A multi-method joint pruning-based computation amount control method in multi-hypothesis tracking

Abstract

The application discloses a kind of based on multi-method joint pruning in multi-hypothesis tracking calculation amount control method, comprising the following steps: S1. establish target track library and update confirmation;S2. obtain radar echo data and carry out target merging;S3. predict track position, and each radar echo data is associated with predicted track position;S4. the same associated radar echo data of same target track is redundantly deassociated;S5. update, branch and calculate track score to predicted track;S6. according to track score, pruning is carried out to predicted track;S7. pruning is carried out to target possession track;S8. cluster updates;S9. generate hypothesis, and calculate global score.The application establishes a stable and efficient pruning system, realizes the effective control to track quantity, avoids the exponential explosion of hypothesis number due to too many track quantities in clustering.

Description

Technical Field

[0001] This invention relates to the field of radar track detection technology, specifically to a computational control method for multi-hypothesis tracking based on multi-method joint pruning. Background Technology

[0002] In the field of radar-based maritime surveillance, complex sea conditions such as sea clutter and secondary reflections, as well as complex ship behaviors such as docking, pose increasingly severe challenges to radar tracking capabilities. Traditional nearest-neighbor correlation tracking is insufficient to handle these challenges. To address these difficulties, multi-hypothesis tracking algorithms are needed. Multi-hypothesis tracking methods generate multiple hypotheses when faced with complex scenarios where the correct echo cannot be determined. These hypotheses continue tracking for a certain number of frames, using subsequent echo data to determine which one is correct—a delayed decision-making scheme.

[0003] Since its inception, multi-hypothesis tracking theory has been considered one of the optimal tracking theories. However, the main reason it hasn't become a mainstream algorithm is its high computational resource requirements. Compared to traditional nearest-neighbor correlation tracking, multi-hypothesis tracking involves more tracks, more functions, and a more complex process. Furthermore, in the hypothesis generation stage, the number of hypotheses increases exponentially with the number of tracks. Therefore, in some complex tracking scenarios, excessive computation and time consumption can easily occur, affecting the real-time performance of the radar. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning, comprising the following steps: S1. Establishing and updating a target track database; S2. Acquiring radar echo data and merging targets; S3. Predicting track positions and associating each radar echo data with the predicted track position; S4. Redundantly deassociating radar echo data with the same target track; S5. Updating, branching, and calculating track scores for the predicted tracks; S6. Pruning the predicted tracks based on track scores; S7. Pruning tracks owned by the target; S8. Clustering and updating; S9. Generating hypotheses and calculating global scores.

[0005] Further, step S2 specifically involves: merging the measurement echo data of multiple units of a target, finding the connectivity position by the amplitude of the echo, and using the centroid position as the measurement position of the echo. After merging, the echo data of the (k+1)th frame is obtained as Z(k+1)={z1(k+1),z2(k+1),z3(k+1)…z M (k+1)}, where M is the number of echoes in this frame.

[0006] Further, step S3 specifically includes the following sub-steps: S31. Limiting the range gate, wherein the distance r between track i and echo j is... ij The calculation formula is:

[0007] In the formula, H is the measurement matrix;

[0008] S32. Calculate the normalized distance g ij Before calculating the normalized distance, the motion state and covariance matrix of the track need to be predicted; the prediction of the motion state of track i in the (k+1)th frame. An iterative update method is used, based on the motion state X of the k-th frame of the track. i (k) is calculated using the following formula:

[0009]

[0010] Where F is the state transition matrix;

[0011] Simultaneously, the covariance of the trajectory is predicted. The predicted covariance P(k+1|k) for the (k+1)th frame is obtained by iterating from the covariance P(k|k) of the k-th frame. The calculation formula is: P(k+1|k)=F(k)·P(k|k)·F'(k)+Q(k), where Q(k) is the process noise covariance. Based on the trajectory prediction results and echo measurement results, the normalized distance g is... ij The calculation formula is:

[0012] Where S is the innovation covariance matrix of the trajectory in the Kalman filtering process, which is calculated from the covariance. The calculation method is: S(k+1)=H·P(k+1│k)·H'+R(k+1), where R(k+1) is the measurement noise matrix;

[0013] S33. Determine the normalized distance g ij Whether it falls within the gate: if g ij If the value is less than the set gate value, the track is considered to be associated with the measured echo.

