A parallel processing method for heterogeneous radar trace fusion in dense clutter environment

Abstract

The present application discloses a parallel processing method for the fusion of heterogeneous radar tracks in a dense clutter environment, comprising: coarsely screening radar tracks based on hypothetical tracks to obtain a coarse correlation between radar tracks and hypothetical tracks; predicting the status of the hypothetical tracks based on the coarse correlation to eliminate false tracks; generating hypothetical track status information for each radar platform in the current processing cycle based on the updated hypothetical tracks and recording scores; clustering hypothetical track trees with correlation according to the platform index of each radar platform, and forming hypothetical track clustering information using weighted undirected graph records; determining the maximum weighted independent set of each hypothetical track cluster based on the hypothetical track cluster information; traversing the hypothetical track clusters, pruning each hypothetical track tree based on the maximum weighted independent set result, and updating the track hypothesis tree. The method of the present application can reduce the processing time of heterogeneous radar track fusion and meet the real-time requirements of actual use.

Description

Technical Field

[0001] The present application relates to the field of radar data processing technology, and in particular to a method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment. Background Art

[0002] To improve the stability and accuracy of target tracking, networked radar collaborative detection technology has developed rapidly. Multi-radar track fusion processing is a key component of this technology. It primarily processes the track information of networked radars through track initiation, point-to-track correlation, and track filtering to generate composite tracking tracks with improved accuracy and stability. However, as the number of networked radars increases, so too will the number of false and interfering tracks, posing a severe challenge to the real-time processing of redundant tracks. Furthermore, the increasing number of collaboratively detected targets and the increasing complexity of the filter tracking algorithm also impact processing time, making it difficult for radar data processing systems based on traditional CPU processing architectures to meet the real-time processing requirements of multi-radar collaborative detection.

[0003] Some scholars have studied radar data processing methods based on GPU processing architectures. In the patent "Maneuvering Multi-Target Tracking Method under Dense Clutter Conditions Based on GPU Architecture", a GPU parallel method of "Joint Maximum Likelihood-Interactive Multi-Model-Probabilistic Data Association Algorithm (CJML-IMM-PAD)" was proposed; in the article "Research on GPU-Based Radar Data Processing Technology", the "JPDA and Modified Hough Transform Track Initiation Algorithm" was optimized for GPU parallelization; in the article "Research on Parallelized Multi-Target Tracking Methods Based on CPU+GPU Collaborative Architecture", the "Nearest Neighbor (NN) Algorithm" was optimized for GPU parallelization. The above patents provide improvement ideas for GPU parallel optimization in principle, but they still have the following common problems:

[0004] The application platforms considered in the disclosed patents are all single-radar platforms and cannot be applied to heterogeneous multi-radar platforms.

[0005] Compared with the MHT association algorithm, the NN and JPDA association algorithms proposed in the published patents have poor tracking performance and are difficult to cope with multi-radar point fusion in dense clutter environments. Summary of the Invention

[0006] The embodiments of the present application provide a parallel processing method for heterogeneous radar trace fusion in a dense clutter environment, which is used to parallel optimize the heterogeneous radar trace fusion algorithm and implement parallel processing of the trace-track association algorithm and the cluster allocation algorithm on the GPU side, saving computing resources on the CPU side, reducing the processing time of heterogeneous radar trace fusion, and meeting the real-time requirements of actual use.

[0007] The present application provides a method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment, including:

[0008] Perform spatial registration on the radar traces received from multiple radar platforms to align each radar trace to the same coordinate system;

[0009] According to the hypothetical track, the radar track is roughly screened to obtain the rough correlation between the radar track and the hypothetical track;

[0010] Using GPU and parallel processing, the state of the hypothetical track is predicted based on the coarse correlation relationship, and false tracks are eliminated to obtain the updated hypothetical track;

[0011] Using the CPU, based on the updated hypothetical track, generate hypothetical track status information for each radar platform in the current processing cycle and record the score;

[0012] Cluster the correlated hypothetical track trees according to the platform index of each radar platform, and use weighted undirected graph records to form hypothetical track clustering information;

[0013] Using GPU and parallel processing, based on the hypothetical track cluster information, the maximum weighted independent set of each hypothetical track cluster is determined;

[0014] Using the CPU, the hypothesis track clusters are traversed, each hypothesis track tree is pruned according to the maximum weighted independent set result, and the track hypothesis tree is updated.

