Method for improving complex scene multi-radar track association through adaptive pre-association technology

Abstract

The invention relates to a method for improving complex scene multi-radar track association through an adaptive pre-association technology. The method comprises the following steps: constructing a track pair matrix, a threshold screening matrix and an association cost matrix; calculating a deviation distance and a pre-association distance threshold of the first track pair according to track information in the first track pair, and if the deviation distance and the pre-association distance threshold meet a preset screening judgment condition, setting an element corresponding to the first track pair in a threshold screening matrix as a passing mark value; calculating an association cost value of the second track pair based on the track information of the second track pair, and setting an element corresponding to the second track pair in the association cost matrix as the association cost value; and generating a track association result based on the association cost matrix in combination with an optimal allocation algorithm. By adopting the method, track sets from different radar nodes can be effectively processed, the method adapts to the condition of multi-radar cooperative work in a complex scene, and powerful support is provided for realizing multi-radar target tracking and recognition.

Description

Technical Field

[0001] The present invention belongs to the field of radar signal processing, and particularly relates to a method for improving multi-radar track association in complex scenarios by an adaptive pre-association technique. Background Art

[0002] In a distributed information processing system, multi-sensor track fusion aims to improve the positioning accuracy of a target by integrating complementary target information from multiple sensors. This process includes two main steps: track association and track fusion state estimation. The task of track association is to match the tracks of the same target from different sensors, and then perform track fusion and filtering output. Essentially, track association is a process of removing duplicates for the multi-sensor repeated tracking of a target. As a key step in multi-sensor track fusion, the effect of track association directly affects the accuracy and stability of subsequent target state estimation. Therefore, ensuring the accuracy and reliability of track association is the basis for improving the overall performance of the fusion system.

[0003] Currently, the existing main track association methods can be roughly divided into three categories: probability and statistics-based, mathematical model-based, and artificial intelligence-based. The first category of methods is based on probability and statistics, and performs track association by calculating the state difference between target tracks, such as the nearest neighbor (NN) method and the weighted track association method. These methods have simple principles and are easy to implement, but in scenarios with dense targets or crossing tracks, it is extremely easy to have problems of incorrect track association and low association accuracy. The second category of methods comprehensively considers the measurement uncertainty of target motion characteristics and uses fuzzy mathematics methods or likelihood functions to compare the similarity of tracks. Such as the track association method based on fuzzy clustering and the fuzzy double-threshold track association method. However, such methods involve relatively high computational complexity and are likely to affect the efficiency of track association in real-time scenarios. The third category of methods uses neural network and deep learning methods for target track association. However, such methods require a large number of matched track pairs as training data to build a mature intelligent model, and it is usually difficult to collect sufficient training data for a specific radar system. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method for improving multi-radar track association in complex scenarios by an adaptive pre-association technique that can reduce computational complexity and improve the efficiency of track association on the premise of ensuring relatively high association accuracy.

[0005] The present application provides a method for improving multi-radar track association in complex scenarios by an adaptive pre-association technique, including:

[0006] Obtain the initialized first-node track set and second-node track set, and construct a track pair matrix, a threshold screening matrix, and an association cost matrix based on the first-node track set and the second-node track set; wherein, the first-node track set is the track set obtained based on the first radar, and the second-node track set is the track set obtained based on the second radar;

[0007] Calculate the deviation distance and the pre-association distance threshold of the first track pair according to the track information in the first track pair. If the deviation distance and the pre-association distance threshold meet the preset screening determination conditions, set the element corresponding to the first track pair in the threshold screening matrix to the passing mark value; wherein, the first track pair is the track pair in the track pair matrix;

[0008] Calculate the association cost value of the second track pair based on the track information of the second track pair, and set the element corresponding to the second track pair in the association cost matrix to the association cost value; wherein, the second track pair is the track pair corresponding to the element set to the passing mark value in the threshold screening matrix;

[0009] Based on the association cost matrix and combined with the optimal assignment algorithm, globally optimize the assignment of the tracks in the first-node track set and the second-node track set to generate a track association result.

[0010] The above adaptive pre-association technology for improving multi-radar track association in complex scenarios can effectively process track sets from different radar nodes, adapt to the situation of multi-radar collaborative work in complex scenarios, and provide strong support for realizing multi-radar target tracking and recognition.

[0011] Specifically, the above adaptive pre-association technology for improving multi-radar track association in complex scenarios can concentrate on processing track pairs with potential association value by screening based on the deviation distance and the pre-association distance threshold, reduce unnecessary computational complexity, and improve data processing efficiency; by comprehensively considering various factors to calculate the association cost value and construct the association cost matrix, the track association result is more accurate, and the probability of mis-association and missed association can be effectively reduced; by globally optimizing the assignment of tracks, the association relationship between all track pairs can be comprehensively considered, so as to obtain a more reasonable overall track association result and improve the reliability and overall performance of the system. Brief Description of the Drawings

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0013] Figure 1Schematic diagram of the application environment of a method for improving multi-radar track association in complex scenarios using an adaptive pre-association technique provided by an embodiment of the present application;

[0014] Figure 2 Flowchart of the steps of a method for improving multi-radar track association in complex scenarios using an adaptive pre-association technique provided by an embodiment of the present application;

[0015] Figure 3 Flowchart of the steps of another method for improving multi-radar track association in complex scenarios using an adaptive pre-association technique provided by an embodiment of the present application;

[0016] Figure 4 Schematic diagram of the observation of the same target by two sensors provided by an embodiment of the present application;

[0017] Figure 5 Algorithm flowchart of a maximum likelihood estimation global optimal track association algorithm introducing a pre-association threshold provided by an embodiment of the present application;

[0018] Figure 6 Schematic diagram of the curve of the association correct rate of three algorithms changing with the number of time steps provided by an embodiment of the present application;

[0019] Figure 7 Schematic diagram of the curve of the association correct rate of three algorithms changing with the initial spacing provided by an embodiment of the present application;

[0020] Figure 8 Schematic diagram of the curve of the running time of three algorithms changing with the number of targets provided by an embodiment of the present application;

[0021] Figure 9 Schematic diagram of the structure of a device for improving multi-radar track association in complex scenarios using an adaptive pre-association technique provided by an embodiment of the present application. Detailed implementation manners

[0022] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0023] The method for improving multi-radar track association in complex scenarios using the adaptive pre-association technique provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 . Among them, the computing device 101 can communicate with the first radar 102 and the second radar 103 through a communication channel. The data storage system can store the data that the computing device 101 needs to process. The data storage system can be integrated on the computing device 101, or placed in the cloud or other network servers.

