elliptical extended target using Log-Cholesky metric after completing flooding communication to form an associated Bernoulli subset;

[0044] A state extraction module is used to fuse the associated Bernoulli subsets to obtain a fused ellipse extended target, perform state extraction on the fused ellipse extended target, and return a state estimation value.

[0045] A third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a multi-sensor multi-extended target tracking method based on improved ellipticity is implemented.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a multi-sensor multi-extended target tracking method based on improved ellipticity.

[0047] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0048] 1. The present invention uses Log-Cholesky distance to improve the step of associating the same Bernoulli component under different sensors in the arithmetic mean fusion algorithm, which can better capture the differences between different shape matrices and is suitable for processing shape matrices that obey IW distribution. The elliptical shape metric index can effectively associate the Bernoulli posterior density information of the same extended target under different sensors when the target center of mass is close;

[0049] 2. In the filtering stage, the present invention addresses the problem of missed target estimation due to dense measurement distribution of neighboring targets, and effectively improves the tracking performance of neighboring extended targets by compensating for the existence probability of the Bernoulli component of lost targets in the neighboring area. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0051] Figure 1 A schematic diagram of a flow chart of a multi-sensor multi-extended target tracking method based on improved ellipticity provided in Embodiment 1 of the present invention;

[0052] Figure 2 A schematic diagram of the structure of a multi-sensor multi-extended target tracking method based on improved ellipticity provided in Example 1 of the present invention;

[0053] Figure 3 The sensor and the real extended target trajectory map provided in Example 1 of the present invention;

[0054] Figure 4 A target quantity estimation diagram provided by Embodiment 1 of the present invention;

[0055] Figure 5The OSPA error diagram provided by Example 1 of the present invention;

[0056] Figure 6 A schematic diagram of the structure of an electronic device provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0058] Example 1

[0059] Figure 1 A flow chart of a multi-sensor multi-extended target tracking method based on improved ellipticity provided in Example 1 of the present invention is shown in FIG. Figure 1 As shown, the multi-sensor multi-extended target tracking method based on improved ellipse measurement includes the following steps:

[0060] Step S1, obtaining elliptical extended targets of multiple sensors, using the GGIW-PMBM algorithm to predict and update the elliptical extended targets, and obtaining posterior density information of multiple sensors;

[0061] Step S2: using a distributed flooding algorithm to iteratively update the posterior density information of multiple sensors, and fusing the iteratively updated posterior density information to complete flooding communication;

[0062] Step S3, after completing the flooding communication, using Log-Cholesky metric to associate the Bernoulli components of the same elliptical extended target to form an associated Bernoulli subset;

[0063] Step S4: fuse the associated Bernoulli subsets to obtain a fused ellipse extended target, extract the state of the fused ellipse extended target, and return a state estimation value.

[0064] It should be noted that in the filtering stage of the local sensor, the GGIW-PMBM algorithm is first used to predict and update the elliptical extended target of the local sensor to determine whether there is a lost target in the vicinity of the local sensor. If there is a lost target, it needs to be compensated, which effectively alleviates the problem of target underestimation caused by the close measurement distribution.

[0065] Secondly, in the multi-sensor information fusion stage, a distributed flooding algorithm is first used to propagate the posterior density information. In the sensor network composed of multiple sensors, each local sensor is equivalently regarded as a fusion center. The posterior density information from other sensors is fused through the distributed flooding algorithm, and the posterior density information of the local sensor is updated to obtain accurate target state estimation in a multi-sensor environment.

[0066] In the multi-sensor information fusion stage, accurately associating the same extended target under different sensors is the core of multi-sensor multi-extended target tracking. However, when the centroids of different extended targets are close, their Bernoulli components are easily associated with the same subset, resulting in multiple extended targets being misestimated as one extended target. Since the centroid of the elliptical extended target obeys the Gaussian distribution, but the shape matrix obeys the IW distribution, this method proposes an improved elliptical extended target shape metric, the Log-Cholesky distance, as an elliptical extended target metric. Compared with the currently used GW distance, the Log-Cholesky distance can better capture the differences between different shape matrices and is suitable for processing shape matrices that obey the IW distribution. When the target centroids are close, the Bernoulli posterior density information of the same extended target under different sensors is effectively associated.

