Radar multi-target tracking method, device and equipment based on Mama network and medium

Abstract

The invention discloses a radar multi-target tracking method and device based on a Mama network, equipment and a medium. The method comprises the following steps: for each target, performing matrix splicing on state information of the target at a preset historical moment to obtain target state information at all moments; calculating the Euclidean distance between the target state information and all the measurement data at the current moment; generating a target matrix of the target trace point features according to the Euclidean distance; inputting the target matrix into a pre-constructed Mama network model to obtain an association probability of the target and measurement; and performing filtering estimation processing on the association probability by using a Kalman filter to obtain a state prediction value of the target at the current moment. According to the embodiment of the invention, the Mama network is combined with the data association, and the historical state information features are fused to predict the association probability between the target and measurement, so that the efficiency of solving the data association problem in multi-target tracking by using a deep learning method is improved, and prior information such as a target motion model and clutter density does not need to be obtained in advance.

Description

Technical Field

[0001] The present application belongs to the field of radar technology, and in particular, relates to a radar multi-target tracking method, device, equipment and medium based on a Mamba network. Background Art

[0002] Data association technology is the basic part of multi-target tracking. Only by correctly matching the relationship between the measurement and the target source can accurate multi-target tracking be achieved. However, when using the traditional airborne radar multi-target tracking data association algorithm, it is necessary to obtain prior information such as the target motion model and clutter density in advance. Summary of the invention

[0003] The embodiments of the present application provide a radar multi-target tracking method, apparatus, device and medium based on a Mamba network, which are used to at least solve the problem in the related art that prior information such as target motion model and clutter density needs to be obtained in advance.

[0004] In a first aspect, an embodiment of the present application provides a radar multi-target tracking method based on a Mamba network, comprising:

[0005] For each target, matrix splicing is performed on the status information of its preset historical moments to obtain the target status information at all moments;

[0006] Calculating the Euclidean distance between the target state information and all the measured data at the current moment;

[0007] Generating a target matrix of target point trace features according to the Euclidean distance;

[0008] Input the target matrix into the pre-built Mamba network model to obtain the association probability between the target and the measurement;

[0009] The Kalman filter is used to perform filtering and estimation processing on the associated probability to obtain a state prediction value of the target at the current moment.

[0010] In a second aspect, an embodiment of the present application provides a radar multi-target tracking device based on a Mamba network, comprising:

[0011] A splicing module is used to perform matrix splicing on the state information of each target at a preset historical moment to obtain the target state information at all moments;

[0012] A calculation module, used for calculating the Euclidean distance between the target state information and all the measurement data at the current moment;

[0013] A generating module, used for generating a target matrix of target point trace features according to the Euclidean distance;

[0014] An association module, used to input the target matrix into a pre-built Mamba network model to obtain the association probability between the target and the measurement;

[0015] The filtering module is used to perform filtering and estimation processing on the associated probability using a Kalman filter to obtain a state prediction value of the target at the current moment.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the radar multi-target tracking method based on the Mamba network as described in any embodiment of the first aspect are implemented.

[0017] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the steps of the radar multi-target tracking method based on the Mamba network as described in any embodiment of the first aspect are implemented.

[0018] In a fifth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the Mamba network-based radar multi-target tracking method provided in the first aspect of the embodiment of the present application.

[0019] The radar multi-target tracking method, device, equipment and medium based on the Mamba network of the embodiment of the present application combine the Mamba network model with data association, and integrate the historical state information characteristics to predict the association probability between the target and the measurement, which greatly improves the efficiency of using deep learning methods to solve the data association problem in multi-target tracking, thereby inputting the association probability into the Kalman filter to predict and update the state of each target at the current moment, without the need to obtain prior information such as the target motion model and clutter density in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 It is a flowchart of a radar multi-target tracking method based on a Mamba network provided in an embodiment of the present application;

[0022] Figure 2 is a schematic diagram of using the Mamba network model to predict the association probability between a target and a measurement provided by an embodiment of the present application;

[0023] Figure 3 The expected number of clutter provided in the embodiment of the present application is E λ =80, the detection probability is p d =0.99 simulation trajectory and measurement graph;

[0024] Figure 4A The expected number of clutter provided in the embodiment of the present application is E λ =80, the detection probability is p d Schematic diagram of the tracking results of the HA algorithm when =0.99;

[0025] Figure 4B The expected number of clutter provided in the embodiment of the present application is E λ =80, the detection probability is p d Schematic diagram of the tracking results of the JPDA algorithm when =0.99;