[0015] Optionally, based on the hypothetical track, the radar track is roughly screened to obtain a rough correlation relationship between the radar track and the hypothetical track, including:

[0016] For the hypothetical track, the TOMHT algorithm is used to traverse the deep nodes of the composite tracking track tree to complete the coarse correlation processing between the hypothetical track and the radar track information, so as to obtain the coarse correlation relationship between the radar track and the hypothetical track.

[0017] The coarse correlation gate of TOMHT is a spherical gate. The rectangular coordinate position of the track is assumed to be the center origin. The size of the spherical correlation gate is determined according to the maximum distance of the target movement as the radius. By calculating the Euclidean distance between the radar point track and the assumed tracking track, all radar point tracks near the assumed track are determined.

[0018] Optionally, using GPU, parallel processing includes:

[0019] The number of grids and thread blocks is determined according to the GPU performance and the size of the coarse correlation relationship, and the point-track correlation kernel function is called to perform ellipsoidal gate correlation processing between the hypothetical track and the radar point track.

[0020] Optionally, performing state prediction on the hypothetical track based on the coarse correlation and eliminating false tracks to obtain an updated hypothetical track further includes:

[0021] The interactive multi-model-conversion measurement Kalman filter (IMM-CMKF) is used to filter and update the hypothetical trajectory;

[0022] According to the filtering update results, the Mahalanobis distance and the update score of the hypothetical track are calculated, and the ellipsoidal gate rule is used to judge the false track. When the Mahalanobis distance is greater than the set threshold, it is determined that the radar point track fails to be associated with the hypothetical track, so as to eliminate the false track.

[0023] Optionally, clustering the correlated hypothetical track trees according to the platform index of each radar platform, and using a weighted undirected graph to record the hypothetical track clustering information includes:

[0024] Each radar platform is pre-configured with a platform index;

[0025] The nodes of the hypothesis track tree with the same platform index are traversed in a specified order to determine all hypothesis tracks updated by the same radar track to form a weighted hypothesis track cluster graph, wherein a node in the hypothesis track cluster graph represents a hypothesis track, a node weight represents a score of the hypothesis track, and an edge represents that the linked hypothesis tracks use the same updated track.

[0026] Optionally, using GPU and parallel processing, based on the hypothetical track cluster information, the maximum weighted independent set of each hypothetical track cluster is determined to include:

[0027] The maximum weighted independent set (MWIS) of the hypothetical track cluster graph is determined so that the sum of the track scores of all nodes in the independent set is maximized, and the hypothetical track nodes in the maximum weighted independent set are determined as the optimal allocation scheme for this processing cycle.

[0028] Optionally, after updating the trajectory hypothesis tree, the following steps are also included:

[0029] The updated results of the hypothetical track trees of each radar platform are used to update and prune the composite tracking track tree, and the root node information of each composite tracking track tree is used as the optimal fusion result of the current processing cycle.

[0030] An embodiment of the present application further provides a computer device comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the aforementioned method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment are implemented.

[0031] The embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment are implemented.

[0032] The embodiment of the present application performs parallel optimization on the heterogeneous radar point track fusion algorithm, and implements parallel processing of the point track-track association algorithm and the cluster allocation algorithm on the GPU side, which saves computing resources on the CPU side, reduces the processing time of heterogeneous radar point track fusion, and can meet the real-time requirements of actual use.

