A Track Reassociation Method for Micro-Moving and Moving Targets in Millimeter-Wave Radar

Abstract

This invention discloses a track reassociation method for micro-moving and moving targets in millimeter-wave radar, applied to the field of millimeter-wave positioning and tracking technology. Addressing the problem of inconsistent track estimation in existing technologies, this invention first determines whether a short track state is static or moving. Then, it selects multiple sets of potentially homogeneous short tracks through threshold detection and constructs corresponding Hankel matrices. Missing tracks are solved based on the minimum rank principle, and then track allocation is performed using a generalized linear allocation model based on defined association probabilities. Next, the historical reassociation count of the track is updated. For micro-moving targets, the historical reassociation count is used to update their track state and track information; for moving targets, the track information is directly updated and they are treated as official tracks. This invention achieves reassociation of moving and micro-moving targets.

Description

Technical Field

[0001] This invention belongs to the field of millimeter-wave positioning and tracking technology, and specifically relates to a technique for re-correlation and supplementation of broken tracks of micro-moving and moving targets. Background Technology

[0002] Millimeter-wave-based target tracking technology primarily utilizes electromagnetic waves reflected from the human body. After signal processing, measurements containing information such as distance, radial velocity, azimuth, and elevation angles are obtained. These measurements are then processed by target tracking algorithms to estimate the target's real-time state and cumulative trajectory. This technology has wide applications in civilian industries such as autonomous driving and home control. However, obstacles or target cessation of movement can lead to missed detections. Furthermore, interactions such as intersections between targets can cause a single target to be associated with all measurements. Both of these situations can result in consecutive frames of target association failure during tracking, causing the target to prematurely terminate its trajectory. When a target is successfully clustered again, it is treated as a newly generated target, and its relationship with the previously terminated trajectory cannot be determined, thus initiating a new trajectory. Therefore, reassociation of broken trajectories and compensation for missing trajectories are crucial in target tracking.

[0003] Many research institutions both domestically and internationally have conducted research on continuous indoor target tracking. The University of Oxford in the UK proposed a target recognition-based method for continuous indoor target tracking (Zhao P, et al., "mID: Tracking and Identifying People with Millimeter Wave Radar", International Conference on Distributed Computing in Sensor Systems (DCOSS), 2019, pp. 33-40.). This method uses a fixed-size bounding box to enclose points in each frame that may belong to human targets, forming a grid. By inputting a series of bounding box sequences into a Long Short-Term Memory (LSTM) network, features such as gait and centroid of the target are extracted for target identification. Target recognition can reduce the possibility of accidental track breaks. However, body shape and human gait are weak biometric technologies, and their effectiveness deteriorates with the increase in the number of targets and the complexity of the indoor environment. The National Institute of Technology in Karnataka, India, proposed a method for linking interrupted tracks (Cao Y, Cao J, Zhou Z. Track Segment Association Method Based on Bidirectional Track Prediction and Fuzzy Analysis[J]. Aerospace, 2022, 9(5):274.). This method predicts past tracks and backtracks new tracks using a bidirectional Holt-Winters-based prediction method, and combines fuzzy analysis to determine the similarity between the two tracks before performing track association. However, due to the simplicity of the trend prediction model, it struggles to handle complex target movements, and it does not supplement missing tracks, resulting in a lack of coherence in the final estimated track. Therefore, research on handling track fragmentation in indoor environments is crucial in the field of indoor human target tracking based on millimeter-wave radar. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a track reassociation method for micro-moving and moving targets in millimeter-wave radar, which can effectively determine the homogeneity of broken short tracks and estimate missing target trajectories.

[0005] The technical solution adopted in this invention is: a track reassociation method for micro-moving and moving targets in millimeter-wave radar, comprising:

[0006] S1. Obtain the target's short flight path;

[0007] S2. Mark the short tracks as micro-moving targets or moving targets according to the state variance of each short track;

[0008] S3. Traverse each short track marked in step S2 to obtain a series of track pairs that may be from the same source;

[0009] S4. By calculating the missing tracks between each pair of short tracks in potentially homogeneous track pairs, the missing tracks are used to supplement each pair of short tracks, thereby obtaining a series of supplemented long tracks.