[0026] Figure 4C The expected number of clutter provided in the embodiment of the present application is E λ =80, the detection probability is p d Schematic diagram of the tracking results of the Bi-LSTM algorithm when =0.99;

[0027] Figure 4D The expected number of clutter provided in the embodiment of the present application is E λ =80, the detection probability is p d Schematic diagram of the tracking results of the Mamba algorithm when =0.99;

[0028] Figure 5 It is a schematic diagram of the OSPA distance comparison of the HA algorithm, JPDA algorithm, Bi-LSTM algorithm, and Mamba algorithm provided in the embodiment of the present application;

[0029] Figure 6 It is a structural schematic diagram of a radar multi-target tracking device based on a Mamba network provided in an embodiment of the present application;

[0030] Figure 7 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application.

[0031] Reference numerals:

[0032] Radar multi-target tracking device based on Mamba network 600, splicing module 601, calculation module 602, generation module 603, association module 604, filtering module 605,

[0033] Electronic device 700 , processor 701 , memory 702 , communication interface 703 , bus 710 . DETAILED DESCRIPTION

[0034] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0035] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0036] Data association technology is the basic part of multi-target tracking. Only by correctly matching the relationship between the measurement and the target source can accurate multi-target tracking be achieved. The traditional multi-target tracking data association algorithm can be traced back to the Hungarian algorithm proposed by the Hungarian mathematician Kuhn. It is an algorithm for solving the maximum weight matching problem of bipartite graphs. In addition, some scholars have proposed using multiple hypothesis tracking (MHT) in data association problems. The MHT algorithm is suitable for solving data association problems in sparse clutter environments. For data association problems in dense clutter environments, the probabilistic data association (PDA) algorithm based on Bayesian probability theory converts the association problem between targets and measurements into a computational problem of probability distribution. In recent years, many researchers have applied deep learning methods to radar multi-target tracking problems. Studies have shown that LSTM-based models have better association performance than classic models such as JPDA and HA.

[0037] However, when using the traditional airborne radar multi-target tracking data association algorithm, it is necessary to obtain prior information such as the target motion model and clutter density in advance.

[0038] In order to solve the problems of the related art, the embodiments of the present application provide a radar multi-target tracking method, device, equipment and medium based on a Mamba network.

[0039] The following is a detailed description of the radar multi-target tracking method based on the Mamba network provided by the embodiment of the present application through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0040] Figure 1 FIG. 2 is a flow chart showing a radar multi-target tracking method based on a Mamba network according to an embodiment of the present application. Figure 1 As shown, the radar multi-target tracking method based on the Mamba network may specifically include the following steps:

[0041] S101, for each target, performing matrix splicing on the status information of its preset historical moments to obtain the target status information at all moments;

[0042] S102, calculating the Euclidean distance between the target state information and all the measurement data at the current moment;

[0043] S103, generating a target matrix of target point trace features according to the Euclidean distance;

[0044] S104, inputting the target matrix into a pre-built Mamba network model to obtain the association probability between the target and the measurement;

[0045] S105: Perform filtering and estimation processing on the association probability using a Kalman filter to obtain a state prediction value of the target at the current moment.

[0046] Therefore, the Mamba network model is combined with data association, and the historical state information features are integrated to predict the association probability between the target and the measurement, which greatly improves the efficiency of using deep learning methods to solve the data association problem in multi-target tracking. The association probability is then input into the Kalman filter to predict and update the state of each target at the current moment, without the need to obtain prior information such as the target motion model and clutter density in advance.

[0047] The specific implementation methods of the above steps are introduced below.

[0048] It should be noted that in the embodiment of the present application, the data association problem is formulated as a classification task performed on as many targets as possible. It can be understood that the data association problem is essentially an optimization combination matching problem, and this problem can be regarded as a multi-classification problem when studied from the perspective of deep learning.

[0049] In the embodiments of the present application, the data association problem to be addressed must satisfy the following requirements: each measurement has at most one target source, and each target is associated with at most one measurement.

[0050] First, assuming that the number of tracked targets is n and the number of measurements acquired by the radar at the current time (time k) is m, the target state set X(k) and the measurement data set Z(k) at the current time (time k) are defined.

[0051] Specifically, in, represents the state of the i-th target at time k, n k Represents the number of targets at time k.

