Intelligent track association method driven by state space model

Abstract

The embodiment of the invention relates to the technical field of multi-radar target association, in particular to an intelligent track association method driven by a state space model, which comprises the following steps of: performing spatial registration on track data captured by a radar sensor, so that the position information of the track data is converted from a radar self-coordinate system to a geocentric geodetic coordinate system; performing time registration on a pair of to-be-associated track data subjected to space registration, and performing time alignment on the two pieces of track data by using a cubic spline interpolation algorithm; inputting two pieces of track data subjected to spatial registration into a feature extractor formed by a state space model, and performing sequence modeling on each piece of track data; inputting the track characteristics of the two pieces of track data into a dual-state space aggregation module, comparing the track characteristics of the two pieces of track data, and generating a decision vector; and inputting the decision vector into a classifier to obtain an association result of the two tracks. According to the method, the accuracy of the association result can be well improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of multi-radar target association, and in particular to a state-space model-driven track intelligent association method. Background Art

[0002] In the field of multi-radar target association technology, in order to correctly determine whether the targets captured by multiple radars are the same individual, it is first necessary to perform data association between "tracks". Therefore, as the basis of the entire decision-making system, the accuracy of track association directly determines the quality of subsequent track fusion and target recognition, and plays a key role in subsequent decision-making tasks. In order to further improve the accuracy of track association, many research teams have conducted a lot of in-depth research on various problems in association scenarios and developed a variety of association algorithms with different characteristics.

[0003] The track association algorithms that are currently widely used can be roughly divided into three types: those based on probability statistics, those based on fuzzy mathematics, and those based on artificial intelligence and neural networks. The core idea of ​​the association algorithm based on probability statistics is to calculate various statistical distances between tracks, and then verify whether the tracks are associated through hypothesis testing. However, with the increase in the number of targets in the observation area and the sensor error, this type of algorithm often fails to achieve the expected results. The association algorithm based on fuzzy mathematics calculates the fuzzy membership between tracks through various fuzzy membership functions, and judges the association relationship between tracks by comparing the size of the fuzzy membership. This changes the association idea and further improves the association accuracy of the algorithm in dense scenes. Although the fuzzy mathematics association algorithm solves the association problems in cases of target merging and intersection, the parameter settings in the algorithm are relatively complex, which is not conducive to engineering implementation.

[0004] With the rapid development of machine learning, more and more research teams are applying machine learning to the field of track association technology. The main idea is to first extract features from track segment data, and then classify the tracks according to the features to complete the association. Although the association algorithm based on artificial intelligence and neural networks has solved the problem of radar track association under homogeneous sensors, in the case of a large number of tracking targets, close distances, repeated and crossed tracks, coupled with the influence of factors such as the measurement error of a single multi-target radar propagating with the target distance, resulting in a decrease in radar accuracy, and the increasingly complex track matching between multiple radars, its association speed and reliability are not good. Summary of the invention

[0005] In view of this, an embodiment of the present application proposes a state-space model-driven track intelligent association method, which aims to utilize the long time series modeling capability of the state-space model to solve the problem of track misassociation in dense scenes and improve the accuracy of association results.

[0006] In the first aspect, an embodiment of the present application proposes a state-space model-driven track intelligent association method, comprising the following steps: spatially aligning the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system, and the track data that has completed the spatial alignment is obtained; temporally aligning a pair of track data that have completed the spatial alignment to be associated, and using a cubic spline interpolation algorithm to align the time of the two track data to obtain two track data that have completed the time-space alignment; inputting the two track data that have completed the time-space alignment into a feature extractor composed of a state-space model, and performing sequence modeling on each track data respectively to obtain the track features of the two track data; inputting the track features of the two track data into a dual-state space aggregation module, comparing the track features of the two track data, and generating a decision vector; inputting the decision vector into a classifier to obtain the association result of the two tracks.