[0024] Schematically, the first radar 102 and the second radar 103 can obtain the track information of the target 104 and send the track information of the target 104 to the computing device 101 through the communication channel. The computing device 101 can match the track information of the target 104 from the first radar 102 and the second radar 103, and then perform track fusion and filtering output. Among them, the computing device 101 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, servers, and Internet of Things devices. The server can be an independent server or a server cluster composed of multiple servers.

[0025] In an exemplary embodiment, as Figure 2 shown, a method for improving multi-radar track association in complex scenarios using an adaptive pre-association technique is provided. Taking the method applied to the Figure 1 computing device 101 as an example, it includes the following steps S201 to S204. Among them:

[0026] Step S201, obtain the initialized first node track set and the second node track set, and construct a track pair matrix, a threshold screening matrix, and an association cost matrix based on the first node track set and the second node track set.

[0027] Specifically, the computing device 101 can obtain the initialized first node track set and the second node track set at time t τ , and can construct a track pair matrix, a threshold screening matrix, and an association cost matrix based on the first node track set and the second node track set. Among them, the first node track set can be the track set obtained based on the first radar, and the second node track set can be the track set obtained based on the second radar.

[0028] Optionally, the computing device 101 can traverse the first node track set collected based on the first radar and the second node track set collected based on the second radar in accordance with the first-in first-out principle of the queue, construct the first track pair and build a track pair matrix. Among them, is the i-th track in the first node track set, is the j-th track in the second node track set, N1 is the number of tracks in the first node track set, and N2 is the number of tracks in the second node track set.

[0029] Optionally, the computing device 101 may initialize a threshold screening matrix GatePass with a size of N1·N2, and initialize the values of all elements in the threshold screening matrix GatePass to 0; the computing device 101 may initialize an association cost matrix C with a size of N1·N2, and set the values of all elements in the association cost matrix C to ∞.

[0030] Step S202: Calculate the deviation distance and the pre-association distance threshold of the first track pair according to the track information in the first track pair. If the deviation distance and the pre-association distance threshold meet the preset screening determination condition, set the element corresponding to the first track pair in the threshold screening matrix to the passing mark value.

[0031] Specifically, the computing device 101 may calculate the deviation distance and the pre-association distance threshold of the first track pair according to the track information in the first track pair. If the deviation distance and the pre-association distance threshold meet the preset screening determination condition, set the element corresponding to the first track pair in the threshold screening matrix to the passing mark value. Herein, the first track pair is the track pair in the track pair matrix.

[0032] Optionally, the computing device 101 may calculate the first track pair according to the ranging r, the course θ information of the track, and the position information of the first node and the second node τ the deviation distance |ΔX(t τ )| and the pre-association distance threshold λ Π ·σ X (t τ ) of the two tracks in τ . If the preferred preset screening determination condition: "|ΔX(t Π )|≤λ X (t τ )" is satisfied, the computing device 101 may set the element GatePass[i][j] corresponding to the first track pair in the threshold screening matrix GatePass to the passing mark value.

[0033] Optionally, the passing mark value may be a non-zero value. Preferably, the passing mark value may be 1.

[0034] Step S203: Calculate the association cost value of the second track pair based on the track information of the second track pair, and set the element corresponding to the second track pair in the association cost matrix to the association cost value.

[0035] Specifically, the computing device 101 may calculate the correlation cost value of the second track pair based on the track information of the second track pair, and set the element corresponding to the second track pair in the correlation cost matrix to the correlation cost value. Among them, the second track pair is the track pair corresponding to the element set to the passing mark value in the threshold screening matrix.

[0036] Optionally, the computing device 101 may traverse the threshold screening matrix GatePass. If the element GatePass[i][j] corresponding to the first track pair in the threshold screening matrix GatePass is the passing mark value, the computing device 101 may calculate the correlation cost c of the second track pair ij , and update the corresponding element C[i][j] of the correlation cost matrix C.

[0037] Step S204: Based on the correlation cost matrix and in combination with the optimal assignment algorithm, perform global optimal assignment on the tracks in the first node track set and the second node track set to generate a track association result.

[0038] Optionally, the computing device 101 may complete the global optimal matching of the tracks between the first node corresponding to the first radar and the second node corresponding to the second radar at time t τ based on the correlation cost matrix C and the optimal assignment algorithm, and generate a track association result.

[0039] Optionally, the optimal assignment algorithm may be but is not limited to the Auction Algorithm (AA), the Hungarian Algorithm (HA), and the Jonker-Volgenant-Castanon (JVC) algorithm.

[0040] In the above method for improving multi-radar track association in complex scenarios by the adaptive pre-association technology, by obtaining the initialized first-node track set and second-node track set, and constructing a track pair matrix, a threshold screening matrix, and an association cost matrix, the track data from different radars can be structurally processed, enabling subsequent calculations and analyses to be carried out based on an ordered and standardized data form, thereby improving the efficiency and accuracy of data processing; by calculating the deviation distance and pre-association distance threshold based on the track information in the track pair, and screening according to the preset screening and determination conditions, irrelevant track pairs can be effectively excluded, reducing the computational amount of subsequent processing and improving the accuracy of track association; by calculating the association cost value based on the track pairs remaining after threshold screening, the correlation degree between these track pairs can be accurately measured; by using the association cost matrix and combining with the optimal assignment algorithm to globally optimize the assignment of tracks in the two-node track sets, the optimal track matching combination can be found as a whole, avoiding the problem of local optimal solutions, and thus improving the overall performance of track association, enabling the tracks obtained by different radars to more accurately correspond to the same target.