[0052] It should be noted that, since the sensor may miss detection or generate false alarm in each scan, the concept of virtual measurement is introduced, and each measurement data set Z(k) is defined to include virtual measurement.

[0053] Specifically, considering the possibility of sensor miss detection and / or false alarm in radar, the concept of virtual measurement is introduced. The measurement set is defined by adding a virtual measurement with index i=0 to each set Z(k), that is:

[0054] in, represents the state of the jth target at time k, Indicates virtual measurement, m k represents the number of measurements at time k.

[0055] In some embodiments, in S101, taking the preset historical moments as the first three historical moments as an example, matrix splicing is performed on the target motion states of the first three historical moments to obtain all target states:

[0056]

[0057] in, represents the state of the i-th target at time k, and T represents transposition.

[0058] In some embodiments, in S102, the Euclidean distance between the target state information and all the measurement data at the current moment is calculated according to the following formula:

[0059]

[0060] Among them, ρ represents the Euclidean distance, represents the state information of the jth target at the current k moment, x' i = indicates target status information.

[0061] In some embodiments, in S103, a target matrix of target point trace features is generated according to the following formula:

[0062] D i = reshape(ρ,1,M*SZ );

[0063] Among them, D i represents the target matrix; ρ represents the Euclidean distance between the state of the i-th target at all times and all measurements at time k; M represents the number of measurements; S Z Represents the measurement state dimension.

[0064] In this way, the dimension transformation of the matrix is ​​realized, so that the point trace features of each target become a one-dimensional matrix, that is, the target matrix of the target point trace features obtained is a one-dimensional matrix.

[0065] Furthermore, in some embodiments, in S104, the target matrix is ​​used as an input of the Mamba network framework, and the Mamba network framework is trained and outputs the association probability between each target and the measurement.

[0066] Specifically, the Mamba model is widely used in the fields of natural language processing and computer vision. The Mamba model is an innovative linear time series modeling method that cleverly combines the characteristics of recurrent neural networks (RNNs) and convolutional neural networks (CNNs) to solve the problem of computational efficiency when processing long sequences.

[0067] One of the core innovations of the Mamba model is the introduction of a selective mechanism, which is based on the traditional state space to form a selective state space, so that the model can dynamically adjust its behavior according to different inputs. In this way, Mamba can effectively filter out irrelevant information while retaining and strengthening task-related information, improving the ability to process long sequence data. In order to optimize computational efficiency, the second core innovation of the Mamba model is the use of hardware-aware algorithms, especially the use of the GPU's memory hierarchy to increase the computational speed of scanning operations and reduce memory requirements. The data interaction of the Mamba structure is mainly concentrated between the GPU and SRAM, and this part of the data interaction is fast. In view of the application of the selective mechanism and hardware-aware algorithms, the advantage of the Mamba network model for the problem of multi-target tracking data association is that it can realize convolution in computation, conduct parallel training, and has high efficiency. It also handles the problems of input addition and modification, random order, and gradient disappearance.

[0068] Optionally, the Mamba network model in the embodiment of the present application includes an input module, a MambaBlock module and an output module.

[0069] The input module includes a Mask layer and a fully connected layer. The input sequence is processed by the Mask layer. The main function of the Mask layer is to fill the variable-length measurement with zeros when the maximum number of measurements is fixed to reduce the complexity of the network. After feature extraction through the fully connected layer, it enters the MambaBlock module.

[0070] The MambaBlock module includes a convolutional layer and a selective state space model (SSM). In this network model, a total of 3 MambaBlock modules are iterated, each of which consists of a convolutional layer and a selective state space model (SSM). At the same time, the RMSNorm layer is added to standardize the activation of the neural network, which can help stabilize and speed up training.

[0071] The output module includes a multi-layer perceptron and a Softmax layer. A multi-layer perceptron is added to the back end of the MambaBlock module for category prediction, and a Softmax layer is added to convert the classification problem into a probability problem. The network model first associates the processed data with the probability of each target and each measurement in turn, and then obtains the association matrix between each target and all measurements, where 0 represents that the target is not associated with the measurement, and 1 represents that the target is associated with the measurement.

[0072] Optionally, make the following assumptions: first, the sensor may miss detections in each scan; second, the sensor may generate false alarms. This assumption means that the number of measurements in each scan is variable and not equal to the number of targets. Then, the Mamba network model is constructed by introducing decision variables that define data associations:

[0073]

[0074] in, represents the decision variable, Indicates that the measurement is associated with the target, Indicates that the measurement is not associated with the target.