[0033] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0035] Figure 1 The following is a schematic diagram of the basic process of the parallel processing method for fusion of heterogeneous radar traces according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0037] The present application embodiment provides a method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment, such as Figure 1 As shown in the figure, this application innovatively uses the CPU and GPU architecture to parallel optimize the IMM-KF and MHT algorithms, realizing a parallel processing method for heterogeneous radar trace fusion in dense clutter environments. Specifically, the method includes the following steps:

[0038] In step S101, the received radar traces from multiple radar platforms are spatially aligned to align each radar trace to the same coordinate system. Before the parallel processing of heterogeneous radar traces, the cudaSetDevice function is called to locate the GPU hardware, and the cudaMalloc function is called to request a data cache for processing. The received radar traces from multiple platforms are then spatially aligned, converting the radar trace information and composite tracking information to the same coordinate system. Radar traces are typically polar coordinate information. In a specific implementation, the radar trace information can first be converted from polar coordinates to rectangular coordinates, then combined with the navigation information of the corresponding platform, converted to geocentric rectangular coordinates to form radar trace information in an absolute coordinate system. Finally, the radar trace information is converted to a fusion origin rectangular coordinate system (east-north-sky coordinate system) for heterogeneous radar trace fusion processing to ensure measurement accuracy. In specific applications, the received radar traces can be stored and periodically processed according to the greatest common divisor of the scanning periods of the heterogeneous radars.

[0039] In step S102, the radar tracks are roughly screened based on the hypothetical track to obtain a rough correlation between the radar tracks and the hypothetical track. Specifically, this can be achieved by using a track-oriented multiple hypothesis correlation algorithm (TOMHT).

[0040] After the rough screening, the point-track coarse correlation information is copied to the GPU. Next, in step S103, the GPU performs parallel processing to perform state prediction, measurement prediction, and state update on the hypothetical tracks based on the coarse correlations, eliminating false tracks to obtain updated hypothetical tracks. The updated state information and track scores for hypothetical tracks that meet the threshold conditions are stored. After the parallel processing of the point-track coarse correlation information is completed, the stored processing results are copied from the GPU to the CPU.

[0041] After receiving the point-track-track association results, the CPU replicates the composite tracking track tree by the number of platforms, forming a platform-labeled hypothetical track tree. The tracking target in this embodiment is represented by a multi-layered tree structure, namely a composite tracking track tree. The branch tracks in the tree represent possible target motions, referred to as hypothetical tracks in this embodiment. In step S104, the CPU generates hypothetical track status information for each radar platform in the current processing cycle based on the updated hypothetical tracks, and records the track scores.

[0042] In step S105, the correlated hypothetical track trees are clustered according to the platform index of each radar platform, and the hypothetical track cluster information is formed using weighted undirected graph records and stored. The hypothetical track cluster information formed by the weighted undirected graph records is copied from the CPU to the GPU.

[0043] In step S106, GPU is used for parallel processing to determine the maximum weighted independent set (MWIS) of each hypothetical track cluster based on the hypothetical track cluster information. In the embodiment of the present application, the maximum weighted independent set result of the cluster is used as the optimal allocation result of the point track-track association and stored.

[0044] In step S107, the CPU is used to traverse the hypothesis track clusters, prune each hypothesis track tree according to the maximum weighted independent set result, and update the track hypothesis tree.

[0045] The embodiment of the present application performs parallel optimization on the heterogeneous radar point track fusion algorithm, and implements parallel processing of the point track-track association algorithm and the cluster allocation algorithm on the GPU side, which saves computing resources on the CPU side, reduces the processing time of heterogeneous radar point track fusion, and can meet the real-time requirements of actual use.

[0046] The tracking target in this embodiment is represented by a multi-layer tree structure, namely a composite tracking track tree. The branch tracks in the tree are possible target motions, which are referred to as hypothetical tracks in this application embodiment. In some embodiments, based on the hypothetical tracks, the radar tracks are roughly screened to obtain a rough correlation between the radar tracks and the hypothetical tracks, including:

[0047] For the hypothetical track, the TOMHT algorithm is used to traverse the deep nodes of the composite tracking track tree to complete the coarse correlation processing between the hypothetical track and the radar track information, so as to obtain the coarse correlation relationship between the radar track and the hypothetical track.