[0067] In an optional embodiment, the step S1 of using the GGIW-PMBM algorithm to predict and update the elliptical extended target includes the following steps:

[0068] Step S11, arbitrarily selecting two elliptical extended targets, and obtaining the extended target states of the two elliptical extended targets at time k; wherein the extended target states include Poisson measurement rate, motion state vector and shape matrix;

[0069] Step S12: calculating the center-of-mass Euclidean distance of the two elliptical extended objects based on the motion state vector; if the center-of-mass Euclidean distance is less than or equal to a distance threshold, the two elliptical extended objects are in a neighboring area;

[0070] Step S13, calculating the number of targets at time k, comparing the number of targets at time k with the number of targets at time k-1, if the number of targets decreases and the reduced elliptical expanded targets are located in the adjacent area, compensating for the lost targets in the adjacent area.

[0071] It should be noted that the purpose of step S1 is to determine whether there is a lost target in the vicinity of the local sensor. If there is a lost target, a probability value compensation is required for it.

[0072] Specifically, in two-dimensional space, the extended target state of the i-th elliptical extended target at time k is recorded as in, represents the Poisson measurement rate of the i-th extended target at time k, represents the motion state vector of the i-th extended target at time k, Represents the shape matrix of the i-th extended target at time k.

[0073] The motion state vectors of the two elliptical extension targets at time k are used to calculate the center-of-mass Euclidean distance. When the center-of-mass Euclidean distance of the two elliptical extension targets meets the condition: When , the two targets are considered to be in the adjacent area. Record the number of targets at the current time k and compare it with the number of targets at the previous time k-1. If the number of targets decreases and the reduced targets are in the adjacent area, the target is considered lost, and the probability value of the Bernoulli component of the lost target in the adjacent area is compensated, that is: r k|k =r k|k ×N,r k|k is the existence probability, and N is the number of targets in the neighboring area.

[0074] If there is no target loss or there is a target loss and the low-weight Bernoulli components are pruned and merged after compensating for the lost target, the processed multi-Bernoulli mixture distribution will participate in the prediction process at the next moment. At the same time, the Bernoulli mixture distribution at time k will be used as the posterior density information for subsequent processing.

[0075] In an optional embodiment, the step S2 of using a distributed flooding algorithm to communicate the posterior density information of multiple sensors includes the following steps:

[0076] The sensor serving as the fusion center is denoted as sensor s;

[0077] Record the set of sensors adjacent to the sensor s as a neighbor sensor set, and select any sensor from the neighbor sensor set as a neighbor sensor i;

[0078] The iterative information received by sensor s from neighbor sensor i at the tth iteration is obtained, the iterative information of sensor s at the tth iteration is fused with the iterative information from neighbor sensor i, and the fused iterative information is used as the iterative information of sensor s at the t+1th iteration for the next iteration.

[0079] In an optional embodiment, fusing the iteration information of sensor s at the tth iteration with the iteration information of neighbor sensor i at the tth iteration includes:

[0080]

[0081] In the above formula, is the iteration information at sensor s at the tth iteration, x is the target motion state information, is the iterative information received by sensor s from neighbor sensor i at the tth iteration, S s is the set of neighbor sensors, p i (x) is the posterior density of neighbor sensor i.

[0082] It should be noted that the purpose of using a distributed flooding algorithm to communicate the posterior density information of multiple sensors in step S2 is to fuse the posterior density information from other sensors through the distributed flooding algorithm and update the posterior density information of the local sensor to obtain accurate target state estimation in a multi-sensor environment.

[0083] Sensor s is used as the fusion center, which is responsible for iterating neighborhood communication and combining and fusing information from other sensors. The set of neighbor sensors of sensor s is denoted as S s , any one of the neighbor sensor sets is selected as the neighbor sensor i. In this embodiment 1, the fusion of sensor s and neighbor sensor i is used as an example to implement the communication process.