[0041] In one alternative embodiment, please refer to Figure 3 , the track information of the second track pair includes target ranging information, target angle measurement information, and target speed measurement information. Calculating the association cost value of the second track pair based on the track information of the second track pair includes:

[0042] Step S306, calculate the likelihood estimates of the target ranging information, target angle measurement information, and target speed measurement information to generate a ranging likelihood estimate value, an angle measurement likelihood estimate value, and a speed likelihood estimate value.

[0043] Step S307, obtain the weights of the corresponding ranging likelihood estimate value, angle measurement likelihood estimate value, and speed likelihood estimate value based on the variances of the target ranging information, target angle measurement information, and target speed measurement information to generate a ranging likelihood weight, an angle measurement likelihood weight, and a speed likelihood weight.

[0044] Step S308, calculate the association cost value of the second track pair according to the ranging likelihood estimate value, angle measurement likelihood estimate value, speed likelihood estimate value, ranging likelihood weight, angle measurement likelihood weight, and speed likelihood weight.

[0045] In the above method for improving multi-radar track association in complex scenarios using the adaptive pre-association technology, by calculating the likelihood estimates of the target ranging information, target angle measurement information, and target velocity measurement information, generating the ranging likelihood estimate value, angle measurement likelihood estimate value, and velocity likelihood estimate value, the correlation between tracks can be judged more accurately; by calculating the weights of the corresponding likelihood estimates using the variances of the target ranging information, target angle measurement information, and target velocity measurement information, generating the ranging likelihood weight, angle measurement likelihood weight, and velocity likelihood weight, reasonable weighting of information in different dimensions can be performed, and thus the correlation degree between tracks can be reflected more comprehensively and accurately.

[0046] In one optional embodiment, the expression of the association cost value is:

[0047]

[0048] c m =-ln(γ m )

[0049] In the formula, c ij is the association cost value between the i-th track in the first node track set and the j-th track in the second node track set, m is the number of modalities of the track information of the second track pair, a m is the weight of the m-th modality, c m is the association cost component of the m-th modality, a1 is the ranging likelihood weight, a2 is the angle measurement likelihood weight, a3 is the velocity likelihood weight, γ m is the likelihood estimate value of the m-th modality, γ1 is the ranging likelihood estimate value, γ2 is the angle measurement likelihood estimate value, γ3 is the velocity likelihood estimate value; among them, the first modality of the track information of the second track pair is the target ranging information, the second modality of the track information of the second track pair is the target angle measurement information, and the third modality of the track information of the second track pair is the target velocity measurement information.

[0050] In one optional embodiment, the expressions of the ranging likelihood estimate value, angle measurement likelihood estimate value, velocity likelihood estimate value, ranging likelihood weight, angle measurement likelihood weight, and velocity likelihood weight can be:

[0051]

[0052] In the formula, is the variance of the target ranging information, is the variance of the target angle measurement information, is the variance of the target velocity measurement information, n is the number of historical traces of the i-th track in the first node track set and the j-th track in the second node track set, r i (k) and r j(k) is the target ranging information of the k-th historical trace point of the i-th trace in the first node trace set and the j-th trace in the second node trace set, θ i (k) and θ j (k) are the target angle measurement information of the k-th historical trace point of the i-th trace in the first node trace set and the j-th trace in the second node trace set, v i (k) and v j (k) are the target speed measurement information of the k-th historical trace point of the i-th trace in the first node trace set and the j-th trace in the second node trace set.

[0053] In one alternative embodiment, please refer to Figure 3 . The trace information in the first trace pair includes target ranging information and target angle measurement information. Calculating the deviation distance and pre-association distance threshold of the first trace pair based on the trace information in the first trace pair includes:

[0054] Step S302, calculating the position error, ranging error, and angle measurement error of the first trace pair based on the target ranging information and target angle measurement information.

[0055] Step S303, calculating the position error variance of the position error based on the ranging error and angle measurement error.

[0056] Step S304, obtaining the deviation distance and pre-association distance threshold based on the position error and position error variance.

[0057] Preferably, the expressions for the deviation distance and pre-association distance threshold can be:

[0058] D Δ (t τ ) = |ΔW(t τ )|

[0059]

[0060] In the formula, D Δ (t τ ) is the deviation distance at time t τ , |ΔW(t τ )| is the ranging error at time t τ , Π D (t τ ) is the pre-association distance threshold, λ ∏ is the Gaussian probability distribution coefficient, σ W (t τ ) is the standard deviation of the position error at time t τ , is the position error variance at time t τ .

[0061] In the method for improving multi-radar track association in complex scenarios using the above-mentioned adaptive pre-association technology, by calculating the deviation distance and the pre-association distance threshold to preliminarily screen track pairs, irrelevant track pairs can be quickly excluded from a large number of track pairs, significantly reducing the computational amount and improving the efficiency of track association. Furthermore, the system can respond and process target information more quickly.

[0062] Furthermore, in the method for improving multi-radar track association in complex scenarios using the above-mentioned adaptive pre-association technology, by setting an adjustable parameter, the Gaussian probability distribution coefficient λ, in the pre-association distance threshold Π , the threshold can be flexibly adjusted according to different application scenarios and requirements.

[0063] In one alternative embodiment, the position error can be the error of the position in any coordinate axis direction in the coordinate system of the track.

[0064] Optionally, the expressions for the position error and the position error variance can be:

[0065]

[0066] In the formula, ΔW(t τ ) is the position error at time t τ , ΔX(t τ ) is the position error in the X-axis direction at time t τ , ΔY(t τ ) is the position error in the Y-axis direction at time t τ , represents that the position error is the error of the position in the X coordinate axis direction in the coordinate system of the track, represents that the position error is the error of the position in the Y coordinate axis direction in the coordinate system of the track, is the position error variance at time t τ , is the variance of the position error in the X-axis direction at time t τ , is the variance of the position error in the Y-axis direction at time t τ .