[0048] The coarse correlation gate of TOMHT is a spherical gate. Assuming the rectangular coordinate position of the track as the center origin, the size of the spherical correlation gate is determined according to the maximum distance of the target movement as the radius. By calculating the Euclidean distance between the radar point track and the assumed tracking track, all radar point tracks near the assumed track are determined, thereby establishing a point track-track coarse correlation relationship (point-track coarse correlation relationship).

[0049] After traversing the deepest nodes of the composite tracking track tree, the point-to-flight coarse correlation information that satisfies the spherical gate is stored. In some specific examples, when the number of stored nodes meets the GPU parallel thread count of 10240, the cudaMemcpy function is called to copy the point-to-flight coarse correlation information from the CPU to the GPU. The size of the point-to-flight coarse correlation information is: the coarse correlation radar point track information and the target track status information. In some embodiments, using the GPU, parallel processing includes:

[0050] The number of grids and blocks is determined based on the GPU performance and the size of the coarse correlation relationship, and the point-track correlation kernel function is called to perform ellipsoidal gate correlation processing between the hypothetical track and the radar point track. For example, the grid value can be 10 and the block value can be 1024.

[0051] After the point-to-flight correlation kernel function is called, the GPU allocates parallel computing resources based on the number of grids and blocks. Within each independent GPU thread, a set of coarse point-to-flight correlation information is indexed by the thread block size, thread block number, and thread number. In some embodiments, predicting the state of the hypothetical track based on the coarse correlation and eliminating false tracks to obtain an updated hypothetical track also includes:

[0052] The interactive multi-model-conversion measurement Kalman filter (IMM-CMKF) is used to filter and update the hypothetical trajectory;

[0053] Based on the filter update results, the Mahalanobis distance and the hypothetical track update score are calculated, and the ellipsoidal gate rule is used to identify false tracks. If the Mahalanobis distance exceeds the set threshold, the radar point track is determined to have failed to associate with the hypothetical track, and the associated track is considered a false track. The Mahalanobis distance value is -1, and the track will not be included in subsequent hypothetical track updates to eliminate false tracks. The hypothetical track status update information is then stored according to the GPU thread index, recording information such as the Mahalanobis distance, hypothetical track score, and track update status. The hypothetical track status update information is then copied to the CPU. Specifically, after the GPU completes the parallel processing of the point-to-track association, the cudaMemcpy function is called on the CPU to copy the hypothetical track status update information from the GPU to the CPU.

[0054] In some embodiments, clustering the correlated hypothetical track trees according to the platform index of each radar platform and forming hypothetical track clustering information using weighted undirected graph records includes:

[0055] Each radar platform is pre-configured with a platform index. All composite track trees are replicated based on the number of platforms and tagged with the platform index to create a hypothetical track tree. Based on the point-to-track association results and the platform number of the associated radar track, the hypothetical track tree with the same platform index is updated, and a hypothetical track child node is created to update the status and track score of the hypothetical track.

[0056] Clustering is then performed as follows: the nodes of the hypothetical track tree with the same platform index are traversed in a specified order to identify all hypothetical tracks updated by the same radar track, forming a weighted hypothetical track cluster graph. A node in the hypothetical track cluster graph represents a hypothetical track, a node weight represents the score of the hypothetical track, and an edge indicates that the linked hypothetical tracks use the same updated track. After the hypothetical track clustering process is completed, the hypothetical track clustering information is stored as an undirected graph. When the number of stored nodes meets the GPU's parallel thread count of 1024, the cudaMemcpy function is called to copy the clustering information from the CPU to the GPU.