[0007] Optionally, the spatial registration of the track data captured by the radar sensor so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system to obtain the track data with completed spatial registration, includes:

[0008] Assume the coordinates of the radar sensor in the latitude and longitude coordinate system are 1. H represents the longitude, latitude, and altitude of the radar sensor in the longitude and latitude coordinate system, respectively;

[0009] Convert the coordinates of the radar sensor in the latitude and longitude coordinate system to the geocentric coordinate system. The coordinates of the radar sensor in the geocentric coordinate system are:

[0010]

[0011] Where N is the radius of the circle, (x ′ ,y ′ ,z ′ ) are the coordinates of the radar sensor in the geocentric coordinate system;

[0012] The coordinate axis of the radar self-coordinate system is rotated to be parallel to the coordinate axis of the geocentric coordinate system, and then the position of the track data captured by the radar sensor is added to the coordinate of the radar sensor in the geocentric coordinate system, so as to complete the conversion of the track data from the radar self-coordinate system to the geocentric coordinate system, and obtain the track data that has completed the spatial registration;

[0013] The conversion formula for converting track data from the radar coordinate system to the geocentric coordinate system is:

[0014]

[0015] Among them, (x 1i ,y1i ,z 1i ) represents the coordinates of a point i in the track data in the radar coordinate system, (x ei ,y ei ,z ei ) represents the coordinates of a point i in the track data in the geocentric coordinate system, R T is the rotation matrix determined based on the coordinates of the radar sensor in the latitude and longitude coordinate system.

[0016] Optionally, performing time registration on a pair of spatially registered track data to be associated, using a cubic spline interpolation algorithm to align the time of the two track data to obtain two track data that have completed time-space registration, includes:

[0017] The track data to be associated and completed with spatial registration are recorded as the first track data and the second track data;

[0018] The latest M seconds of the common time of the first track data and the second track data are intercepted, and the two track data are sampled into a time series of a preset length according to a cubic spline interpolation function to obtain two track data that have completed time-space registration;

[0019] According to the cubic spline interpolation function, the two track data are sampled into a time series of preset length, and two track data with time-space registration are obtained, which can be expressed by the formula:

[0020] A align,1 =S[I(A ori,1 )];

[0021] A align,2 =S[I(A ori,2 )];

[0022] Among them, A ori,1 and A ori,2 denote the first track data and the second track data respectively, I(·) denotes the cubic spline interpolation function, S(·) denotes the sampling function, and A align,1 and A align,2 They respectively represent the first track data after completing the time-space registration and the second track data after completing the time-space registration.

[0023] Optionally, the two track data after time-space registration are input into a feature extractor composed of a state space model, and sequence modeling is performed on each track data to obtain track features of the two track data, including:

[0024] The two track data that have completed time-space registration are input into the feature extractor composed of the state space model, and are first normalized. Then, the sequence feature dimensions of the two normalized track data are projected to P through the linear projection layer;

[0025] The two track data after linear projection are input into the state space model to obtain the track characteristics of the two track data. The track characteristics of the two track data are expressed by the formula:

[0026] F1 = SSM(T1);

[0027] F2 = SSM(T2);

[0028] T1=Linear(A align,1 );

[0029] T2=Linear(A align,2 );

[0030] Among them, Linear(·) represents the linear projection layer, SSM(·) represents the state space model, F1 represents the track characteristics of the first track data of the two track data, and F2 represents the track characteristics of the second track data of the two track data.

[0031] Optionally, the track features of the two track data are input into a dual-state space aggregation module, the track features of the two track data are compared, and a decision vector is generated, which is implemented by the following formula:

[0032] O=Linear[SSM(H1),SSM(H2)];

[0033] H1=Norm[Linear(F1)];

[0034] yQ2=Norm[Linear(F2)];

[0035] Where Norm(·) represents the normalization layer and O represents the decision vector output by the dual state space aggregation module.

[0036] Optionally, the classifier is composed of a multi-layer perceptron, and the classification effect of any nonlinear function is achieved by stacking linear layers and activation functions. The decision vector is input into the classifier to obtain the association result of the two tracks, which is expressed by the formula:

[0037] c=Mlp(O);

[0038] Among them, Mlp(·) represents a multi-layer perceptron, and c represents the association result of two tracks.