[0014] Furthermore, step S4 specifically includes the following sub-steps: S41. Assigning the correlation between the track and the echo according to the normalized distance g. ijThe values ​​are divided into levels 1, 2, and 3 from smallest to largest; S42. Extract one of the target's tracks; S43. Extract a radar echo associated with the track and determine whether the association between the track and the radar echo is level 1; S44. Re-extract another track of the target and determine whether the re-extracted track is associated with the radar echo extracted in step S43; S45. Determine whether the track extracted in step S42 has any associated echoes; S46. Determine whether the target has any other tracks.

[0015] Furthermore, step S43 specifically includes the following sub-steps: when the association between the track and the radar echo is a level 1 association, step S44 is executed; when the association between the track and the radar echo is not a level 1 association, step S43 is executed repeatedly.

[0016] The S44 step specifically includes the following sub-steps: when the association between the re-extracted track and the radar echo is a level 1 association, repeat step S44; when the association between the re-extracted track and the radar echo is not a level 1 association, remove the association and determine whether there are any remaining tracks, then execute step S45.

[0017] The S45 step specifically includes the following sub-steps: when the track extracted in the S42 step still has associated echoes, repeat step S43; when the track extracted in the S42 step does not have associated echoes, execute step S46.

[0018] The S45 step specifically includes the following sub-steps: if there are other tracks for the target, repeat step S42; if there are no other tracks for the target, execute step S5.

[0019] Furthermore, the formula for calculating the track score in step S5 is: Score(k+1)=Score(k)+ΔScore(k+1); In the formula, P D β represents the radar detection probability. FT denoted as radar false alarm rate; S is the information covariance, which is calculated from the covariance. The calculation method is: S(k+1)=H·P(k+1│k)·H'+R(k+1), where R(k+1) is the measurement noise matrix; g is the normalized distance between the track and the associated echo.

[0020] Further, step S6 specifically includes the following sub-steps: S61. Set a threshold value and calculate the average value of the track scores; S62. Determine whether the average value is greater than or equal to the threshold value; S63. Calculate the sum of the scores of the most recent N frames and determine whether the sum of the scores of the N frames is greater than or equal to the threshold value; S64. Determine whether the target track number is 1; wherein, the formula for calculating the average value of the track scores in step S61 is: AveScore=(Score(k)) / Step; the formula for calculating the sum of the scores of the N frames in step S63 is: In the above formula, Step is the number of track frames, and N ≤ Step.

[0021] Further, step S62 specifically includes the following steps: when the average value is greater than or equal to the threshold value, execute step S63; when the average value is less than the threshold value, delete this track; step S63 specifically includes the following steps: when the sum of the scores of N shots is greater than or equal to the threshold value, execute step S64; when the sum of the scores of N shots is less than the threshold value, delete this track; step S64 specifically includes the following steps: when the number of target tracks is equal to 1, execute step S8; when the number of target tracks is greater than 1, execute step S7.

[0022] Further, step S7 specifically includes the following steps: S71. Merging similar tracks to the same target; S72. N-SCAN pruning based on radar echoes; S73. Limiting the number of tracks for the target; wherein, step S72 specifically involves: recording the radar echo sequence number used by each track in the most recent M frames, controlling the branch depth to N, comparing the usage of the radar echo in the Nth frame of each track, and deleting all tracks with long branches based on the track with the highest score.

[0023] Furthermore, step S9 specifically includes the following steps:

[0024] Hypothesis generation: Build a conflict table for the tracks, search for it, combine non-conflicting tracks as a hypothesis, then use another track that has not been included as the head to search again, build the next track, and repeat until all tracks have been included with hypotheses;

[0025] Calculate the global score: Based on the score for each hypothesis, calculate the global score of the track; the formula for calculating the global score is as follows:

[0026] In the formula, T i H represents the i-th track. j Assuming the location of this flight path, H k For all hypotheses in the cluster, score(H) j H is the hypothesis of the j-th hypothesis. j Score,

[0027] This invention provides a computational control method for multi-hypothesis tracking based on multi-method joint pruning, which has the following beneficial effects:

[0028] This invention establishes a stable and efficient pruning system to effectively control the number of tracks and avoid the exponential explosion of hypotheses caused by an excessive number of tracks in clusters. This invention focuses on the number of tracks in clusters, controlling the number of tracks before generating hypotheses to prevent hypothesis explosion; the echo-based N-SCAN pruning method more efficiently and quickly controls the branch depth. After hypotheses are completed, they are used to calculate the global score, and tracks with low global scores are pruned to reduce the computational load for the next frame. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0030] Figure 1 A flowchart of the method provided by the present invention;

[0031] Figure 2 A flowchart of the method for removing redundant echoes provided by the present invention;

[0032] Figure 3 A flowchart for merging similar flight paths provided by this invention;

[0033] Figure 4 The flowchart of N-SCAN pruning based on radar echo provided for this invention. Detailed Implementation

[0034] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0035] The following detailed description of the implementation method of the present invention is in conjunction with the accompanying drawings. The description is only a partial embodiment and not all embodiments. For clarity, representations and descriptions unrelated to the present invention are omitted in the drawings and description.