[0084] Specifically, the initial information p of sensor s is s (x) and p as the initial information of neighbor sensor i i (x) The following formula is used for fusion:

[0085]

[0086] Get the iteration information required by sensor s for the next iteration.

[0087] For the iterative information newly received by sensor s from neighbor sensor i at the tth iteration Calculate using the following formula:

[0088]

[0089] For the 0th iteration, the posterior density p of sensor s is s (x) is the initial information, i.e., the information of sensor s at the 0th iteration, For the iterative information received by sensor s from neighbor sensor i at iteration 0, the posterior density p of neighbor sensor i is taken at iteration 0. i (x).

[0090] For the t-th iteration (when t≥1), for the iterative information of sensor s at the t-th iteration, the result of the previous iterative fusion is taken as the iterative information of this time. For the iterative information newly received from the neighbor sensor i by sensor s at the t-th iteration, the set difference between the iterative information of sensor s at the t-th iteration and the iterative information of sensor s at the t-1-th iteration is taken as the iterative information received by sensor s from the neighbor sensor i at the t-th iteration.

[0091] The iteration information of sensor s at the tth iteration is fused with the iteration information received from neighbor sensor i at the tth iteration as the iteration information of sensor s at the t+1th iteration for the next iteration until the iteration maximum value is reached, thereby completing the flooding communication of adjacent sensors.

[0092] The number of communication iterations is t=0, 1, ..., T, and the maximum iteration value T is set by those skilled in the art according to actual needs.

[0093] In an optional embodiment, the fusing of the posterior density information after communication in step S2 includes:

[0094]

[0095] In the above formula, Represents the information set at sensor s after the iteration The set of sensors involved, p i (x) is the posterior density of sensor i, and x is the target motion state information.

[0096] It should be noted that after completing flooding communication, sensor s will receive information from all sensors that are T hops away from it, where T is the maximum number of communication iterations; all sensors participating in the iteration in flooding communication are considered to be the information set The sensor set involved fuses the posterior density in the above sensor set to achieve the fusion of the posterior density information.

[0097] Among them, the information set The set of sensors involved includes sensor s itself.

[0098] In an optional embodiment, the step S3 of associating Bernoulli components of the same elliptical extended target using Log-Cholesky metric includes the following steps:

[0099] Obtain the shape matrices of the two elliptical extended targets, use the Cholesky algorithm to inversely decompose the shape matrices, and generate two inverse decomposition matrices;

[0100] The Log-Cholesky metric is used to calculate the Log-Cholesky distance of two inverse decomposition matrices in the triangular matrix space, and the Log-Cholesky distance of the shape matrix is ​​obtained;

[0101] Calculate the centroid Euclidean distance of the shape matrices of the two elliptical extended targets, and combine the centroid Euclidean distance and the shape matrix Log-Cholesky distance to obtain the distance between the two elliptical extended targets;

[0102] The distance between the two elliptical extended targets is compared with a threshold value. If the distance between the two elliptical extended targets is less than the threshold value, the Bernoulli components of the two elliptical extended targets are associated.

[0103] It should be noted that the distance between different elliptical extended targets includes the centroid distance between the elliptical extended targets and the shape matrix distance of the two elliptical extended targets. The current problem is that the centroid distance and the shape matrix distance are both calculated using the GW distance. Among them, the elliptical measurement method currently used in the field of extended target tracking is the GW distance, which is defined as:

[0104]

[0105] Among them, ξ x =(x,X) and Represent the states of different extended targets respectively, ||x|| represents the 2-norm of x, and Tr(·) is the trace of the matrix. However, the GW distance is defined as the distance between two Gaussian probability distributions, while the center of mass of the elliptical extended target obeys the Gaussian distribution, and the shape matrix obeys the IW distribution. When two extended targets are in a neighboring area, the distance between the centers of mass is relatively close. At this time, if different elliptical extended targets are to be distinguished, an indicator that can better distinguish the slight differences between different shape matrices is needed. It is more appropriate to use a distance measurement method designed for the covariance matrix than to use the GW distance. For the elliptical shape matrix that obeys the IW distribution, the present invention uses the Log-Cholesky metric to measure the relationship between different symmetric positive definite (SPD) matrices.