[0067] In one alternative embodiment, the screening determination condition is that the deviation distance is less than or equal to the pre-association distance threshold.

[0068] Optionally, the Gaussian probability distribution coefficient λ Π can be 3. At this time, according to the Gaussian distribution function, the probability that the screening determination condition: "|ΔX(t τ )|≤λ Π ·σ X (t τ )" is true is 99.74%.

[0069] In one optional embodiment, the position error includes a first position error and a second position error, and the position error variance includes a first position error variance of the first position error and a second position error variance of the second position error; wherein, the first position error is the error of the position in the direction of the first coordinate axis in the coordinate system of the track, and the second position error is the error of the position in the direction of the second coordinate axis in the coordinate system of the track;

[0070] The calculation formulas for the first position error variance and the second position error variance are as follows:

[0071]

[0072] In the formula, is the first position error variance at time t τ , is the second position error variance at time t τ , is the variance of the position error in the X-axis direction at time t τ , is the variance of the position error in the Y-axis direction at time t τ , r i (t τ ) and r j (t τ ) are the target ranging information of the i-th track in the first node track set and the j-th track in the second node track set at time t τ respectively, θ i (t τ ) and θ j (t τ ) are the target angle measurement information of the i-th track in the first node track set and the j-th track in the second node track set at time t τ respectively, Δr i (t τ ) and Δr j (t τ ) are the ranging errors of the i-th track in the first node track set and the j-th track in the second node track set at time t τ respectively, Δθ i (t τ ) and Δθ j (t τ ) are the angle measurement errors of the i-th track in the first node track set and the j-th track in the second node track set at time t τ .

[0073] In one alternative embodiment, the screening determination condition is that the first deviation distance is less than or equal to the first pre - association distance threshold and the second deviation distance is less than or equal to the second pre - association distance threshold;

[0074] Wherein, the first deviation distance is the deviation distance obtained based on the first position error, the second deviation distance is the deviation distance obtained based on the second position error, the first pre - association distance threshold is the pre - association distance threshold obtained based on the first position error variance, and the second pre - association distance threshold is the pre - association distance threshold obtained based on the second position error variance.

[0075] Optionally, the screening determination condition can be: " and ", wherein, |ΔW1(t τ )| is the first deviation distance obtained based on the first position error corresponding to the position error in the X - axis direction, |ΔW2(t τ )| is the second deviation distance obtained based on the second position error corresponding to the position error in the Y - axis direction, is the pre - association distance threshold obtained based on the first position error variance of the first position error, is the pre - association distance threshold obtained based on the second position error variance of the second position error.

[0076] Optionally, the Gaussian probability distribution coefficients λ Π,X and λ Π,Y can both be set to 3.

[0077] In one alternative embodiment, the optimal assignment algorithm is the JVC optimal assignment algorithm.

[0078] Illustratively, the Jonker - Volgenant - Castanon (JVC) algorithm is an efficient linear assignment algorithm based on network flow and duality theory, which can be used to solve the minimum - cost assignment problem in bipartite graph matching. The JVC algorithm has a high time - complexity performance and can find the optimal solution in a relatively short time. Especially for large - scale assignment problems, its efficiency advantage is more obvious.

[0079] In an exemplary embodiment, as Figure 3 shown, a method for improving multi - radar track association in complex scenarios using an adaptive pre - association technique is provided, including the following steps S301 to step S310. Wherein:

[0080] Step S301, obtain the initialized first - node track set and second - node track set, and construct a track - pair matrix, a threshold - screening matrix, and an association - cost matrix based on the first - node track set and the second - node track set.

[0081] Step S302: Calculate the position error, ranging error, and angle measurement error of the first track pair based on the target ranging information and target angle measurement information.

[0082] Step S303: Calculate the position error variance of the position error based on the ranging error and angle measurement error. Step S304: Obtain the deviation distance and pre - associated distance threshold based on the position error and position error variance.

[0083] Step S305: If the deviation distance and pre - associated distance threshold meet the preset screening and determination conditions, set the element corresponding to the first track pair in the threshold screening matrix to the passing mark value.

[0084] Step S306: Calculate the likelihood estimates of the target ranging information, target angle measurement information, and target velocity information, and generate a ranging likelihood estimate value, an angle measurement likelihood estimate value, and a velocity likelihood estimate value.

[0085] Step S307: Obtain the weights of the corresponding ranging likelihood estimate value, angle measurement likelihood estimate value, and velocity likelihood estimate value based on the variances of the target ranging information, target angle measurement information, and target velocity information, and generate a ranging likelihood weight, an angle measurement likelihood weight, and a velocity likelihood weight.

[0086] Step S308: Calculate the association cost value of the second track pair according to the ranging likelihood estimate value, angle measurement likelihood estimate value, velocity likelihood estimate value, ranging likelihood weight, angle measurement likelihood weight, and velocity likelihood weight.

[0087] Step S309: Set the element corresponding to the second track pair in the association cost matrix to the association cost value.

[0088] Step S310: Based on the association cost matrix and in combination with the optimal assignment algorithm, perform global optimal assignment on the tracks in the first - node track set and the second - node track set to generate a track association result.

[0089] In the above - mentioned method for improving multi - radar track association in complex scenarios by the adaptive pre - association technology, through the setting of the passing mark value in the threshold screening matrix, track pairs that do not meet the preset conditions can be quickly excluded, and track pairs with potential association value can be concentrated for processing, thereby reducing unnecessary computational complexity and improving data processing efficiency; by using the optimal assignment algorithm to perform global optimal assignment on the tracks, the association relationships between all track pairs can be comprehensively considered, so that a more reasonable overall track association result can be obtained, and further improve the overall performance and reliability of the system.

[0090] To further illustrate the solution of the embodiments of the present application, a specific example is given below for explanation.