[0057] After determining the number of grids and thread blocks based on GPU performance and the size of the hypothetical track clustering information, the point-to-track allocation kernel function is called to find the optimal hypothesis association solution. In this embodiment, the grid value is 1 and the block value is 1024.

[0058] After calling the point-to-flight allocation kernel function, the GPU allocates parallel computing resources based on the number of grids and blocks. Within each independent thread of the GPU, a hypothetical track cluster is indexed by the thread block size, thread block number, and thread number. In some embodiments, the GPU is used to parallelize the hypothetical track cluster information and determine the maximum weighted independent set for each hypothetical track cluster, including:

[0059] Determine the maximum weighted independent set (MWIS) of the hypothetical track cluster graph so that the sum of the track scores of all nodes in the independent set is maximized, and determine the hypothetical track nodes in the maximum weighted independent set as the optimal allocation scheme for this processing cycle. After completing the relevant calculations to determine the optimal allocation scheme, the maximum weighted independent set information of the cluster is stored according to the thread index of the GPU, and the graph node number of the maximum weighted independent set is recorded. After completing the parallel processing of the optimal point-to-navigation allocation on the GPU side, call the cudaMemcpy function on the CPU side to copy the maximum weighted independent set information from the GPU side to the CPU side.

[0060] In some embodiments, after updating the track hypothesis tree, it also includes: updating the hypothesis track tree results of each radar platform, updating and pruning the composite tracking track tree, and using the root node information of each composite tracking track tree as the optimal fusion result of the current processing cycle.

[0061] Specifically, the hypothesis track clustering graph information can be traversed in sequence, and the optimal hypothesis track in each hypothesis track tree can be found using the maximum weighted independent set information. Then, the hypothesis tracks of other sibling nodes can be deleted and pruned back to the root node to form the optimal hypothesis track tree.

[0062] After pruning all hypothetical track trees, the hypothetical track trees from multiple platforms are merged. The heterogeneous radar points associated with each hypothetical track are sorted in chronological order and updated sequentially into the composite tracking track tree using the IMM-CMKF algorithm. This generates hypothetical track subnodes and updates the hypothetical track status and track score.

[0063] After completing the composite tracking tree merge and update process, all hypothetical track trees are deleted. Then, the hypothetical track with the highest track score in the composite tracking tree is found as the optimal hypothetical track. Similarly, the hypothetical tracks of other sibling nodes are deleted and pruned back to the root node to form the optimal composite tracking tree.

[0064] After updating and pruning all composite tracking track trees, the root node of each track tree is determined to be the optimal composite tracking track for the current processing cycle. The track status information of each root node is output, completing the parallel processing of heterogeneous radar track fusion in dense clutter environments.

[0065] The method of the present application can effectively make up for the problem of insufficient computing resources in the CPU processing architecture, and utilize the multi-threaded resources of the GPU to realize parallel processing of algorithms such as point-to-flight association, track filtering and cluster allocation, effectively improving the processing efficiency of heterogeneous radar point-to-flight fusion. In terms of target tracking algorithm, the method of the present application adopts the interactive multi-model Kalman filter (IMM-KF) algorithm and the multi-hypothesis tracking (MHT) association algorithm, which has strong anti-interference ability and can effectively deal with the negative impact of multiple heterogeneous radars in dense clutter environments, effectively improve the accuracy of point-to-flight association, and obtain continuous and stable target tracking tracks.

[0066] An embodiment of the present application further provides a computer device comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the aforementioned method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment are implemented.

[0067] The embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aforementioned method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment are implemented.

[0068] It should be noted that, in the various embodiments of the present application, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0069] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0070] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0071] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are protected by this application.