[0036] To provide a clearer understanding of the technical features, objectives, and beneficial effects of this invention, the following detailed description of the technical solution is provided. Obviously, the described embodiments are only a portion of the embodiments of this invention, not all of them, and should not be construed as limiting the scope of implementation of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the protection scope of this invention.

[0037] like Figure 1 As shown, this invention provides a computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning, comprising the following steps: S1. Establishing a target track database and updating and confirming it; S2. Acquiring radar echo data and merging targets; S3. Predicting track positions and associating each radar echo data with the predicted track position; S4. Redundantly deassociating radar echo data with the same target track; S5. Updating, branching, and calculating track scores for the predicted tracks; S6. Pruning the predicted tracks based on the track scores; S7. Pruning tracks owned by the target; S8. Clustering and updating; S9. Generating hypotheses and calculating global scores.

[0038] Step S2 specifically involves: merging the measurement echo data from multiple units of a target, finding connectivity points based on the echo amplitude, and using the centroid position as the measurement location of the echo. After merging, the echo data Z(k+1) of the (k+1)th frame is obtained: Z(k+1) = {z1(k+1), z2(k+1), z3(k+1)...z...} M (k+1)}, where M is the number of echoes in this frame.

[0039] Step S3 involves associating tracks and echoes, with each track paired with each echo. This includes the following sub-steps: S31. Applying range gate constraints, where the distance r between track i and echo j... ij The calculation formula is:

[0040] In the formula, H is the measurement matrix.

[0041] S32. Calculate the normalized distance g ij Before calculating the normalized distance, the motion state and covariance matrix of the track need to be predicted; the prediction of the motion state of track i in the (k+1)th frame. An iterative update method is used, based on the motion state X of the k-th frame of the track. i (k) is calculated using the following formula:

[0042] Where F is the state transition matrix.

[0043] Simultaneously, the covariance of the trajectory is predicted. The predicted covariance P(k+1|k) for the (k+1)th frame is obtained by iterating from the covariance P(k|k) of the k-th frame. The calculation formula is: P(k+1|k)=F(k)·P(k|k)·F'(k)+Q(k), where Q(k) is the process noise covariance. Based on the trajectory prediction results and echo measurement results, the normalized distance g is... ij The calculation formula is:

[0044] Where S is the information covariance matrix of the trajectory during the Kalman filtering process, which is calculated from the covariance. The calculation method is: S(k+1)=H·P(k+1│k)·H'+R(k+1), where R(k+1) is the measurement noise matrix.

[0045] S33. Determine the normalized distance g ij Whether it falls within the gate: if g ij If the value is less than the set gate value, the track is considered to be associated with the measured echo.

[0046] In step S4, the correlation between tracks and echoes is rated, categorized into levels 1, 2, and 3 based on the normalized distance g value, from smallest to largest. For all tracks of a target, if an echo has a correlation level of 1 with a track, then if there are other tracks of the same target with a correlation level of 2 or 3 with that echo, the latter's correlation is removed. If two tracks have a correlation level of 1 with a track and an echo, then both correlations are retained. This includes the following sub-steps, such as... Figure 2 As shown: S41. The correlation between the track and the echo is calculated according to the normalized distance g. ij The values ​​are divided into levels 1, 2, and 3 from smallest to largest; S42. Extract one of the target's tracks; S43. Extract a radar echo associated with the track and determine whether the association between the track and the radar echo is level 1; S44. Re-extract another track of the target and determine whether the re-extracted track is associated with the radar echo extracted in step S43; S45. Determine whether the track extracted in step S42 has any associated echoes; S46. Determine whether the target has any other tracks.

[0047] Specifically, step S43 includes the following sub-steps: when the association between the track and the radar echo is a level 1 association, step S44 is executed; when the association between the track and the radar echo is not a level 1 association, step S43 is executed repeatedly.