[0106] In an optional embodiment, calculating the Log-Cholesky distance of two inverse decomposition matrices in a triangular matrix space using the Log-Cholesky metric includes:

[0107]

[0108] In the above formula, X and represents the shape matrix of two elliptical extension targets, P and Q represent the two inverse decomposition matrices of the shape matrix of two elliptical extension targets, Indicates L + The distance function of the P and Q matrices in space, Indicated in The distance function of the shape matrices of two ellipsoidally extended targets in space, It is the lower triangular part in the strict sense; represents the diagonal part of the matrix, ||·|| F Represents the 2-norm of the matrix.

[0109] It should be noted that the principle of the present invention using Log-Cholesky metric to calculate the distance between shape matrices is to introduce the Log-Cholesky metric in the lower triangular matrix space with diagonal elements, and then expand it to the SPD matrix space through Cholesky decomposition.

[0110] Among them, the shape matrices X and For two d*d dimensional SPD matrices, use the formula:

[0111]

[0112] For two shape matrices X and Calculate and get The distance between the shape matrices of two elliptical extended targets in space, Defined as a collection of m×m SPD matrices.

[0113] In the present invention, L + Space to study The properties of the space, where the space of m×m lower triangular matrices is denoted by L, L + represents a subset of L whose diagonal elements are all positive.

[0114] Specifically, the shape matrices of different elliptical expansion targets are subjected to Cholesky inverse decomposition, and P and Q are called X and The Cholesky factorization of X = PP T , Recorded as In L + Calculate the distance between P and Q matrices in space and then transform them into In space, the shape matrix Log-Cholesky distance between the two elliptical extended target shape matrices is obtained.

[0115] At the same time, since the distance between two elliptical extension targets includes the centroid Euclidean distance and the shape matrix Log-Cholesky distance, to obtain the distance between the elliptical extension targets, it is necessary to combine the centroid Euclidean distance and the shape matrix Log-Cholesky distance, which is calculated as follows:

[0116]

[0117] A threshold Dmax is set to control the associated subset. In the present invention, the distance between two elliptical extended targets is compared with the threshold Dmax. If the distance between the two elliptical extended targets is less than the threshold Dmax, it is considered that the two Bernoulli components represent a unified extended target and are associated with the same subset.

[0118] The associated Bernoulli subsets are fused by arithmetic mean, and the state is extracted, and the estimated value is returned to complete the tracking of multi-sensor multi-extended targets.

[0119] Figure 2 The schematic diagram of the structure of the multi-sensor multi-extended target tracking method based on improved ellipticity provided in Example 1 of the present invention is as follows: Figure 2 As shown, the present invention is based on the existing GGIW-PMBM algorithm and improves the MBM mixed distribution using the Log-Cholesky metric, so that it obeys the IW distribution in the shape matrix association, and effectively associates the Bernoulli posterior density information of the same extended target under different sensors when the target center of mass is close. AA fusion is performed after accurately associating different extended targets. Compared with the multi-target tracking method that uses the GW distance indicator to associate extended targets, the present invention effectively improves the tracking performance of adjacent extended targets.

[0120] Example 1 of the present invention uses a specific implementation case to demonstrate the superiority of the multi-sensor multi-extended target tracking method based on improved ellipticity provided by the present invention.

[0121] Specifically, the AA-GGIW-PMBM algorithm and the improved algorithm are used to track the extended targets in the monitoring area, and the MATLAB simulator is used for comprehensive simulation. For all scenarios, the moving targets follow the linear Gaussian approximate constant velocity (CV) model, and the measurement model is also a linear Gaussian model.