[0010] S5. Based on the series of supplemented long tracks obtained in step S4, calculate the short track correlation matrix;

[0011] S6. Obtain successfully associated tracks based on the short track correlation matrix, and then update the tracks.

[0012] Step S3 specifically includes:

[0013] For two short tracks marked as micro-moving targets, if the difference in the mean state of the two short tracks both meet the requirements, then the two short tracks are marked as track pairs that may be from the same source.

[0014] For two short tracks marked as moving targets, if the time difference, slope difference, and distance difference between the two short tracks meet the requirements, then the two short tracks are marked as a track pair that may be of the same origin.

[0015] The process of calculating the missing tracks in step S4 is as follows:

[0016] Let a pair of short tracks be denoted as short track i and short track j; define the original track as... The supplemented track definition is in For missing flight tracks;

[0017] The target state model is represented by the following linear regression relationship:

[0018]

[0019] The missing tracks are obtained by minimizing the rank of the following Hankel matrix:

[0020]

[0021] in This represents the target state at the start of the selected short flight path. This represents the target status when the trajectory ends.

[0022] Step S5 is as follows:

[0023] Let a certain long trajectory in a series of supplemented long tracks be denoted as Original flight path The corresponding Hankel matrices are denoted as follows: and Long trajectory The corresponding Hankel matrix is ​​denoted as

[0024] right Perform SVD decomposition on each to obtain their respective singular value matrices;

[0025] Will The number of singular values ​​that are greater than the threshold is used as Their respective ranks;

[0026] according to The rank of each short track i and short track j is used to calculate the association probability.

[0027] The following minimum cost problem is solved using the Jonker-Volgenant method to obtain the short track correlation matrix A;

[0028]

[0029]

[0030]

[0031] A ij ∈{0,1}

[0032] Among them, A ij The elements representing A.

[0033] Step S6 specifically involves: if the target is a moving target, then delete the new short track. Information and update old tracks. The system will update the track status to a formal track; if the target is a slightly moving target, the new short track will be deleted. The information, and update the old track. The number of times N has been re-associated ra When the number of reassociations is N ra Greater than the threshold If it does, its track status will be changed to a formal track; otherwise, its track status will be changed to a temporary track.

[0034] The beneficial effects of this invention are as follows: This invention proposes a track reassociation method for micro-moving and moving targets in indoor tracking based on millimeter waves. This method can effectively determine the homology of broken short tracks and estimate missing target trajectories. Compared to the methods mentioned in the background section, this invention can achieve reassociation of multiple short tracks in small-scale scenarios with simultaneous tracking of multiple targets, and also considers the possibility of short-term micro-movements of targets indoors, ensuring the continuity and robustness of indoor tracking and providing strong support for operators to make correct decisions. Attached Figure Description

[0035] Figure 1 This is a flowchart for track reassociation in an indoor environment.

[0036] Figure 2 This is a schematic diagram of a simulation scenario in a specific implementation method.

[0037] Figure 3 This is a simulation diagram of accumulating one hundred frames in a specific implementation method.

[0038] Figure 4 This is the trajectory result for frame 75 in the specific implementation method.

[0039] Figure 5 This is the result of accumulating 100 frames of track data that are not re-associated in the specific implementation.