[0075] As an optional embodiment, in order to make the Mamba network suitable for the case where the number of measurements is unknown, a loss function (MCD) combining masked cross entropy loss and overlap loss is proposed for model training.

[0076] Specifically, the MCD loss function is:

[0077] Loss MCD =Loss mask +Loss dice

[0078]

[0079] Among them, pi represents the predicted association probability; y i represents the true association probability; represents the predicted association probability between target i and measurement j; represents the true association probability between target i and measurement j; γ is a smoothing term, which takes a value of 1 to prevent loss overflow; m represents the number of measurements obtained by the radar at time k; n represents the number of targets; mask represents a vector composed of Mask layer markers.

[0080] In this way, when the number of measurements at time k does not reach the maximum number of measurements, the measurements at time k will be filled with 0 values, and 1 will be used to mark the real measurement, 0 will be used to mark the filled measurement, and then the marks will be combined into a Mask vector. In order to avoid the influence of the filled measurement on the loss calculation, an improvement is made on the basis of the cross entropy loss function to obtain Loss mask Since there are more clutter than real measurements of the target in multi-target scenarios, this will lead to sample imbalance. In order to push the model to pay more attention to the probability of association between the learning target and the correct measurement, the Dice loss is introduced. dice The final MCD loss function used is the sum of the above two losses, that is, the above formula.

[0081] thus, Figure 2 FIG. 2 shows a schematic diagram of using the Mamba network model to predict the association probability between a target and a measurement in an embodiment of the present application. Figure 2 As shown, the target matrix is ​​input into the pre-built Mamba network model, the Mask layer is used to fill the variable-length measurements with zeros when the maximum number of measurements is fixed, the fully connected layer is used to extract features, the multi-layer perceptron is used to perform category prediction to obtain classified data, and the Softmax layer is used to perform probability association between the target and each measurement in turn to obtain an association matrix between each target and all measurements.

[0082] In this way, the matching relationship between the target and the measurement can be learned from the training samples without prior information such as the target motion model and clutter density. The constructed and trained Mamba network can extract the motion feature information of the target and learn the data association between the trajectory and the measurement, thereby predicting the output classification matrix to obtain the association probability between the target and the measurement.

[0083] In other words, the data association problem is regarded as a time series problem. By constructing a Mamba network model, combining the Mamba model with data association, and integrating the historical state information characteristics to predict the association probability between the target and the measurement, the efficiency of using deep learning methods to solve the data association problem in multi-target tracking is greatly improved.

[0084] Furthermore, the associated probability is input into the Kalman filter to predict and update the state of each target at the current moment, that is, execute S105 to finally obtain the state prediction value of the target at the current moment.

[0085] In summary, the radar multi-target tracking method based on the Mamba network of the embodiment of the present application formulates the data association problem as a classification task performed on as many targets as possible; performs matrix splicing on the target motion state at the historical moment, and calculates the Euclidean distance between each target and the measurement; performs dimension transformation on the matrix, so that the point trace feature of each target becomes a one-dimensional matrix; uses the matrix as the input of the Mamba network framework, trains through the Mamba network framework and outputs the association probability of each target and the measurement; finally, inputs the association probability into the Kalman filter to predict and update the state of each target at the current moment. It is a data-driven deep learning network algorithm, which does not require prior knowledge of the target motion model and clutter density and other prior information; because a loss function combining masked cross entropy loss and overlap loss is used for model training, the network can be applied to the situation where the number of measurements is unknown; virtual measurement is introduced to take into account the missed detection and false alarm of the sensor.

[0086] Furthermore, a simulation experiment is used to evaluate the performance of the radar multi-target tracking method based on the Mamba network in the embodiment of the present application. The specific evaluation process and results are as follows.

[0087] Network parameters and evaluation indicators include:

[0088] The experiment was conducted based on the Ubuntu 20.04 system, using the deep learning framework Pytorch. The hardware configuration of the experiment includes: CPU is Intel (R) i5-10400F, GPU is GeForce RTX 3090Ti, memory is 16GB, and CUDA11.6 is used to call the GPU for training acceleration. The key parameter settings are as follows: the initial learning rate of the network is set to 0.001, the model is trained from scratch, the number of epoch iterations is 1000, and the batch size of the training is 256; the model dimension of the Mamba network is 16, the state expansion coefficient of the SSM module is 16, the local convolution factor is 4, and the expansion coefficient of the MambaBlock module is 2; the Adam optimizer is used for parameter update.