[0122] (1) The tracking performance of the proposed AA-GGIW-PMBM filter and other tracking schemes is evaluated using the Optimal Sub-Pattern Assignment (OSPA) distance. The smaller the index, the better. OSPA can be described as:

[0123]

[0124] Where X = {x1, x2, …, x m} represents the true state set of the observed target, Represents the estimated state set of the target after algorithm processing. X and Respectively represent the true state vector and estimated state vector of an independent target at a certain observation time. n Represents all permutations and combinations of m elements from the sequence set {1,2,...,n}. The number of combinations is It means finding the pairing with the smallest distance error from all the combined pairings of estimated tracks and true tracks. is the estimated error distance of the track state. To prevent the excessive distance error in the indicator from affecting the associated error, take:

[0125]

[0126] Among them, p≥1 is the order parameter and c is the penalty parameter. is a single target state x and The benchmark metric.

[0127] The dynamic equation of the extended target is in, represents the motion state vector of the i-th extended target at time k, F k is the state transfer matrix, ω k The covariance is Q k The process Gaussian white noise. The measurement equation of the extended target is in is the jth measurement at time k, H k is the measurement matrix, The covariance is R k The measurement of Gaussian white noise.

[0128] In the simulation experiment of the algorithm proposed in the present invention, the parameters are set as follows: the target detection probability is P D =0.9, the target survival probability is P S = 0.99, the target Poisson rate is randomly sampled in the interval γ∈[7,9], and the clutter Poisson rate is λ = 60. State transfer matrix Process noise covariance Sampling interval T = 1. Measurement matrix

[0129] In the algorithm implementation, multi-Bernoulli (MB) components with weights lower than 0.01 are pruned, and all Bernoulli estimates with existence probability greater than 0.5 are selected from the MB component with the largest MB weight, and their average vector is calculated to obtain the estimate of the target set.

[0130] In a distributed sensor network with 9 sensors, the range of the experimental scene is set to [-60m, 60m]×[-60m, 60m]. 100 time steps are simulated for 4 extended targets in the two-dimensional monitoring area. The following table describes the initial state and birth and death time of the extended targets in the scene:

[0131] The target's initial state and birth and death times

[0132]

[0133] The actual trajectories and shape ranges of the four targets are as follows: Figure 3 As shown, the position of the sensor is marked with a triangle; the dotted line represents that information can be exchanged between sensors; the solid line represents the center of mass trajectory of the target; and the ellipse represents the shape range of the target, which is 8 standard deviations in size.

[0134] Run the Monte Carlo simulation experiment 100 times and get the average simulation result as follows Figure 4-5 shown.

[0135] Depend on Figure 4 The results in show that the proposed algorithm significantly reduces the estimation error of the number of extended targets. When the targets are at the intersection moment or in the adjacent area, it can effectively alleviate the situation of missed estimation of the number of targets, achieve a more accurate estimation of the number of targets, and improve the tracking performance of extended targets.

[0136] Depend on Figure 5 From the results of the OSPA error variation over time, we can see that the error of the improved algorithm is significantly smaller in the target intersection stage (40s-53s) and the target adjacent stage (82s-98s), which shows that the proposed algorithm can better improve the missed detection problem caused by the proximity of the target. Overall, the OSPA error of the proposed algorithm is at a lower level, which shows that the proposed algorithm has a better tracking effect.

[0137] Example 2

[0138] Embodiment 2 of the present invention provides a multi-sensor multi-extended target tracking system based on improved ellipticity, including:

[0139] A target detection module, wherein the target detection module is used to obtain elliptical extended targets of multiple sensors, and use a GGIW-PMBM algorithm to predict and update the elliptical extended targets to obtain posterior density information of multiple sensors;

[0140] A flooding communication module, wherein the flooding communication module is used to iteratively update the posterior density information of multiple sensors using a distributed flooding algorithm, and to fuse the iteratively updated posterior density information to complete flooding communication;

[0141] An association module, wherein the association module is used to associate Bernoulli components of the same elliptical extended target using Log-Cholesky metric after completing flooding communication to form an associated Bernoulli subset;

[0142] A state extraction module is used to fuse the associated Bernoulli subsets to obtain a fused ellipse extended target, perform state extraction on the fused ellipse extended target, and return a state estimation value.