[0039] Optionally, the feature extractor composed of the state space model, the dual state space aggregation module and the classifier together constitute a track association model, and the track association model is trained by the following steps:

[0040] Inputting the sample track data pairs with real association labels into the track association model, and obtaining the association results of the sample track data pairs output by the track association model;

[0041] Based on the cross entropy loss function, the loss value between the association result of the sample track data pair output by the track association model and the true association label is calculated; the calculated loss value is expressed by the formula:

[0042]

[0043] Among them, Batch represents the batch size, represents the true associated label of the bth pair of sample track data, c b represents the association result of the bth pair of sample track data output by the track association model, and Loss represents the calculated loss value;

[0044] Based on the loss value, the network parameters of the track association model are updated using the stochastic gradient descent method until the track association model converges, the optimal network parameters are saved, and the trained track association model is obtained.

[0045] The embodiment of the present application proposes a state-space model driven track intelligent association method, which first performs spatial registration on the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system, and then performs temporal registration on a pair of track data to be associated that have completed spatial registration, so that the two are aligned in time, and obtain two regular track data that have completed spatiotemporal registration. The spatiotemporal registration process facilitates the processing of subsequent models and helps to improve the determination speed of the association results. A feature extractor composed of a state-space model is used for feature extraction, and then a dual-state space aggregation module is used for comparison and decision-making, and finally the classifier outputs the association result. As a component of the feature extractor and feature aggregation network, the state-space model has excellent performance in capturing local correlation and long sequence modeling, and effectively solves the association problem in the case of dense tracks. In addition, the state-space model also has outstanding advantages in hardware acceleration processing, which can improve the association speed and is easier to deploy in actual scenarios.

[0046] In the second aspect, an embodiment of the present application proposes a state-space model-driven track intelligent association system, the system comprising: a spatial registration unit, a temporal registration unit and a track association model, the track association model comprising a feature extractor composed of a state-space model, a dual-state space aggregation module and a classifier; a spatial registration unit, used to perform spatial registration on the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system, and the track data with spatial registration is obtained; a temporal registration unit, used to perform temporal registration on a pair of track data to be associated that have completed spatial registration, and use a cubic spline interpolation algorithm to align the time of the two track data to obtain two track data with time-space registration; a feature extractor composed of a state-space model, used to perform sequence modeling on each track data respectively, and obtain the track characteristics of the two track data; a dual-state space aggregation module, used to compare the track characteristics of the two track data, and generate a decision vector; a classifier, used to input the decision vector into the classifier to obtain the association result of the two tracks.

[0047] It can be understood that the beneficial effects of the second aspect mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related technologies, the drawings required for use in the embodiments of the present application or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 is a flow chart of a state space model driven track intelligent association method provided in one embodiment of the present application;

[0050] Figure 2 is a schematic diagram of the structure of a track association model provided in an embodiment of the present application;

[0051] Figure 3 is a schematic diagram of the structure of a feature extractor provided in one embodiment of the present application;

[0052] Figure 4 is a structural schematic diagram of a track intelligent association system driven by a state space model provided in another embodiment of the present application;

[0053] Figure 5 It is a structural diagram of an electronic device deployed in another embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. In the various embodiments of the present application, many technical details are proposed in order to make the reader better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is only for the convenience of description, and the specific implementation mode of the present application should not constitute any limitation. The various embodiments can be combined with each other and referenced to each other under the premise of no contradiction.

[0055] An embodiment of the present application proposes a state-space model-driven intelligent track association method, which is applied to an electronic device, wherein the electronic device can be a terminal or a server. In this embodiment and the following embodiments, the electronic device is described using the server as an example. The implementation details of the state-space model-driven intelligent track association method proposed in this embodiment are described in detail below. The following content is only the relevant implementation details provided for ease of understanding and is not necessary for the implementation of this solution.

[0056] The specific process of the state space model driven track intelligent association method proposed in this embodiment can be as follows: Figure 1 As shown, including:

[0057] Step 101 , spatially aligning the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system, and the track data with completed spatial alignment is obtained.

[0058] In the specific implementation, the track data captured by the radar sensor is recorded based on the radar auto-coordinate system, and needs to be converted to the geocentric coordinate system to be more scientific and reasonable. Therefore, the server first needs to spatially register the track data captured by the radar sensor, so that the position information of the track data is converted from the radar auto-coordinate system to the geocentric coordinate system to obtain the track data that has completed spatial registration.