[0040] Figure 6 This is the track result accumulated after reassociation in the specific implementation method for one hundred frames. Detailed Implementation

[0041] The method of this invention first determines whether a short track is a static or dynamic target. Then, it selects multiple sets of potentially homogeneous short tracks through threshold detection and constructs the corresponding Hankel matrix. Missing tracks are solved based on the minimum rank principle. Finally, based on the defined association probability, a generalized linear allocation model is used to assign tracks. Next, the historical reassociation count of the track is updated. For slightly moving targets, the historical reassociation count is used to update their track status and track information; for moving targets, the track information is directly updated and they are treated as formal tracks. This method of the invention achieves reassociation of moving and slightly moving targets. The processing flow of this method is as follows: Figure 1 As shown, it includes the following steps:

[0042] Step 1: Track State Estimation and Logical Judgment

[0043] Tracks are acquired using millimeter-wave radar, and short tracks are obtained using a sliding window method to obtain the short track in the i-th m-frame at time k. Calculate the state mean of this short track. and variance

[0044] When the variance Less than the threshold If the target moves slightly, mark it as a micro-moving target; otherwise, mark it as a moving target.

[0045] Repeat the above steps until all existing short tracks have been traversed and all existing short tracks have been marked.

[0046] For short tracks i and j marked as slightly moving targets, when their state mean difference is less than a threshold At that time, they are marked as track pairs that may be of the same origin;

[0047] For tracks a and b marked as moving targets, calculate the time difference between their existence. slope difference and distance difference Then, it checks whether all three values ​​are less than a threshold. If they are, they are marked as potentially originating track pairs. This logical judgment process is repeated until all possible track pairs are traversed, ultimately resulting in a series of potentially originating track pairs.

[0048] Step 2: Solve for missing tracks based on the Hankel matrix

[0049] Select short tracks i and j of length m from potentially homogeneous track pairs, and define the original track as follows: The supplemented track definition is in The flight path is missing.

[0050] For a continuously moving human target, its current target state x k Compared with historical target state x k-i Correlation, i.e., using weight a i The current target state can be represented by a weighted average of historical target states. The target state model can be expressed by the following linear regression relationship:

[0051]

[0052] in This represents the current target status of the selected short flight path. This represents the historical target status of the flight path.

[0053] To calculate the missing tracks We need to minimize the order of the above regression model. Minimizing the order of this model is equivalent to minimizing the rank of the Hankel matrix. The n-order Hankel matrix of the trajectory can be calculated as follows:

[0054]

[0055] in This represents the target state at the start of the selected short flight path. This represents the target status when the trajectory ends.

[0056] Then, this invention requires minimizing the Hankel matrix. rank:

[0057]

[0058]

[0059] in, ε is a sufficiently small threshold, which can take the value 0.01, and ⊙ represents the Hadamard product. After convex relaxation, the problem of minimizing the matrix rank is transformed into the following convex optimization problem:

[0060] minTr(X)+Tr(Z)

[0061]

[0062]

[0063] Where X and Z are free variables, and Tr(·) is the trace of the matrix. By traversing all potentially homogeneous track pairs and calculating the Hankel matrix and its missing tracks, a series of supplemented long tracks can be obtained.

[0064] Step 3: Track Association

[0065] Select The flight path in Hankel matrix of the original short track and With the supplementary long track Hankel matrix Perform SVD decomposition. Obtain the singular value matrix S, and use the number of singular values ​​above a threshold as the rank r(H) of the Hankel matrix. Then calculate the association probability between short tracks i and j:

[0066]

[0067] Then, the Jonker-Volgenant method is used to calculate the following minimum cost problem, and the short track correlation matrix A is obtained.

[0068]

[0069]

[0070]

[0071] A ij ∈{0,1}

[0072] Step 4: Track Information Update

[0073] For track pairs successfully associated in step 3, mark the old and new tracks in chronological order. If the target is a moving target, delete the new short track directly in the track management module. Information and update old tracks. The information is used to modify the old track status to the official track; if the target is a slightly moving target, the new short track is deleted. If the information is correct, then update the old flight path. The number of times N has been re-associated ra When the number of reassociations is N ra Greater than the threshold If the old track status is changed to a formal track, then change it to a temporary track to avoid the potential impact of noise.

[0074] The following is a detailed implementation of the present invention based on a MATLAB simulation example.