[0143] Example 3

[0144] Figure 6 A schematic diagram of the structure of an electronic device provided in Embodiment 2 of the present invention, such as Figure 6 As shown, the electronic device includes a processor 21, a memory 22, an input device 23 and an output device 24; the number of processors 21 in the computer device can be one or more. Figure 6 A processor 21 is taken as an example; the processor 21, the memory 22, the input device 23 and the output device 24 in the electronic device can be connected by a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0145] The memory 22 is a computer-readable storage medium that can be used to store software programs, computer executable programs, and modules. The processor 21 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 22, that is, the multi-sensor multi-extended target tracking method based on improved ellipticity of embodiment 1 is implemented.

[0146] The memory 22 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 22 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 22 may further include a memory remotely arranged relative to the processor 21, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0147] The input device 23 can be used to receive the ID and password input by the user, etc. The output device 24 is used to output the network configuration page.

[0148] Example 4

[0149] Embodiment 4 of the present invention further provides a computer-readable storage medium, wherein the computer-executable instructions, when executed by a computer processor, are used to implement the multi-sensor multi-extended target tracking method based on improved ellipticity as provided in Embodiment 1.

[0150] An embodiment of the present invention provides a storage medium containing computer executable instructions, and its computer executable instructions are not limited to the method operations provided in Example 1, and can also execute related operations in the multi-sensor multi-extended target tracking method based on improved ellipticity provided in any embodiment of the present invention.

[0151] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-sensor multi-extended target tracking method based on improved ellipticity, characterized in that: The steps include: Step S1, obtaining elliptical extended targets of multiple sensors, using the GGIW-PMBM algorithm to predict and update the elliptical extended targets, and obtaining posterior density information of multiple sensors; Step S2: using a distributed flooding algorithm to iteratively update the posterior density information of multiple sensors, and fusing the iteratively updated posterior density information to complete flooding communication; Step S3, after completing the flooding communication, using Log-Cholesky metric to associate the Bernoulli components of the same elliptical extended target to form an associated Bernoulli subset; Step S4, fusing the associated Bernoulli subsets to obtain a fused ellipse extended target, extracting the state of the fused ellipse extended target, and returning a state estimation value; Wherein, in step S3, associating the Bernoulli components of the same elliptical extended target using Log-Cholesky metric includes the following steps: Obtain the shape matrices of the two elliptical extended targets, use the Cholesky algorithm to inversely decompose the shape matrices, and generate two inverse decomposition matrices; The Log-Cholesky metric is used to calculate the Log-Cholesky distance of two inverse decomposition matrices in the triangular matrix space, and the Log-Cholesky distance of the shape matrix is ​​obtained; Calculate the centroid Euclidean distance of the shape matrices of the two elliptical extended targets, and combine the centroid Euclidean distance and the shape matrix Log-Cholesky distance to obtain the distance between the two elliptical extended targets; The distance between the two elliptical extended targets is compared with a threshold value. If the distance between the two elliptical extended targets is less than the threshold value, the Bernoulli components of the two elliptical extended targets are associated.

2. The multi-sensor multi-extended target tracking method based on improved ellipticity according to claim 1 is characterized in that: The step S1 in which the GGIW-PMBM algorithm is used to predict and update the elliptical extended target comprises the following steps: Step S11, arbitrarily selecting two elliptical extended targets, and obtaining the extended target states of the two elliptical extended targets at time k; wherein the extended target states include Poisson measurement rate, motion state vector and shape matrix; Step S12: calculating the center-of-mass Euclidean distance of the two elliptical extended objects based on the motion state vector; if the center-of-mass Euclidean distance is less than or equal to a distance threshold, the two elliptical extended objects are in a neighboring area; Step S13, calculating the number of targets at time k, comparing the number of targets at time k with the number of targets at time k-1, if the number of targets decreases and the reduced elliptical expanded targets are located in the adjacent area, compensating for the lost targets in the adjacent area.