[0089] In the experiment, the present invention uses the optimal sub-pattern assignment (OSPA) distance index to evaluate the tracking performance of different algorithms. OSPA is used to measure the distance between the estimated state of the target and the true target state, which has better accuracy sensitivity for verifying the tracking algorithm. The true state set of all targets is defined as: X = {x1, x2, ..., x m}, where m is the number of true targets; the predicted state set of all targets is defined as: P = {p1, p2, ..., p n}, where n is the number of predicted targets, and these two sets belong to the true state and predicted state of the target at the same time. OSPA is calculated as follows:

[0090]

[0091] Wherein, s (1≤s≤∞) represents the distance sensitivity, which is used to adjust the distance error weight between the target real state set and the predicted state set; c (c>0) represents the association sensitivity parameter, which is used to adjust the quantity error weight between the target real state set and the predicted state set; Π n Represents all permutations and combinations of m elements taken from X.

[0092] The simulation test scenarios include:

[0093] A total of 1,200 trajectories of various motion modes were simulated, with every 6 trajectories forming a multi-target scene. A total of 200 multi-target scenes were simulated, of which 60% were used for training, 20% for verification, and 20% for testing.

[0094] By adding mean μ = 0 and variance σ to the trajectory data x =130m(σ y =130m) is converted to the spatial rectangular coordinate system to simulate the target measurement, and uniformly distributed clutter points are added to simulate the interference of the environment. The clutter obeys a uniform distribution in the target motion scene, and the number of clutter obeys a Poisson distribution with a density of λ. The expected definition of the number of clutter is:

[0095] E λ =λ·(x max -x min )·(y max -y min ), where x max and x min Represents the maximum and minimum values ​​of the x coordinate within the range of motion, and y max and max Respectively represent the maximum and minimum y-coordinate values ​​within the motion range.

[0096] Furthermore, in order to better display the target tracking results and compare the tracking performance of different algorithms, all measurements are uniformly displayed using the Cartesian coordinate system through coordinate transformation.

[0097] Figure 3 shows the expected clutter number E λ =80, detection probability p d =0.99 simulation trajectory and measurement diagram. Figure 3 In the scene shown, each color of the measurement point corresponds to a target of the same color. The six targets move in a straight line at a uniform speed in different directions within the observation range. The sampling interval is 1s, and the total number of samplings is 50. During the movement of these six targets, there are 7 cases of two-target intersections and one case of three-target intersection.

[0098] Figure 4A , Figure 4B , Figure 4C , Figure 4D E λ =80,p d =0.99, the tracking results of four different algorithms: HA algorithm, JPDA algorithm, Bi-LSTM algorithm, and Mamba algorithm. Figure 5 The OSPA distance comparison of these four algorithms is shown, where the parameters of OSPA are p=2, c=500. Before k=10, there is a situation where the OSPA distance of the Mamba algorithm is higher than that of the JPDA algorithm and the Bi-LSTM algorithm. This is because the measurements determined by different algorithms are different, which will affect the update of the covariance and the filter gain, and thus affect the tracking results. In the first target intersection, the mutual proximity of multiple targets produces a more complex data association problem. It can be qualitatively analyzed that the OSPA distance of the HA algorithm is the largest. After the target data association occurs, the OSPA distance of the JPDA algorithm and the Bi-LSTM algorithm shows an upward trend. The OSPA distance of the HA algorithm in both intersections will increase significantly. However, the OSPA distance of the Mamba algorithm of the embodiment of the present application is generally the smallest. By analyzing the OSPA distance under different detection probabilities, it can be obtained that the HA algorithm is greatly affected by the detection probability, and the Bi-LSTM algorithm is less affected by the detection probability. The OSPA distance of the Mamba algorithm of the embodiment of the present application is the smallest under different detection probabilities.

[0099] It can be seen that the Mamba algorithm of the embodiment of the present application has a smaller optimal sub-pattern assignment (Optimal Sub-Pattern Assignment, OSPA) distance error, and in the tracking experiments with different detection probabilities, the performance of the Mamba algorithm of the embodiment of the present application is better than the classic data association algorithm. In addition, the Mamba network constructed and trained by the embodiment of the present application has the following advantages: when multiple targets intersect, the Mamba network can solve the matching problem of multiple targets and multiple measurements; in a certain noise interference environment, the Mamba network has better tracking robustness; under different detection probability conditions, the OSPA distance of the Mamba algorithm of the embodiment of the present application is the smallest, so it can improve the tracking performance of multiple targets.