[0075] Simulation scenarios such as Figure 2 As shown, the room size is 6m × 8m, and the coordinate system zero point is located at the center of the scene. Target 1 walks using a CT model, and target 2 randomly moves slightly within a range of 0.25m from (0m, 0m). To facilitate observation of the simultaneous correlation between the slight movements and the moving target, it is set that both target 1 and target 2 have measurement gaps between frames 50 and 60. The cumulative measurement images of one hundred frames from the simulation are shown below. Figure 3 As shown.

[0076] like Figure 1 As shown, the solution of the present invention includes the following steps:

[0077] Step 1: Track state estimation and logical judgment

[0078] Since the minimum short track length is set to 16 frames, the re-association module for the two targets is entered at frame 75. At this point, as... Figure 4 As shown, there are four short tracks. Short tracks 1 and 3 belong to target 1, and short tracks 2 and 4 belong to target 2, located near (0m, 0m).

[0079] First, the mean and variance of the four short tracks were calculated as follows:

[0080]

[0081] Select the variance threshold for a single track The threshold is set to 0.5, and the target's static / dynamic state is determined. Short tracks 2 and 4 are marked as slightly moving target tracks, while short tracks 1 and 3 are marked as moving target tracks. Then, for slightly moving targets, it is determined whether they might originate from the same source. The mean difference between slightly moving short tracks 2 and 4 is [0.02; 0.01], which is less than the selected threshold. Therefore, the micro-moving target tracks 2 and 4 are marked as potentially homologous track pairs [2,4]. For moving target tracks 1 and 3, since the prerequisite for homology is that the existence times of the two tracks cannot overlap, the time difference between the tracks is... The conditions are met; slope difference The threshold of 0.3419 is taken as the maximum slope between the two short tracks; distance difference Therefore, the moving target tracks 1 and 3 are marked as potentially homogeneous track pairs [1,3]. Finally, a series of potentially homogeneous track pairs {[1,3], [2,4]} are obtained.

[0082] Step 2: Solve for missing tracks based on the Hankel matrix

[0083] For potentially homogeneous track pairs, compute the following convex optimization problem:

[0084] minTr(X)+Tr(Z)

[0085]

[0086]

[0087] Then, this invention can obtain a series of supplemented long tracks. Table 1 shows the supplemented tracks for track pair [1,3]. Due to the long length of the supplemented tracks, this invention only selects the results of the first ten supplemented points as examples, where the first row is the x-value and the second row is the y-value:

[0088] Table 1 shows the supplementary tracks for track pairs [1,3].

[0089] 2.11 2.15 2.19 2.21 2.27 2.35 2.37 2.42 2.44 2.47 -0.78 -0.69 -0.58 -0.52 -0.46 -0.31 -0.25 -0.14 -0.06 0.02

[0090] Table 2 shows the supplementary tracks for track pair [2,4]. Due to the long length of the supplementary tracks, this invention only selects the results of the first ten supplementary points as examples, where the first row is the x-value and the second row is the y-value:

[0091] Table 2 shows the supplementary tracks for track pairs [2,4].

[0092] -0.05 0.03 0.05 0.12 0.01 0.02 0.08 0.09 0.07 0.05 0.05 0.12 0.02 0.08 0.07 0.09 0.07 0.05 -0.01 0.08

[0093] Step 3: Track Association

[0094] Select track pair [1,3], and set the SVD threshold to 10. Calculate the Hankel matrix for the two original short tracks. and Singular value decomposition (SVD) was performed to obtain singular value matrices S1 and S3, where each matrix contains only one value greater than 10. The supplemented long track was then calculated. Hankel matrix Singular value matrix S 13 Only one of these values ​​is greater than 10. The number of singular values ​​exceeding the threshold is used as the rank r(H) of the Hankel matrix, and then the association probability between short tracks 1 and 3 is calculated:

[0095]

[0096] Similarly, the same steps are used to calculate the association probabilities of short tracks 2 and 4:

[0097]

[0098] Then, the Jonker-Volgenant method is used to calculate the minimum cost problem, and the short track correlation matrix A is obtained.