[0059] In one example, the coordinates of the radar sensor in the latitude and longitude coordinate system are 1. H represents the longitude, latitude and altitude of the radar sensor in the longitude and latitude coordinate system. When the server performs spatial registration of the track data, it first needs to convert the coordinates of the radar sensor in the longitude and latitude coordinate system to the geocentric coordinate system.

[0060] The coordinates of the radar sensor in the geocentric coordinate system can be expressed as:

[0061]

[0062] Among them, N is the radius of the y-axis circle, a represents the major radius of the earth, e represents the first eccentricity of the earth, (x ′ ,y ′ ,z ′ ) is the coordinate of the radar sensor in the geocentric coordinate system.

[0063] After obtaining the coordinates of the radar sensor in the geocentric coordinate system, the server can rotate the coordinate axis of the radar self-coordinate system to be parallel to the coordinate axis of the geocentric coordinate system, and then add the position value of the track data captured by the radar sensor to the coordinates of the radar sensor in the geocentric coordinate system, completing the conversion of the track data from the radar self-coordinate system to the geocentric coordinate system, and obtaining the track data that has completed spatial registration.

[0064] The conversion formula of track data from radar coordinate system to geocentric coordinate system can be expressed as:

[0065]

[0066] Among them, (x 1i ,y 1i ,z 1i ) represents the coordinates of a point i in the track data in the radar self-coordinate system (also known as the north-east coordinate system), (x ei ,y ei ,z ei ) represents the coordinates of a point i in the track data in the geocentric coordinate system, R T is the rotation matrix determined based on the coordinates of the radar sensor in the latitude and longitude coordinate system.

[0067] Step 102 , time-align a pair of spatially-aligned track data to be associated, and use a cubic spline interpolation algorithm to align the time of the two track data to obtain two track data that have completed time-space alignment.

[0068] In a specific implementation, after completing the spatial registration of the track data captured by the radar sensor, the server can perform temporal registration on a pair of spatially registered track data to be associated. The specific method is to use a cubic spline interpolation algorithm to align the time of the two track data, thereby obtaining two track data that have completed time-space registration.

[0069] In one example, the track data to be associated and completed spatial registration are respectively the first track data and the second track data. The server intercepts the latest M seconds (generally set to 60 seconds) of the common time of the first track data and the second track data, and samples the two track data into a time series of a preset length according to the cubic spline interpolation function to obtain two track data that have completed spatial registration. The preset length is generally set to 16, and the specific value can be set by a technician in this field according to actual needs.

[0070] In one example, the server samples two track data into a time series of preset length according to the cubic spline interpolation function, and obtains two track data that have completed time-space registration, which can be expressed by the formula:

[0071] A align,1 =S[I(A ori,1 )];

[0072] A align,2 =S[I(A ori,2 )];

[0073] Among them, A ori,1 and A ori,2 denote the first track data and the second track data respectively, I(·) denotes the cubic spline interpolation function, S(·) denotes the sampling function, and A align,1 and A align,2 They respectively represent the first track data after completing the time-space registration and the second track data after completing the time-space registration.

[0074] Step 103, input the two track data that have completed time-space registration into a feature extractor composed of a state space model, perform sequence modeling on each track data, and obtain track features of the two track data.

[0075] In the specific implementation, after completing the spatiotemporal registration of the track data, the track association model can be used to determine whether two track data are associated. The network structure of the track association model can be as follows: Figure 2 As shown in the figure, the track association model consists of a feature extractor, a dual state space aggregation module and a classifier. The feature extractor is based on the state space model. The server inputs two track data that have completed time-space registration into the feature extractor composed of the state space model, and performs sequence modeling on each track data to obtain the track features of the two track data.

[0076] In one example, the network structure of the feature extractor can be as follows Figure 3As shown, the server inputs the two track data that have completed time-space registration into the feature extractor composed of the state-space model, first normalizes them, and then projects the sequence feature dimension of the two normalized track data from 3 to P through the linear projection layer, and then inputs the two track data after linear projection into the state-space model to obtain the track features of the two track data.

[0077] In one example, the track characteristics of two track data are expressed by the formula:

[0078] F1 = SSM(T1);

[0079] F2 = SSM(T2);

[0080] T1=Linear(A align,1 );

[0081] T2=Linear(A align,2 );

[0082] Where Linear(·) represents the linear projection layer, SSM(·) represents the state space model, T1 represents the first track data after linear projection, T2 represents the second track data after linear projection, and T1∈R 16×P , T2∈R 16×P , F1 represents the track characteristics of the first track data of the two track data, and F2 represents the track characteristics of the second track data of the two track data.