[0100] It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0101] Based on the same technical concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a radar multi-target tracking device 600 based on a Mamba network.

[0102] like Figure 6 As shown, the radar multi-target tracking device 600 based on the Mamba network may include:

[0103] The splicing module 601 is used to perform matrix splicing on the state information of each target at a preset historical moment to obtain the target state information at all moments;

[0104] A calculation module 602, used to calculate the Euclidean distance between the target state information and all the measurement data at the current moment;

[0105] A generating module 603 is used to generate a target matrix of target point trace features according to the Euclidean distance;

[0106] An association module 604 is used to input the target matrix into a pre-built Mamba network model to obtain an association probability between the target and the measurement;

[0107] The filtering module 605 is used to perform filtering and estimation processing on the association probability using a Kalman filter to obtain a state prediction value of the target at the current moment.

[0108] Optionally, the target matrix is ​​a one-dimensional matrix.

[0109] In some embodiments, the generating module 603 is specifically used to generate a target matrix of target point trace features according to the following formula:

[0110] D i = reshape(ρ,1,M*S Z );

[0111]

[0112] Among them, D i represents the target matrix; ρ represents the Euclidean distance; M represents the number of measurements; S Z Indicates the measurement state dimension; represents the state information of the jth target at the current k moment; x' i = indicates target status information.

[0113] Optionally, the Mamba network model includes an input module, a MambaBlock module and an output module, wherein the input module includes a Mask layer and a fully connected layer, the MambaBlock module includes a convolutional layer and a selective state space model, and the output module includes a multilayer perceptron and a Softmax layer.

[0114] In some embodiments, the association module 604 is specifically used to input the target matrix into a pre-built Mamba network model, perform zero padding on the variable-length measurement when the maximum number of measurements is fixed through the Mask layer, perform feature extraction through the fully connected layer, perform category prediction through the multi-layer perceptron to obtain classified data, and perform probability association of the target and each measurement on the classified data in turn through the Softmax layer to obtain an association matrix between each target and all measurements.

[0115] It should be noted that, in the embodiment of the present application, each measurement has at most one target source, and each target is associated with at most one measurement.

[0116] In some optional embodiments, the MCD loss function used for Mamba network model training is:

[0117] Loss MCD =Loss mask +Loss dice

[0118]

[0119] Among them, p i represents the predicted association probability; y i represents the true association probability; represents the predicted association probability between target i and measurement j; represents the true association probability between target i and measurement j; γ is a smoothing term, which takes a value of 1 to prevent loss overflow; m represents the number of measurements obtained by the radar at time k; n represents the number of targets; mask represents a vector composed of Mask layer markers.

[0120] It should be noted that, for the convenience of description, the above device is described in various modules according to their functions. Of course, when implementing this application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0121] The device of the above embodiment is used to implement the corresponding radar multi-target tracking method based on Mamba network in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0122] Based on the same technical concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides an electronic device.

[0123] Figure 7 A more specific schematic diagram of the hardware structure of an electronic device provided by this embodiment is shown.

[0124] The electronic device 700 may include a processor 701 and a memory 702 storing computer program instructions.

[0125] Specifically, the processor 701 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0126] The memory 702 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 702 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 702 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 702 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 702 is a non-volatile solid-state memory.

[0127] In certain embodiments, the memory may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present application.

[0128] The processor 701 reads and executes the computer program instructions stored in the memory 702 to implement any one of the radar multi-target tracking methods based on the Mamba network in the above embodiments.

[0129] In some examples, the electronic device 700 may further include a communication interface 703 and a bus 710. Figure 7 As shown, the processor 701, the memory 702, and the communication interface 703 are connected via a bus 710 and communicate with each other.

[0130] The communication interface 703 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0131] Bus 710 includes hardware, software or both, and the parts of online data flow billing equipment are coupled to each other. For example, but not limitation, bus 710 may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front-end bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 710 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the present application considers any suitable bus or interconnection.

[0132] Exemplarily, the electronic device 700 may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA).

[0133] Based on the same technical concept, corresponding to any of the above-mentioned embodiments, the present application also provides a non-transitory computer-readable storage medium. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any of the radar multi-target tracking methods based on the Mamba network in the above-mentioned embodiments is implemented. Examples of computer-readable storage media include non-transitory computer-readable storage media, such as portable disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, etc.