[0091] 1 Algorithm implementation

[0092] 1.1 Pre - associated Threshold Design

[0093] In the Cartesian coordinate system, the measurement model of the track of a certain target by sensor S at time conforms to:

[0094]

[0095] Among them, represents the true value of the range measurement of sensor S with respect to the target, represents the true value of the angle measurement of sensor S with respect to the target, Δr s,j (t τ ) and Δθ s,j (t τ ) respectively represent the range measurement error and the angle measurement error, and the errors follow a Gaussian distribution, that is: and

[0096] Assume that in the global coordinate system, the coordinates of sensor S are Through equation (2), the local polar coordinates of the track at time can be converted into global coordinates:

[0097]

[0098] As Figure 4 shown, at time t τ , sensor 1 and sensor 2 observe the same target, generating corresponding local tracks track i and track j , and converting the position observations into the global coordinate system. Due to the existence of process noise and measurement noise, in the global coordinate system, there is a deviation in the position observation information between sensor 1 and sensor 2, and the corresponding position deviation can be expressed as:

[0099]

[0100] Among them, ΔX i,j (t τ ) and ΔY i,j (t τ ) respectively represent the position deviations of the two tracks at time t τ in the X - direction and Y - direction.

[0101] Equation (1) can be substituted into equation (3) to further obtain the position deviation expression containing process error and measurement error:

[0102]

[0103] Further performing a first - order Taylor expansion, the approximate value expression of the position deviation of the two tracks in the X - direction at time t τ can be obtained:

[0104]

[0105] Since the true value of the target is uncertain, in practical applications, the measured value of the sensor is substituted into Equation (5).

[0106] In the case where the measurement errors all follow a normal distribution, the linear summation based on the measurement errors still follows a normal distribution, where The calculation is shown in Equation (6):

[0107]

[0108]

[0109] Similarly, the expression of the position deviation in the Y direction can be obtained, as shown in Equation (7):

[0110]

[0111] According to the Gaussian distribution function, the probability that |ΔX(t τ )| ≤ 3σ X (t τ ) is 99.74%. Therefore, 3σ X (t τ ) and 3σ Y (t τ ) are used as the distance thresholds at a single moment. When |ΔX(t τ )| ≤ 3σ x (t τ ) and |ΔY(t τ )| ≤ 3σ Y (t τ ), it is determined that the screening through the pre-association threshold is passed and enters the global optimal allocation.

[0112] 1.2 JVC Global Optimal Allocation Based on Maximum Likelihood Estimation

[0113] For the track association pairs that pass the threshold screening, their subsequent association process can be regarded as a typical two-dimensional allocation problem. Classical optimal allocation algorithms include the auction algorithm, the Kuhn–Munkres (Munkres) algorithm, and the JVC (Jonker–Volgenant–Castanon, JVC) algorithm. Kadar et al. compared the performance and computational efficiency of these three algorithms. In terms of the correct rate of allocation, the Munkres algorithm and the JVC algorithm are significantly better than the auction algorithm, while in terms of computational complexity, the JVC algorithm has higher computational efficiency. Based on the above discussion, the JVC algorithm can be selected for global optimal allocation.

[0114] In an alternative embodiment, the optimal assignment of track pairs can be achieved by minimizing the association cost based on the idea of global optimality to achieve the association effect. For the optimal assignment of tracks from two sensors at a certain moment, the whole process can be modeled as:

[0115]

[0116] where δ ij represents the matching state of two tracks and . δ ij = 1 indicates that and are successfully matched, that is, and belong to the same target. The track of sensor 1 can be associated with at most one track in sensor 2, and vice versa.

[0117]

[0118] The global optimal matching is based on the principle of minimizing the association cost. Therefore, it is necessary to construct a suitable association cost matrix. In an alternative embodiment of the present invention, the association cost of each association pair passing through the threshold can be calculated based on the maximum likelihood estimation. Assuming that the measurement errors of the three kinematic parameter features of range r, heading θ, and speed v of the sensor for the target are independent of each other, the likelihood values of these three features between the association pairs can be calculated, as shown in Equation (10):

[0119]

[0120] where n represents the number of historical measurements of the track and at the current moment. The larger n is, the more historical information of the track is obtained, and the more accurate the association is, but the computational complexity will also increase accordingly.

[0121] Let c m = -ln(γ m ), where m = 1, 2, 3. The final association cost can be expressed as the weighted sum of c m , that is:

[0122]

[0123] where a m is the weighted coefficient of the likelihood value of each motion feature and can be calculated through their respective measurement variances, as shown in Equation (12):

[0124]

[0125] Finally, the implementation steps of the MLE global optimal track association algorithm introducing the pre-association threshold are as follows. The overall framework of the algorithm is asFigure 5 As shown in the figure, it includes:

[0126] Step 1: Initialize t τ The track set at time node 1 And the track set of node 2 According to the first-in, first-out principle of the queue, traverse the two sets to construct track pairs Enter the decision of the pre-association threshold; initialize the threshold screening matrix GatePass with a size of N1·N2, and set the values of all elements to 0; initialize the association cost matrix C with a size of N1·N2, and set the values of all elements to ∞.

[0127] Step 2: According to the ranging r, course θ information of the track and the position information of the node, calculate the distance |ΔX(t τ )| between the two tracks at time t τ ) and the pre-association distance threshold 3σ X (t τ ). If |ΔX(t τ )| ≤ 3σ X (t τ ), set GatePass[i][j] to 1, repeat Step 1 to Step 2, continue to construct associated track pairs and perform threshold decisions until all track points in the two track lists are screened.

[0128] Step 3: Traverse the threshold screening matrix GatePass. If GatePass[i][j] is 1, calculate the association cost c ij of the track association pair, and update the corresponding element C[i][j] of the association cost matrix.

[0129] Step 4: Based on the association cost matrix C and the JVC optimal assignment algorithm, complete the global optimal matching of the tracks of the two nodes at time t τ , and the algorithm ends.

[0130] 2 Simulation Experiments

[0131] The following will conduct a comparative analysis of the global optimal association algorithm without using a threshold and the global optimal association algorithm with a pre-association threshold proposed in the optional embodiment of the present invention in terms of association accuracy and computational efficiency through simulation experiments. The optional embodiment of the present invention can compare the performance of the following three algorithms: the NN global optimal track association algorithm, the MLE global optimal track association algorithm without a threshold, and the MLE global optimal track association algorithm with a pre-association threshold. The experimental indicators can include the association accuracy P and the average running time of each experiment. In order to reduce the contingency of the experiment, 100 Monte Carlo simulations can be used. The performance of each algorithm can be comprehensively evaluated by comparing the average association accuracy P and the average running time of each experiment.