[0048] The S44 step specifically includes the following sub-steps: when the association between the re-extracted track and the radar echo is a level 1 association, repeat step S44; when the association between the re-extracted track and the radar echo is not a level 1 association, remove the association and determine whether there are any remaining tracks, then execute step S45.

[0049] The S45 step specifically includes the following sub-steps: when the track extracted in the S42 step still has associated echoes, repeat step S43; when the track extracted in the S42 step does not have associated echoes, execute step S46.

[0050] The S45 step specifically includes the following sub-steps: if there are other tracks for the target, repeat step S42; if there are no other tracks for the target, execute step S5.

[0051] In step S5, if a track is associated with only one echo or has no echo, the track is updated. If a track is associated with M (M>1) echoes, the track is updated using one echo, and a new track is generated as a branch using the remaining M-1 echoes. Simultaneously, the score is calculated. The score is calculated using the log-likelihood ratio, iteratively based on the score from the previous step. Different calculation methods apply to unassociated and associated echoes. The formula for calculating the track score is: Score(k+1) = Score(k) + ΔScore(k+1);

[0052] In the formula, P D β represents the radar detection probability. FT denoted as radar false alarm rate; S as information covariance; and g as normalized distance between track and associated echo.

[0053] Step S6 specifically includes the following sub-steps: S61. Set a threshold value and calculate the average track score; S62. Determine if the average score is greater than or equal to the threshold value; S63. Calculate the sum of the scores of the most recent N frames and determine if the sum of the scores of the N frames is greater than or equal to the threshold value; S64. Determine if the target track number is 1; wherein, the formula for calculating the average track score in step S61 is: AveScore=(Score(k)) / Step; the formula for calculating the sum of the scores of the N frames in step S63 is: In the above formula, Step is the number of track frames, and N ≤ Step.

[0054] Specifically, step S62 includes the following steps: when the average value is greater than or equal to the threshold value, execute step S63; when the average value is less than the threshold value, delete this track; step S63 specifically includes the following steps: when the sum of the scores of N shots is greater than or equal to the threshold value, execute step S64; when the sum of the scores of N shots is less than the threshold value, delete this track; step S64 specifically includes the following steps: when the number of target tracks is equal to 1, execute step S8; when the number of target tracks is greater than 1, execute step S7.

[0055] Step S7 specifically includes the following steps: S71. Merging similar tracks to the same target; S72. N-SCAN pruning based on radar echoes; S73. Limiting the number of tracks for a target. To avoid a target having too many tracks, the number of its tracks is limited. First, similar tracks are merged, then the duration of track branching is controlled. If the number of tracks is still too high, low-scoring tracks are deleted.

[0056] Specifically, step S71 involves, for example... Figure 3 As shown: For each target with multiple tracks, internal pruning is performed to limit the number of tracks. If the number of tracks for a target is greater than 1, similar tracks are first merged. Tracks of the target are compared pairwise in a loop, and if they are determined to be similar, the lower-scoring track is deleted. The echoes of the most recent 4 frames are compared. In the k-th frame (the current frame), if two tracks of the same target use the same echo in frames k and k-1, and one of frames k-2 and k-3 has the same echo, then these two tracks are considered similar and can be merged. During merging, the track with the lower score is deleted.

[0057] Specifically, step S72 involves, for example: Figure 4 As shown: Record the radar echo sequence number used in the most recent M frames for each track, control the branch depth to N (N<=M), compare the usage of the radar echo in the Nth frame for each track, and delete all tracks with long branches with the highest-scoring track as the benchmark.

[0058] Step S73 specifically involves limiting the number of tracks for each target. Each target is limited to a maximum of H tracks. After the pruning process, if the number of tracks for a target still exceeds the limit H, then these tracks are sorted from highest to lowest score, the top H tracks are retained, and the rest are deleted.

[0059] The S8 step is as follows: The clustering is updated iteratively. The clustering is updated based on the echo usage of the track contained in the cluster. The update is carried out in two steps. First, the divisible clusters are split into several smaller clusters. Second, all clusters are merged.

[0060] 8.1 Cluster Splitting: If a multi-objective cluster has been pruned, it needs to be split. Each time a new cluster needs to be generated, a track that has not been added to any cluster is used as the head. A track is selected from the remaining tracks. If this track shares the same objective as a track in the cluster or has a shared echo, it is added to that cluster. This process is repeated for all remaining tracks to generate a new cluster. This process continues until all tracks have been added to a cluster, at which point the original clusters are deleted.

[0061] 8.2 Cluster merging: Compare all clusters pairwise to check if they use the same echo. If so, merge the two clusters into one.