[0134] Based on the same technical concept, corresponding to any of the above-mentioned embodiments, the present application also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer so that the computer and / or the processor execute the Mamba network-based radar multi-target tracking method. Corresponding to the execution subject corresponding to each step in each embodiment of the Mamba network-based radar multi-target tracking method, the processor that executes the corresponding step may belong to the corresponding execution subject.

[0135] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0136] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0137] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0138] The above reference is according to the method of the embodiment of the present application, the flow chart of the device (system) and the computer program product and / or the block diagram described various aspects of the present application.It should be understood that each square box in the flow chart and / or the block diagram and the combination of each square box in the flow chart and / or the block diagram can be realized by computer program instructions.These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the realization of the function / action specified in one or more square boxes of the flow chart and / or the block diagram.Such a processor can be but is not limited to a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit.It can also be understood that each square box in the block diagram and / or the flow chart and the combination of the square boxes in the block diagram and / or the flow chart can also be realized by the dedicated hardware that performs the specified function or action, or can be realized by the combination of dedicated hardware and computer instructions.

[0139] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A radar multi-target tracking method based on Mamba network, characterized in that: include: For each target, matrix splicing is performed on the status information of its preset historical moments to obtain the target status information at all moments; Calculating the Euclidean distance between the target state information and all the measured data at the current moment; Generating a target matrix of target point trace features according to the Euclidean distance; Input the target matrix into the pre-built Mamba network model to obtain the association probability between the target and the measurement; The Kalman filter is used to perform filtering and estimation processing on the associated probability to obtain a state prediction value of the target at the current moment.

2. The method according to claim 1, characterized in that The target matrix is ​​a one-dimensional matrix; The step of generating a target matrix of target point trace features according to the Euclidean distance includes: Generate the target matrix of the target point feature according to the following formula: D i =reshape(ρ,1,M*S Z ); Among them, D i represents the target matrix; ρ represents the Euclidean distance; M represents the number of measurements; S Z Indicates the measurement state dimension; represents the state information of the jth target at the current k moment; x' i Indicates target status information.

3. The method according to claim 1, characterized in that The Mamba network model includes an input module, a MambaBlock module and an output module, wherein the input module includes a Mask layer and a fully connected layer, the MambaBlock module includes a convolutional layer and a selective state space model, and the output module includes a multilayer perceptron and a Softmax layer.

4. The method according to claim 3, characterized in that The step of inputting the target matrix into a pre-built Mamba network model to obtain the association probability between the target and the measurement includes: The target matrix is ​​input into the pre-built Mamba network model, the variable-length measurement is padded with zeros when the maximum number of measurements is fixed through the Mask layer, feature extraction is performed through the fully connected layer, and category prediction is performed through the multi-layer perceptron to obtain classified data, and the classified data is probabilistically associated with the target and each measurement in turn through the Softmax layer to obtain the association matrix between each target and all measurements.

5. The method according to claim 1, characterized in that Each measurement has at most one target source, and each target is associated with at most one measurement.

6. The method according to claim 1, characterized in that The MCD loss function used for Mamba network model training is: Loss MCD =Loss mask +Loss dice Among them, p i represents the predicted association probability; y i represents the true association probability; represents the predicted association probability between target i and measurement j; represents the true association probability between target i and measurement j; γ is a smoothing term, which takes a value of 1 to prevent loss overflow; m represents the number of measurements obtained by the radar at time k; n represents the number of targets; mask represents a vector composed of Mask layer markers.

7. A radar multi-target tracking device based on Mamba network, characterized in that: include: A splicing module is used to perform matrix splicing on the state information of each target at a preset historical moment to obtain the target state information at all moments; A calculation module, used for calculating the Euclidean distance between the target state information and all the measurement data at the current moment; A generating module, used for generating a target matrix of target point trace features according to the Euclidean distance; An association module, used to input the target matrix into a pre-built Mamba network model to obtain the association probability between the target and the measurement; The filtering module is used to perform filtering and estimation processing on the associated probability using a Kalman filter to obtain a state prediction value of the target at the current moment.

8. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; when the processor calls the computer program instructions, the radar multi-target tracking method based on the Mamba network as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are called by the processor, the radar multi-target tracking method based on the Mamba network as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the radar multi-target tracking method based on the Mamba network as described in any one of claims 1 to 6.

Images