Claims

1. A parallel processing method for fusion of heterogeneous radar traces in dense clutter environment, characterized by: include: Perform spatial registration on the radar traces received from multiple radar platforms to align each radar trace to the same coordinate system; According to the hypothetical track, the radar track is roughly screened to obtain the rough correlation between the radar track and the hypothetical track; Using GPU and parallel processing, the state of the hypothetical track is predicted based on the coarse correlation relationship, and false tracks are eliminated to obtain the updated hypothetical track; Using the CPU, based on the updated hypothetical track, generate hypothetical track status information for each radar platform in the current processing cycle and record the score; Cluster the correlated hypothetical track trees according to the platform index of each radar platform, and use weighted undirected graph records to form hypothetical track clustering information; Using GPU and parallel processing, based on the hypothetical track cluster information, the maximum weighted independent set of each hypothetical track cluster is determined; Use the CPU to traverse the hypothetical track clusters, prune each hypothetical track tree according to the maximum weighted independent set result, and update the track hypothesis tree; Based on the hypothetical track, the radar track is roughly screened, and the rough correlation between the radar track and the hypothetical track is obtained, including: For the hypothetical track, the TOMHT algorithm is used to traverse the deep nodes of the composite tracking track tree to complete the coarse correlation processing between the hypothetical track and the radar track information, so as to obtain the coarse correlation relationship between the radar track and the hypothetical track. The TOMHT coarse correlation gate is a spherical gate. The rectangular coordinate position of the assumed track is assumed as the center origin. The size of the spherical correlation gate is determined according to the maximum distance of the target movement as the radius. The Euclidean distance between the radar track and the assumed track is calculated to determine all radar track points near the assumed track. Using GPU and parallel processing, the state prediction of the hypothetical track based on the coarse correlation includes: The number of grids and thread blocks is determined based on the GPU performance and the size of the coarse correlation relationship, and the point-track correlation kernel function is called to perform ellipsoidal gate correlation processing between the hypothetical track and the radar point track. The state of the hypothetical track is predicted based on the coarse correlation, and false tracks are eliminated to obtain the updated hypothetical track. The following also includes: The interactive multi-model-conversion measurement Kalman filter (IMM-CMKF) is used to filter and update the hypothetical trajectory. According to the filtering update results, the Mahalanobis distance and the update score of the hypothetical track are calculated, and the ellipsoidal gate rule is used to judge the false track. When the Mahalanobis distance is greater than the set threshold, it is determined that the radar point track fails to be associated with the hypothetical track, so as to eliminate the false track.

2. The method for fusion and parallel processing of heterogeneous radar traces in a dense clutter environment according to claim 1, characterized in that: Cluster the correlated hypothetical track trees according to the platform index of each radar platform, and use weighted undirected graph records to form the hypothetical track clustering information including: Each radar platform is pre-configured with a platform index; The nodes of the hypothesis track tree with the same platform index are traversed in a specified order to determine all hypothesis tracks updated by the same radar track to form a weighted hypothesis track cluster graph, wherein a node in the hypothesis track cluster graph represents a hypothesis track, a node weight represents a score of the hypothesis track, and an edge represents that the linked hypothesis tracks use the same updated track.

3. The method for fusion and parallel processing of heterogeneous radar traces in a dense clutter environment as claimed in claim 2, characterized in that: Using GPU and parallel processing, based on the hypothetical track cluster information, the maximum weighted independent set of each hypothetical track cluster is determined to include: The maximum weighted independent set MWIS of the hypothetical track cluster graph is determined so that the sum of the track scores of all nodes in the independent set is maximized, and the hypothetical track nodes in the maximum weighted independent set are determined as the optimal allocation scheme for this processing cycle.

4. The method for fusion and parallel processing of heterogeneous radar traces in a dense clutter environment as claimed in claim 3, characterized in that: After updating the track hypothesis tree, it also includes: The updated results of the hypothetical track trees of each radar platform are used to update and prune the composite tracking track tree, and the root node information of each composite tracking track tree is used as the optimal fusion result of the current processing cycle.

5. A computer device, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment according to any one of claims 1 to 4 are implemented.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for parallel processing of heterogeneous radar trace fusion in a dense clutter environment according to any one of claims 1 to 4.

Images