[0132] 2.1 Experimental Environment

[0133] The initial target parameter settings are shown in Table 1. Initially, the targets are arranged in a matrix with a target interval of 20 m, and then they move in a straight line along a random direction.

[0134] Table 1 Initial Setup of Simulation Target Parameters

[0135] Tab.1 Initial Setup of Simulation Target Parameters

[0136]

[0137] 2.2 Experimental Results and Analysis

[0138] 2.2.1 Analysis of the Variation of the Correct Association Rate with Time

[0139] Figure 6 The curves showing the variation of the association correct rate with time for three algorithms (the NN global optimal algorithm without threshold, the MLE global optimal algorithm without threshold, and the MLE global optimal algorithm with a pre-association threshold) when the initial spacing is 20 m are presented. It can be seen from the figure that the association correct rates of the three algorithms all show an increasing trend over time.

[0140] For the NN global optimal algorithm without threshold (marked with red circles), the association correct rate tends to be stable after 25 s. This phenomenon may be due to the fact that the distance between the initial targets is relatively close in the initial stage, resulting in difficult association; as time progresses, the distance between the targets gradually increases, the observation conditions are improved, and thus the association performance is enhanced. After that, the distance between the targets further increases, and the distance between the targets is no longer the bottleneck of the performance, so the association correct rate curve tends to be stable; the overall trends of the association correct rate curves of the MLE global optimal algorithm without threshold (marked with blue squares) and the MLE global optimal algorithm with a pre-association threshold (marked with green triangles) are similar. Since the MLE algorithm requires a certain amount of historical data for accurate parameter estimation, in the initial stage, due to the threshold condition of the MLE global optimal algorithm with a pre-association threshold, it better filters out incorrect matches, and thus is slightly superior to the MLE algorithm without threshold in terms of association accuracy.

[0141] To comprehensively evaluate the association performance of the three algorithms, referring to the measurement criteria in the literature, the average correct association rate Fc and the average incorrect association rate Fe of the three algorithms when the initial spacing is 20 m are further compared, and the results are shown in Table 2.

[0142] Table 2 Comparison of the Association Performance of the Three Algorithms

[0143] Table 2 Comparison of Association Performance for Three Algorithms

[0144]

[0145] According to the results shown in Table 2, it can be seen that compared with the NN global optimal algorithm without threshold, the MLE global optimal algorithm with pre-association threshold has a 6.22% increase in the correct association rate and a 9.37% decrease in the false association rate; compared with the MLE global optimal algorithm, the gaps in various indicators are relatively small, and the gaps in the correct association rate and false association rate are within 2%.

[0146] 2.2.2 Analysis of the Variation of the Correct Association Rate with the Initial Spacing

[0147] Figure 7 The curves showing the variation of the association correct rates of the three algorithms with the initial target spacing are presented. It can be seen from the figure that the association correct rates of the three algorithms gradually increase as the initial target spacing increases, indicating that as the target spacing expands, the observation conditions are improved. Specifically, as the initial spacing increases, the distance between targets gradually widens, the distinguishability of the system for targets is enhanced, and thus the association correct rate is effectively improved. Especially in an environment with sparse targets, all three algorithms exhibit excellent association performance, and the association correct rates are all stable above 90%.

[0148] Further analyzing the performance of each algorithm, combined with Figure 7 the experimental results, it can be found that the NN global optimal algorithm without threshold is greatly affected by the target spacing. In an environment with dense targets (initial spacing less than 30 meters), the association correct rate of this algorithm is relatively low, and in this scenario, the stability of the algorithm is poor, showing large fluctuations. This may be due to the fact that the target spacing is too small, making it difficult for the algorithm to effectively distinguish different targets, increasing the probability of false associations; in contrast, the MLE global optimal algorithm with pre-association threshold performs more stably. Especially in the dense target scenario, its association correct rate can be maintained above 90%, approaching the MLE global optimal algorithm without threshold.

[0149] 2.2.3 Analysis of the Algorithm Running Efficiency

[0150] The average single-run time consumption of the three algorithms under different numbers of targets was compared, and the initial target spacing was still maintained at 20m. The results are as Figure 8 shown. Since all three algorithms need to traverse all the track data from two nodes, construct candidate track association pairs and calculate the track association cost, their time complexity is at least O(N2). Therefore, as the number of targets increases, the running time consumption also increases.

[0151] The NN global optimal algorithm without threshold only needs to calculate the relatively simple Euclidean distance on candidate association pairs, so it can still maintain a relatively good operation efficiency when the number of targets increases; the MLE global optimal algorithm without threshold calculates the likelihood function including exponential and logarithmic operations for all track association pairs. As the number of targets increases, the amount of this operation rises sharply, resulting in a significant increase in operation time. The MLE global optimal algorithm with pre-association threshold introduces a screening mechanism for track pre-association, screens out a large number of unmatched candidates in the pre-association stage, significantly reduces the repeated calculation of the likelihood function, and thus can still control the operation time below 100 ms in scenarios with a large number of targets. Compared with the MLE global optimal algorithm without threshold, when there are 50 targets, its operation efficiency is increased by about 70%.

[0152] Based on the above analysis, the MLE global optimal track association algorithm with pre-association threshold shows high association accuracy and good calculation efficiency in scenarios with different target densities and numbers of targets. By establishing the likelihood function of track association pairs, it can effectively alleviate the association difficulties caused by small target spacing and make up for the problem of the decrease in association accuracy rate that is prone to occur in the NN global optimal algorithm in dense target scenarios. At the same time, it introduces a screening mechanism for track pre-association, reduces the repeated calculation of the likelihood function in large-scale target scenarios, reduces the operation complexity, and effectively alleviates the problem of low calculation efficiency caused by the increase in the number of targets. Overall, the MLE global optimal algorithm with pre-association threshold can provide association performance comparable to that of the MLE global optimal algorithm in dense target situations and still maintain a high calculation efficiency when the number of targets increases, demonstrating good effectiveness and robustness.