[0062] Step S9 specifically includes the following steps:

[0063] Hypothesis generation: A conflict table is created for the tracks. A search is performed, and non-conflicting tracks are combined as a hypothesis. Then, another track without a hypothesis is used as the head to search again, creating the next track. This process is repeated until all tracks have had hypotheses added to them.

[0064] Hypothesis generation follows Blackman's method. First, a conflict table is created for the tracks. A search is performed, and tracks without conflict are grouped together as a hypothesis. Then, another track without a hypothesis is used as the head to search again, creating the next track. This process is repeated until all tracks have been included with hypotheses. The number of hypotheses in each cluster is influenced by the number of tracks in the cluster and the conflict relationships between the tracks. For cluster A, there are m targets and n possible values ​​for each target. k If there are (k = 1, 2, 3…m) tracks, then the number of hypotheses formed is Num: The score for each hypothesis is derived from the tracks it contains, with the j-th hypothesis H... j The score is T i This represents the i-th track.

[0065] Calculate the global score: Based on the score for each hypothesis, calculate the global score of the track; the formula for calculating the global score is as follows:

[0066] In the formula, H j Assuming the location of this flight path, H k For all hypotheses in the cluster, score(H) j H is the hypothesis of the j-th hypothesis. j Score. If the score is below the threshold, then delete the track.

[0067] This invention employs multiple methods to control the number of tracks fed into the hypothesis generation process. Compared to other multi-hypothesis tracking designs, this approach controls the number of tracks more efficiently and effectively, preventing an explosion in the number of generated hypotheses. Furthermore, the various pruning methods in this invention exhibit good compatibility, preventing one pruning method from affecting another.

[0068] This invention limits the reuse of measurement echoes by tracks targeting the same target, thereby controlling the generation of branches. To control the number of repeated tracks within a target and avoid redundancy, this invention proposes a method for merging similar tracks targeting the same target. This method identifies similar tracks and deletes redundant tracks based on echo usage. Although named "merging," it does not perform weighting; it only needs to retain high-scoring tracks to achieve correct tracking of measurement echoes. This invention judges the branching status of tracks by echo usage and deletes one of the excessively long branches, thereby controlling the depth (branch duration). Compared to traditional N-SCAN pruning, this method has a lower computational cost.

[0069] This invention establishes a stable and efficient pruning system to effectively control the number of tracks and avoid the exponential explosion of hypotheses caused by an excessive number of tracks in clusters. This invention focuses on the number of tracks in clusters, controlling the number of tracks before generating hypotheses to prevent hypothesis explosion; the echo-based N-SCAN pruning method more efficiently and quickly controls the branch depth. After hypotheses are completed, they are used to calculate the global score, and tracks with low global scores are pruned to reduce the computational load for the next frame.

[0070] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A computational control method for multi-hypothesis tracking based on multi-method joint pruning, characterized in that, Includes the following steps: S1. Establish and update the target track database; S2. Acquire radar echo data and perform target merging; S3. Predict the track position and associate each radar echo data with the predicted track position; S4. Redundantly deassociate radar echo data with the same target track and the same correlation; S5. Update, branch, and calculate track scores for the predicted tracks; S6. Prune the predicted trajectory based on the trajectory score; S7. Prune the target's flight path; S8. Clustering Update; S9. Generate hypotheses and calculate the global score.

2. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, The S2 step specifically involves: merging the measurement echo data from multiple units of a target, finding connectivity points by analyzing the amplitude of the echoes, and using the centroid position as the measurement position of the echoes. After merging, the echo data Z(k+1) = {z1(k+1), z2(k+1), z3(k+1)...z} of the (k+1)th frame is obtained. M (k+1)}, where M is the number of echoes in this frame.

3. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, The S3 step specifically includes the following sub-steps: S31. Apply range gate limitation, where the distance r between track i and echo j is... ij The calculation formula is: In the formula, H is the measurement matrix; S32. Calculate the normalized distance g ij Before calculating the normalized distance, it is necessary to predict the motion state and covariance matrix of the trajectory; Prediction of the motion state of track i in frame k+1 An iterative update method is used, based on the motion state X of the k-th frame of the track. i (k) is calculated using the following formula: Where F is the state transition matrix; Simultaneously, the covariance of the trajectory is predicted. The predicted covariance P(k+1|k) for the (k+1)th frame is obtained by iterating from the covariance P(k|k) of the k-th frame. The calculation formula is: P(k+1|k)=F(k)·P(k|k)·F'(k)+Q(k), where Q(k) is the process noise covariance. Based on the trajectory prediction results and echo measurement results, the normalized distance g is... ij The calculation formula is: Where S is the innovation covariance matrix of the trajectory in the Kalman filtering process, which is calculated from the covariance. The calculation method is: S(k+1)=H·P(k+1│k)·H'+R(k+1), where R(k+1) is the measurement noise matrix; S33. Determine the normalized distance g ij Whether it falls within the gate: if g ij If the value is less than the set gate value, the track is considered to be associated with the measured echo.

4. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, Step S4 specifically includes the following sub-steps: S41. The correlation between the track and the echo is determined according to the normalized distance g. ij The values ​​are divided into three levels, from smallest to largest: 1, 2, and 3. S42. Extract one of the target's flight paths; S43. Extract a radar echo associated with the track and determine whether the association between the track and the radar echo is a level 1 association; S44. Re-extract another track of the target and determine whether the re-extracted track is associated with the radar echo extracted in step S43 at level 1. S45. Determine whether the track extracted in step S42 still has associated echoes; S46. Determine if the target has any other flight paths.

5. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 4, characterized in that, The S43 step specifically includes the following sub-steps: when the association between the track and the radar echo is a level 1 association, execute step S44; when the association between the track and the radar echo is not a level 1 association, repeat step S43. The S44 step specifically includes the following sub-steps: when the association between the re-extracted track and the radar echo is a level 1 association, repeat step S44; when the association between the re-extracted track and the radar echo is not a level 1 association, remove the association and determine whether there are any remaining tracks, then execute step S45. The S45 step specifically includes the following sub-steps: when the track extracted in the S42 step still has associated echoes, repeat step S43; when the track extracted in the S42 step does not have associated echoes, execute step S46. The S45 step specifically includes the following sub-steps: if there are other tracks for the target, repeat step S42; if there are no other tracks for the target, execute step S5.

6. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, The formula for calculating the track score in step S5 is: Score(k+1)=Score(k)+ΔScore(k+1); In the formula, P D β represents the radar detection probability. FT Radar false alarm rate; S is the innovation covariance, which is calculated from the covariance. The calculation method is: S(k+1)=H·P(k+1│k)·H'+R(k+1), where R(k+1) is the measurement noise matrix; g is the normalized distance between the track and the associated echo.

7. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, Step S6 specifically includes the following sub-steps: S61. Set threshold values ​​and calculate the average track score; S62. Determine whether the average value is greater than or equal to the threshold value; S63. Calculate the sum of the scores of the most recent N beats, and determine whether the sum of the scores of the N beats is greater than or equal to the threshold value; S64. Determine if the target track count is 1; The formula for calculating the average track score in step S61 is: AveScore=(Score(k)) / Step; The formula for calculating the sum of N scores in step S63 is as follows: In the above formula, Step is the number of track frames, and N ≤ Step.

8. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 7, characterized in that, The S62 step specifically includes the following steps: when the average value is greater than or equal to the threshold value, execute step S63; when the average value is less than the threshold value, delete this track; The S63 step specifically includes the following steps: when the sum of the scores of N shots is greater than or equal to the threshold value, execute step S64; when the sum of the scores of N shots is less than the threshold value, delete this track. The S64 step specifically includes the following steps: when the number of target tracks is equal to 1, execute step S8; when the number of target tracks is greater than 1, execute step S7.

9. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, Step S7 specifically includes the following steps: S71. Merge similar tracks to the same target; S72. N-SCAN pruning based on radar echo; S73. Limit the number of target tracks; Specifically, step S72 involves: recording the radar echo sequence number used in the most recent M frames for each track, controlling the branch depth to N, comparing the usage of the radar echo in the Nth frame for each track, and deleting all tracks with long branches from the track with the highest score as the benchmark.

10. The computational complexity control method for multi-hypothesis tracking based on multi-method joint pruning according to claim 1, characterized in that, The S9 step specifically includes the following steps: Hypothesis generation: Build a conflict table for the tracks, search for it, combine non-conflicting tracks as a hypothesis, then use another track that has not been included as the head to search again, build the next track, and repeat until all tracks have been included with hypotheses; Calculate the global score: Based on the score for each hypothesis, calculate the global score of the track; the formula for calculating the global score is as follows: In the formula, T i H represents the i-th track. j Assuming the location of this flight path, H k For all hypotheses in the cluster, score(H) j H is the hypothesis of the j-th hypothesis. j Score,

Images