[0083] In an example, the specific calculation process of the state space model can be expressed as:

[0084]

[0085] Among them, A, B, and C are all learnable network parameters in the state space model, and h t represents the hidden state, h -1 =0,x t and t They represent the input and output of the state space model respectively, and Δ represents x t The change step size.

[0086] Step 104: input the track features of the two track data into a dual-state space aggregation module, compare the track features of the two track data, and generate a decision vector.

[0087] In the specific implementation, the dual-state space aggregation module is also built based on the state space model. The feature extractor inputs the track features of the two track data into the dual-state space aggregation module. After passing through the normalization layer and the projection layer, it enters the state space model to compare the track features of the two track data and finally generate a decision vector.

[0088] In one example, the track features of two track data are input into the dual-state space aggregation module, and the track features of the two track data are compared to generate a decision vector, which can be achieved by the following formula:

[0089] O=Linear[SSM(H1),SSM(H2)];

[0090] H1=Norm[Linear(F1)];

[0091] H2=Norm[Linear(F2)];

[0092] Where Norm(·) represents the normalization layer and O represents the decision vector output by the dual state space aggregation module.

[0093] Step 105, input the decision vector into the classifier to obtain the association result of the two tracks.

[0094] In the specific implementation, the dual-state space aggregation module inputs the decision vector into the classifier, and the classifier classifies whether it is associated or not to obtain the association result of the two tracks.

[0095] In one example, the classifier is composed of a multi-layer perceptron. By stacking linear layers and activation functions, the classification effect of any nonlinear function is achieved. The decision vector is input into the classifier to obtain the association result of the two tracks, which can be expressed by the formula:

[0096] c=Mlp(O);

[0097] Among them, Mlp(·) represents a multi-layer perceptron, and c represents the association result of two tracks.

[0098] In an example, a feature extractor composed of a state space model, a dual state space aggregation module, and a classifier together constitute a track association model. The track association model needs to be trained before use, and the server needs to perform supervised iterative training on the track association model.

[0099] The server inputs the sample track data pairs with real association labels into the track association model to obtain the association results of the sample track data pairs output by the track association model. Then, based on the cross entropy loss function, the loss value between the association results of the sample track data pairs output by the track association model and the real association labels is calculated. Finally, based on the loss value, the network parameters of the track association model are updated using the stochastic gradient descent method until the track association model converges, the optimal network parameters are saved, and the trained track association model is obtained.

[0100] The calculated loss value can be expressed by the formula:

[0101]

[0102] Among them, Batch represents the batch size, represents the true associated label of the bth pair of sample track data, c b It represents the association result of the bth pair of sample track data output by the track association model, and Loss represents the calculated loss value.

[0103] In one example, when training a track association model, an Adam optimizer with a learning rate of 1e-3 is used, the learning rate decrease rate is set to decrease to 0.1 of itself every 1,000 epochs, and the learning (training) process is completed after 10,000 epochs.

[0104] The present embodiment proposes a state-space model driven track intelligent association method, which performs spatial registration on the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system, and then performs temporal registration on a pair of track data to be associated that have completed spatial registration, so that the time of the two is aligned, and two regular track data that have completed spatiotemporal registration are obtained. The spatiotemporal registration process facilitates the processing of subsequent models and helps to improve the determination speed of the association results. A feature extractor composed of a state-space model is used for feature extraction, and then a dual-state space aggregation module is used for comparison and decision-making, and finally the classifier outputs the association result. As a component of the feature extractor and feature aggregation network, the state-space model has excellent performance in capturing local correlation and long sequence modeling, and effectively solves the association problem in the case of dense tracks. In addition, the state-space model also has outstanding advantages in hardware acceleration processing, which can improve the association speed and is easier to deploy in actual scenarios.

[0105] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this application.