[0153] In the above method for improving multi-radar track association in complex scenarios with adaptive pre-association technology, aiming at the track association problem in multi-sensor systems, the association ambiguity problem caused by measurement errors and the similarity of target motion parameters in dense target and trajectory intersection scenarios is mainly analyzed. An algorithm combining a threshold and the maximum likelihood estimation (MLE) method is proposed to improve the efficiency of track association while ensuring a high association accuracy. This algorithm calculates the threshold based on the Euclidean distance through statistical methods and pre-screens all track pairs. For the screened track pairs, the association cost is calculated based on the likelihood function, and the final target track pairs are determined through the optimal assignment algorithm.

[0154] The results of simulation experiments show that in scenarios with dense targets and crossing trajectories, the algorithm proposed in this paper can effectively solve the problem of association ambiguity. While ensuring a high association accuracy, it significantly improves the computational efficiency. This algorithm is applicable to the track association scenarios with dense targets and crossing trajectories, with a small amount of calculation and a fast running speed, making it suitable for applications in engineering practice. Future work will combine more features to address the multi-sensor track association problem in complex clutter environments.

[0155] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps does not have a strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or in turn with at least a part of the steps or stages in other steps or other steps.

[0156] Based on the same inventive concept, the embodiments of the present application also provide an apparatus for implementing the adaptive pre-association technology for enhancing multi-radar track association in complex scenarios, which is used to implement the method for enhancing multi-radar track association in complex scenarios by the adaptive pre-association technology. The solution provided by this apparatus for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the apparatus for implementing the adaptive pre-association technology for enhancing multi-radar track association in complex scenarios provided below can refer to the limitations on the method for enhancing multi-radar track association in complex scenarios by the adaptive pre-association technology in the above text, and will not be repeated here.

[0157] In an exemplary embodiment, as Figure 9 shown, a 900 apparatus is provided, including:

[0158] A track information initialization module 901, which can be used to obtain the initialized first-node track set and second-node track set, and construct a track pair matrix, a threshold screening matrix, and an association cost matrix based on the first-node track set and the second-node track set.

[0159] A pre-association distance threshold determination module 902, which can be used to calculate the deviation distance and the pre-association distance threshold of the first track pair according to the track information in the first track pair. If the deviation distance and the pre-association distance threshold meet the preset screening and determination conditions, set the element corresponding to the first track pair in the threshold screening matrix to the passing mark value.

[0160] The associated cost value calculation module 903 can be used to calculate the associated cost value of the second track pair based on the track information of the second track pair, and set the element corresponding to the second track pair in the associated cost matrix to the associated cost value.

[0161] The global optimal allocation module 904 can be used to perform global optimal allocation on the tracks in the first node track set and the second node track set based on the associated cost matrix and in combination with the optimal allocation algorithm, and generate a track association result.

[0162] In the device embodiment, in addition to the embodiment of the device independent claim, embodiments of the device items corresponding one by one to all the method dependent claims need to be written.

[0163] In an optional embodiment, the associated cost value calculation module 903 can also be used for:

[0164] Calculate the likelihood estimates of the target ranging information, target angle measurement information, and target speed measurement information, and generate a ranging likelihood estimate value, an angle measurement likelihood estimate value, and a speed likelihood estimate value;

[0165] Based on the variances of the target ranging information, target angle measurement information, and target speed measurement information, obtain the weights of the corresponding ranging likelihood estimate value, angle measurement likelihood estimate value, and speed likelihood estimate value, and generate a ranging likelihood weight, an angle measurement likelihood weight, and a speed likelihood weight;

[0166] Calculate the associated cost value of the second track pair according to the ranging likelihood estimate value, angle measurement likelihood estimate value, speed likelihood estimate value, ranging likelihood weight, angle measurement likelihood weight, and speed likelihood weight.

[0167] In an optional embodiment, the pre-associated distance threshold determination module 902 can also be used for:

[0168] Calculate the position error, ranging error, and angle measurement error of the first track pair based on the target ranging information and target angle measurement information;

[0169] Calculate the position error variance of the position error based on the ranging error and angle measurement error;

[0170] Obtain the deviation distance and the pre-associated distance threshold based on the position error and the position error variance.

[0171] In an embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of a method for improving multi-radar track association in complex scenarios by an adaptive pre-association technology as described above are implemented.

[0172] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0173] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial descriptions of the method embodiments. The device embodiments described above are only illustrative. The components described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0174] The above embodiments only represent several implementation manners of the embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the embodiments of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several deformations and improvements can still be made, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A method for improving the correlation of multiple radar tracks in complex scenes using adaptive pre-correlation technology, characterized in that: The method comprises: Acquire an initialized first node track set and a second node track set, and construct a track pair matrix, a threshold screening matrix, and an associated cost matrix based on the first node track set and the second node track set; wherein the first node track set is a track set acquired based on the first radar, and the second node track set is a track set acquired based on the second radar; Calculate the deviation distance and pre-association distance threshold of the first track pair according to the track information in the first track pair, and if the deviation distance and the pre-association distance threshold meet the preset screening judgment condition, set the element corresponding to the first track pair in the threshold screening matrix as a pass mark value; wherein the first track pair is a track pair in the track pair matrix; Calculating the associated cost value of the second track pair based on the track information of the second track pair, and setting the element corresponding to the second track pair in the associated cost matrix to the associated cost value; wherein the second track pair is the track pair corresponding to the element set as the pass mark value in the threshold screening matrix; Based on the association cost matrix and in combination with an optimal allocation algorithm, a global optimal allocation is performed on the tracks in the first node track set and the second node track set to generate a track association result.