[0099]

[0100] Step 4: Track Information Update

[0101] Based on the correlation matrix, the successfully associated track pairs in step 3 are {[1,3], [2,4]}. Then, the new short track is deleted in the track management module. and Information and update old tracks. The information was then used to change the track status to an official track; then the old track was updated. The number of times N has been re-associated ra =1, due to the number of reassociations N ra Less than the threshold Then its track status will be changed to a temporary track. After multiple subsequent frames are successfully associated, track 2 will become a formal track again.

[0102] One hundred frames of accumulated track images without reassociation, as shown Figure 5 As shown, Figure 5 The presence of four tracks is due to premature track termination caused by multiple missing frames of measurement; the accumulated track image of one hundred frames after reassociation is as follows: Figure 6 As shown, Figure 6 Only two tracks appeared in the measurement, and in the missing part, the present invention estimated the optimal supplementary track, which basically conformed to the pre-set actual situation.

[0103] Simulation results show that the track reassociation method provided by this invention for indoor tracking of micro-moving and moving targets does not cause breakage of target tracks and can estimate missing target tracks, thus verifying the correctness and effectiveness of this invention.

[0104] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A track reassociation method for micro-moving and moving targets in millimeter-wave radar, characterized in that, include: S1. Obtain the target's short trajectory; calculate the state mean and variance of the short trajectory; S2. Mark the short tracks as either slightly moving targets or moving targets based on the state variance of each short track; Step S2 specifically involves: when the state variance of a short track is less than a threshold... If the target is moving, mark it as a micro-moving target; otherwise, mark it as a moving target. S3. Traverse all short tracks marked in step S2 to obtain a series of track pairs that may originate from the same source; step S3 specifically includes: For two short tracks marked as micro-moving targets, if the difference in the mean state of the two short tracks both meet the requirements, then the two short tracks are marked as track pairs that may be from the same source. For two short tracks marked as moving targets, if the time difference, slope difference, and distance difference between the two short tracks meet the requirements, then the two short tracks are marked as a track pair that may be of the same origin. S4. By calculating the missing tracks between each pair of short tracks in potentially homogeneous track pairs, the missing tracks are used to supplement each pair of short tracks, thereby obtaining a series of supplemented long tracks. S5. Based on the series of supplemented long tracks obtained in step S4, calculate the short track correlation matrix; S6. Obtain successfully associated tracks based on the short track correlation matrix, and then update the tracks; specifically, if the target is a moving target, delete the new short track. Information and update old tracks. The system will update the track status to a formal track; if the target is a slightly moving target, the new short track will be deleted. The information, and update the old track. Number of reassociations When the number of reassociations Greater than the threshold If the condition is met, its track status will be changed to a formal track; otherwise, its track status will be changed to a temporary track.

2. The track reassociation method for micro-moving and moving targets in millimeter-wave radar according to claim 1, characterized in that, The process of calculating the missing tracks in step S4 is as follows: Let a pair of short tracks be denoted as: short track and short flight path ; Define the original trajectory as The supplemented track is defined as ,in For missing flight tracks; The target state model is represented by the following linear regression relationship: ； in, Indicates the weight; The missing tracks are obtained by minimizing the rank of the following Hankel matrix: ； in This represents the target state at the start of the selected short flight path. This represents the target status when the trajectory ends.

3. The track reassociation method for micro-moving and moving targets in millimeter-wave radar according to claim 2, characterized in that, Step S5 is as follows: Let a certain long trajectory in a series of supplemented long tracks be denoted as Original flight path The corresponding Hankel matrices are denoted as follows: and Long trajectory The corresponding Hankel matrix is ​​denoted as ; right , , Perform SVD decomposition on each to obtain their respective singular value matrices; Will , , The number of singular values ​​that are greater than the threshold is used as , , Their respective ranks; according to , , Each rank calculation short track and short flight path The probability of association; The following minimum cost problem is solved using the Jonker-Volgenant method to obtain the short track correlation matrix. ; ； in, Indicates a short flight path and The probability of association, express Element.

Images