[0106] Another embodiment of the present application proposes a state-space model driven track intelligent association system. The implementation details of the state-space model driven track intelligent association system proposed in this embodiment are specifically described below. The following content is only for the convenience of understanding the implementation details provided, and is not necessary for implementing this example. Figure 4 It is a structural diagram of a state-space model-driven track intelligent association system proposed in this embodiment, the system includes: a spatial alignment unit 201, a temporal alignment unit 202 and a track association model 300, the track association model 300 includes a feature extractor 301 composed of a state-space model, a dual-state space aggregation module 302 and a classifier 303.

[0107] The spatial registration unit 201 is used to perform spatial registration on the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system to obtain the track data with completed spatial registration.

[0108] The time registration unit 202 is used to perform time registration on a pair of track data to be associated and complete spatial registration, and use a cubic spline interpolation algorithm to align the time of the two track data to obtain two track data that have completed time and spatial registration.

[0109] The feature extractor 301 composed of the state space model is used to perform sequence modeling on each track data to obtain track features of two track data.

[0110] The dual-state space aggregation module 302 is used to compare the track features of two track data and generate a decision vector.

[0111] The classifier 303 is used to input the decision vector into the classifier to obtain the association result of the two tracks.

[0112] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0113] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in conjunction with the above method embodiment. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the above embodiments.

[0114] Another embodiment of the present application deploys an electronic device, whose specific structure is as follows Figure 5 As shown, it includes: at least one processor 401; and a memory 402 that is communicatively connected to the at least one processor 401; wherein the memory 402 stores instructions that can be executed by the at least one processor 401, and the instructions are executed by the at least one processor 401 so that the at least one processor 401 can execute a state-space model-driven track intelligent association method described in the above embodiments.

[0115] Among them, the memory and the processor can be connected in a bus manner, and the bus can include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all well known in the art and will not be further described in this article. The bus interface is responsible for providing an interface between the bus and the transceiver. The transceiver can be one component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium through an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0116] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0117] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (such as a single-chip microcomputer, chip, etc.) or a processor (processor) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.

[0118] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A state space model driven track intelligent association method, characterized in that: include: Perform spatial registration on the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system, and obtain the track data with completed spatial registration; Performing time registration on a pair of spatially registered track data to be associated, using a cubic spline interpolation algorithm to align the two track data in time, and obtaining two track data that have completed time-space registration; The two track data after time-space registration are input into the feature extractor composed of the state space model, and sequence modeling is performed on each track data to obtain the track features of the two track data; Input the track features of the two track data into the dual-state space aggregation module, compare the track features of the two track data, and generate a decision vector; The decision vector is input into the classifier to obtain the association result of the two tracks.

2. A state space model driven track intelligent association method as claimed in claim 1, characterized in that: The spatial registration of the track data captured by the radar sensor is performed so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system to obtain the track data that has completed the spatial registration, including: Assume the coordinates of the radar sensor in the latitude and longitude coordinate system are 1. H represents the longitude, latitude, and altitude of the radar sensor in the longitude and latitude coordinate system, respectively; Convert the coordinates of the radar sensor in the latitude and longitude coordinate system to the geocentric coordinate system. The coordinates of the radar sensor in the geocentric coordinate system are: Where N is the radius of the circle, (x ′ ,y ′ ,z ′ ) are the coordinates of the radar sensor in the geocentric coordinate system; The coordinate axis of the radar self-coordinate system is rotated to be parallel to the coordinate axis of the geocentric coordinate system, and then the position of the track data captured by the radar sensor is added to the coordinate of the radar sensor in the geocentric coordinate system, so as to complete the conversion of the track data from the radar self-coordinate system to the geocentric coordinate system, and obtain the track data that has completed the spatial registration; The conversion formula for converting track data from the radar coordinate system to the geocentric coordinate system is: Among them, (x 1i ,y 1i ,z 1i ) represents the coordinates of a point i in the track data in the radar coordinate system, (x ei ,y ei ,z ei ) represents the coordinates of a point i in the track data in the geocentric coordinate system, R T is the rotation matrix determined based on the coordinates of the radar sensor in the latitude and longitude coordinate system.