2. The method according to claim 1, characterized in that The track information of the second track pair includes target distance measurement information, target angle measurement information, and target speed measurement information. The calculating of the associated cost value of the second track pair based on the track information of the second track pair includes: Calculating likelihood estimates of the target distance measurement information, the target angle measurement information, and the target speed measurement information to generate a distance measurement likelihood estimate value, an angle measurement likelihood estimate value, and a speed likelihood estimate value; Based on the variance of the target distance measurement information, the target angle measurement information and the target speed measurement information, obtain the corresponding weights of the distance measurement likelihood estimation value, the angle measurement likelihood estimation value and the speed likelihood estimation value, and generate the distance measurement likelihood weight, the angle measurement likelihood weight and the speed likelihood weight; The associated cost value of the second track pair is calculated according to the ranging likelihood estimate value, the angle measurement likelihood estimate value, the speed likelihood estimate value, the ranging likelihood weight, the angle measurement likelihood weight and the speed likelihood weight.

3. The method according to claim 2, characterized in that The expression of the associated cost value is: c m =-ln(γ m ) In the formula, c ij is the association cost value between the i-th track in the first node track set and the j-th track in the second node track set, m is the modal number of the track information of the second track pair, a m is the weight of the mth mode, c m is the associated cost component of the mth mode, a1 is the distance measurement likelihood weight, a2 is the angle measurement likelihood weight, a3 is the speed likelihood weight, γ m is the likelihood estimate of the mth mode, γ1 is the distance measurement likelihood estimate, γ2 is the angle measurement likelihood estimate, and γ3 is the speed likelihood estimate; wherein the first mode of the track information of the second track pair is the target distance measurement information, the second mode of the track information of the second track pair is the target angle measurement information, and the third mode of the track information of the second track pair is the target speed measurement information.

4. The method according to claim 3, characterized in that The expressions of the distance measurement likelihood estimate value, the angle measurement likelihood estimate value, the speed likelihood estimate value, the distance measurement likelihood weight, the angle measurement likelihood weight and the speed likelihood weight are: In the formula, is the variance of the target ranging information, is the variance of the target angle measurement information, is the variance of the target speed measurement information, n is the number of historical track points of the i-th track in the first node track set and the j-th track in the second node track set, r i (k) and r j (k) are the target ranging information of the kth historical point track of the i-th track in the first node track set and the j-th track in the second node track set, respectively, i (k) and θ j (k) the target angle measurement information of the kth historical point track of the i-th track in the first node track set and the j-th track in the second node track set, respectively; v i (k) and v j (k) are the target speed measurement information of the kth historical point track of the ith track in the first node track set and the jth track in the second node track set.

5. The method according to claim 1, characterized in that The track information in the first track pair includes target distance measurement information and target angle measurement information, and calculating the deviation distance and the pre-association distance threshold of the first track pair according to the track information in the first track pair includes: Calculate the position error, the distance measurement error and the angle measurement error of the first track pair based on the target distance measurement information and the target angle measurement information; Calculate a position error variance of the position error based on the distance measurement error and the angle measurement error; Calculate the deviation distance and the pre-association distance threshold based on the position error and the position error variance; The expressions of the deviation distance and the pre-association distance threshold are: D Δ (t τ )=|ΔW(t τ )| Where D Δ (t τ ) is t τ The deviation distance at time, |ΔW(t τ ) is t τ The ranging error at time, Π D (t τ ) is the pre-association distance threshold, λ Π is the Gaussian probability distribution coefficient, σ W (t τ ) is t τ The standard deviation of the position error at time instant, t τ The position error variance at time.

6. The method according to claim 5, characterized in that: The position error is the error of the position in any coordinate axis direction in the coordinate system of the track; The expressions of the position error and the position error variance are: In the formula, ΔW(t τ ) is t τ The position error at time, ΔX(t τ ) is t τ The position error in the X-axis direction at the time, ΔY(t τ ) is t τ The position error in the Y-axis direction at the moment, The position error represents the position error in the X-axis direction of the coordinate system of the track, The position error represents the position error in the Y-axis direction of the coordinate system of the track, t τ The position error variance at time, t τ The variance of the position error in the X-axis direction at the moment, t τ The variance of the position error in the Y-axis direction at the time.

7. The method according to claim 6, characterized in that: The screening determination condition is that the deviation distance is less than or equal to the pre-association distance threshold.

8. The method according to claim 5, characterized in that: The position error includes a first position error and a second position error, and the position error variance includes a first position error variance of the first position error and a second position error variance of the second position error; wherein the first position error is an error of a position in a first coordinate axis direction in a coordinate system of the track, and the second position error is an error of a position in a second coordinate axis direction in a coordinate system of the track; The calculation formulas of the first position error variance and the second position error variance are: In the formula, t τ The first position error variance at time instant, t τ The second position error variance at time instant, t τ The variance of the position error in the X-axis direction at the moment, t τ The variance of the position error in the Y-axis direction at the moment, r i (t τ ) and r j (t τ ) are respectively the i-th track in the first node track set and the j-th track in the second node track set at t τ The target distance information at time, θ i (t τ ) and θ j (t τ ) are respectively the i-th track in the first node track set and the j-th track in the second node track set at t τ The target angle measurement information at the time, Δr i (t τ ) and Δr j (t τ ) are respectively the i-th track in the first node track set and the j-th track in the second node track set at t τ The ranging error at time, Δθ i (t τ ) and Δθ j (t τ ) are respectively the i-th track in the first node track set and the j-th track in the second node track set at t τ The angle measurement error at time.

9. The method according to claim 8, characterized in that: The screening determination condition is that the first deviation distance is less than or equal to the first pre-association distance threshold and the second deviation distance is less than or equal to the second pre-association distance threshold; Among them, the first deviation distance is the deviation distance obtained based on the first position error, the second deviation distance is the deviation distance obtained based on the second position error, the first pre-association distance threshold is the pre-association distance threshold obtained based on the first position error variance, and the second pre-association distance threshold is the pre-association distance threshold obtained based on the second position error variance.

10. The method according to any one of claims 1 to 9, characterized in that: The optimal allocation algorithm is the JVC optimal allocation algorithm.

Images