3. The multi-sensor multi-extended target tracking method based on improved ellipticity according to claim 1, characterized in that: The step S2 of using a distributed flooding algorithm to communicate the posterior density information of multiple sensors includes the following steps: The sensor serving as the fusion center is denoted as sensor s; Record the set of sensors adjacent to the sensor s as a neighbor sensor set, and select any sensor from the neighbor sensor set as a neighbor sensor i; The iterative information received by sensor s from neighbor sensor i at the tth iteration is obtained, the iterative information of sensor s at the tth iteration is fused with the iterative information from neighbor sensor i, and the fused iterative information is used as the iterative information of sensor s at the t+1th iteration for the next iteration.

4. The multi-sensor multi-extended target tracking method based on improved ellipticity according to claim 3 is characterized in that: The fusion of the iterative information of sensor s at iteration t and the iterative information of neighbor sensor i at iteration t includes: In the above formula, is the iteration information at sensor s at the tth iteration, x is the target motion state information, is the iterative information received by sensor s from neighbor sensor i at the tth iteration, S s is the set of neighbor sensors, p i (x) is the posterior density of neighbor sensor i.

5. The multi-sensor multi-extended target tracking method based on improved ellipticity according to claim 1, characterized in that: The step S2 of fusing the posterior density information after communication includes: In the above formula, is the information set at sensor s after the iteration ends, Represents the information set at sensor s after the iteration The set of sensors involved, p i (x) is the posterior density of sensor i, and x is the target motion state information.

6. The multi-sensor multi-extended target tracking method based on improved ellipticity according to claim 1, characterized in that: Using the Log-Cholesky metric to calculate the Log-Cholesky distance between two inverse decomposition matrices in the triangular matrix space includes: In the above formula, X and represents the shape matrix of two elliptical extension targets, P and Q represent the two inverse decomposition matrices of the shape matrix of two elliptical extension targets, Indicates L + The distance function of the P and Q matrices in space, Indicated in The distance function of the shape matrices of two ellipsoidally extended targets in space, It is the lower triangular part in the strict sense; represents the diagonal part of the matrix, ||·|| F Represents the 2-norm of the matrix.

7. A multi-sensor multi-extended target tracking system based on improved ellipticity, characterized in that: include: A target detection module, wherein the target detection module is used to obtain elliptical extended targets of multiple sensors, and use a GGIW-PMBM algorithm to predict and update the elliptical extended targets to obtain posterior density information of multiple sensors; A flooding communication module, wherein the flooding communication module is used to iteratively update the posterior density information of multiple sensors using a distributed flooding algorithm, and to fuse the iteratively updated posterior density information to complete flooding communication; An association module, wherein the association module is used to associate Bernoulli components of the same elliptical extended target using Log-Cholesky metric after completing flooding communication to form an associated Bernoulli subset; A state extraction module, which is used to fuse the associated Bernoulli subsets to obtain a fused ellipse extended target, perform state extraction on the fused ellipse extended target, and return a state estimation value; Wherein, the association module includes: An inverse decomposition unit is used to obtain the shape matrices of the two elliptical extended targets, and to perform inverse decomposition on the shape matrices using the Cholesky algorithm to generate two inverse decomposition matrices; A Log-Cholesky distance calculation unit, used to calculate the Log-Cholesky distance of two inverse decomposition matrices in a triangular matrix space using the Log-Cholesky metric to obtain the Log-Cholesky distance of the shape matrix; An elliptical extended target calculation unit, used to calculate the centroid Euclidean distance of the shape matrices of two elliptical extended targets, and to obtain the distance between the two elliptical extended targets by combining the centroid Euclidean distance and the shape matrix Log-Cholesky distance; The associating unit is used to compare the distance between the two elliptical extension targets with a threshold value, and if the distance between the two elliptical extension targets is less than the threshold value, the Bernoulli components of the two elliptical extension targets are associated.

8. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the multi-sensor multi-extended target tracking method based on improved ellipticity as claimed in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the multi-sensor multi-extended target tracking method based on improved ellipticity as described in any one of claims 1 to 6 is implemented.

Images