3. A state space model driven track intelligent association method as claimed in claim 2, characterized in that: The method of performing time registration on a pair of track data to be associated and completing spatial registration, using a cubic spline interpolation algorithm to align the time of the two track data, and obtaining two track data that have completed time-space registration, includes: The track data to be associated and completed with spatial registration are recorded as the first track data and the second track data; The latest M seconds of the common time of the first track data and the second track data are intercepted, and the two track data are sampled into a time series of a preset length according to a cubic spline interpolation function to obtain two track data that have completed time-space registration; According to the cubic spline interpolation function, the two track data are sampled into a time series of preset length, and two track data with time-space registration are obtained, which can be expressed by the formula: A align,1 =SpI(A ori,1 )]; A align,2 =S[I(A ori,2 )]; Among them, A ori,1 and A ori,2 denote the first track data and the second track data respectively, I(·) denotes the cubic spline interpolation function, S(·) denotes the sampling function, and A align,1 and A align,2 They respectively represent the first track data after completing the time-space registration and the second track data after completing the time-space registration.

4. A state space model driven track intelligent association method as claimed in claim 3, characterized in that: The two track data after time-space registration are input into a feature extractor composed of a state space model, and sequence modeling is performed on each track data to obtain track features of the two track data, including: The two track data that have completed time-space registration are input into the feature extractor composed of the state space model, and are first normalized. Then, the sequence feature dimensions of the two normalized track data are projected to P through the linear projection layer; The two track data after linear projection are input into the state space model to obtain the track characteristics of the two track data. The track characteristics of the two track data are expressed by the formula: F1 = SSM(T1); F2 = SSM(T2); T1=Linear(A aliin,1 ); T2=Linear(A align,2 ); Among them, Linear(·) represents the linear projection layer, SSM(·) represents the state space model, F1 represents the track characteristics of the first track data of the two track data, and F2 represents the track characteristics of the second track data of the two track data.

5. A state space model driven track intelligent association method as claimed in claim 4, characterized in that: The track features of the two track data are input into the dual-state space aggregation module, the track features of the two track data are compared, and a decision vector is generated, which is implemented by the following formula: O=Linear[SSM(H1),SSM(H2)]; H1=Norm[Linear(F1)]; H2=Norm[Linear(F2)]; Where Norm(·) represents the normalization layer and O represents the decision vector output by the dual state space aggregation module.

6. A state space model driven track intelligent association method as claimed in claim 5, characterized in that: The classifier is composed of a multi-layer perceptron, which achieves the classification effect of any nonlinear function by stacking linear layers and activation functions. The decision vector is input into the classifier to obtain the association result of the two tracks, which is expressed by the formula: c=Mlp(O); Among them, Mlp(·) represents a multi-layer perceptron, and c represents the association result of two tracks.

7. A state space model driven track intelligent association method according to any one of claims 1 to 6, characterized in that: The feature extractor composed of the state space model, the dual state space aggregation module and the classifier together constitute the track association model, which is trained through the following steps: Inputting the sample track data pairs with real association labels into the track association model, and obtaining the association results of the sample track data pairs output by the track association model; Based on the cross entropy loss function, the loss value between the association result of the sample track data pair output by the track association model and the true association label is calculated; the calculated loss value is expressed by the formula: Among them, Batch represents the batch size, represents the true associated label of the bth pair of sample track data, c b represents the association result of the bth pair of sample track data output by the track association model, and Loss represents the calculated loss value; Based on the loss value, the network parameters of the track association model are updated using the stochastic gradient descent method until the track association model converges, the optimal network parameters are saved, and the trained track association model is obtained.

8. A state space model driven track intelligent association system, characterized in that: The system comprises: a spatial registration unit, a temporal registration unit and a track association model, wherein the track association model comprises a feature extractor composed of a state space model, a dual state space aggregation module and a classifier; A spatial registration unit is used to perform spatial registration on the track data captured by the radar sensor, so that the position information of the track data is converted from the radar coordinate system to the geocentric coordinate system to obtain track data that has completed spatial registration; A time registration unit is used to perform time registration on a pair of track data to be associated and complete spatial registration, and use a cubic spline interpolation algorithm to align the time of the two track data to obtain two track data that have completed time and space registration; The feature extractor composed of the state space model is used to perform sequence modeling on each track data to obtain the track features of the two track data; Dual-state space aggregation module, used to compare the track characteristics of two track data and generate a decision vector; The classifier is used to input the decision vector into the classifier to obtain the association result of the two tracks.

Images