Multi-stage trace point clustering and trajectory reconstruction method and system

Abstract

The invention provides a multi-stage trace point clustering and trajectory reconstruction method and system, and the method comprises the steps: improving the robustness in a clustering stage through introducing a time frame constraint, motion consistency verification and three-point validity verification mechanism, and guaranteeing the continuity and credibility in a trajectory-level fusion stage; meanwhile, part of clutter points should be allowed to be re-absorbed into a real track on the premise of meeting dynamic constraints, and after the abnormal clustering and re-clustering stage is completed, the remaining clutter points are re-clustered, and whether a new effective track can be formed or not is observed, so that global optimal fusion and robust recognition of multi-target track reconstruction are realized.

Description

Technical Field

[0001] This invention relates to the field of multi-target detection and trajectory recognition technology, and in particular to a multi-stage point clustering and trajectory reconstruction method and system. Background Technology

[0002] In radar detection, electronic countermeasures, and multi-sensor fusion systems, a plot is the smallest unit of information about a target in a single or multiple frame of observation, carrying key feature information such as the target's spatial position, velocity, and Doppler shift. Plot clustering and trajectory reconstruction are core steps in extracting continuous target motion information from discrete observation data, and their results directly affect the accuracy of subsequent target recognition and behavior analysis.

[0003] Existing point clustering methods typically rely on spatial proximity, velocity similarity, or temporal continuity for aggregation. However, under conditions of complex electromagnetic interference, intense target maneuvering, or lack of observation, the robustness of these methods is significantly insufficient. On the one hand, target points are affected by changes in signal-to-noise ratio, interference masking, or multipath effects, leading to sparse point distribution or increased false detection rates, making it difficult for traditional clustering algorithms (such as DBSCAN, MeanShift, or hierarchical clustering based on Euclidean distance) to accurately distinguish targets from clutter. On the other hand, in the presence of temporal discontinuities, abrupt velocity changes, or spatial drift, clustering results often exhibit track breaks, spurious clusters, or false associations, affecting the integrity of the trajectory and the stability of target identification.

[0004] Furthermore, actual observation data often contains anomalous points or fragments, such as isolated points in a single frame, sudden motion segments, or short-term drift points. Traditional methods often employ fixed threshold rejection or linear fitting correction, lacking automatic identification and re-clustering mechanisms, making it difficult to recover potential true trajectory information from clutter. Simultaneously, existing algorithms typically discard "clutter points" directly, ignoring the valuable information that some of these points may originate from target edge observations or short-term deviations from the trajectory. This prevents the initial clustering results from fully utilizing all observation data, limiting further improvements in trajectory continuity and target detection accuracy.

[0005] In multi-target, long-term observation scenarios, segmented trajectories of the same target may exist between different clustering results. Traditional trajectory stitching or trajectory-level fusion methods often rely on fixed thresholds or average velocity determination to connect trajectories, without fully considering higher-order features such as time interval constraints, local acceleration characteristics, and trajectory topological relationships. This can easily lead to mis-stitching or missing stitching problems, thereby disrupting the consistency of the global trajectory structure.

[0006] In summary, existing technologies still have the following shortcomings in complex interference and multi-target dynamic environments:

[0007] Clustering methods are not robust enough to clutter and sparse points;

[0008]  Abnormal points and broken trajectories lack automatic identification and re-clustering mechanisms;

[0009] The trajectory splicing stage does not adequately constrain temporal continuity and motion consistency.

[0010] • The potential value of clutter points is not fully utilized, leading to a decrease in trajectory integrity.

[0011] Therefore, there is an urgent need to propose a multi-stage point clustering and trajectory reconstruction method to solve the aforementioned problems. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of existing radar point clustering and trajectory reconstruction technologies. It proposes a multi-stage point clustering and trajectory reconstruction method and system. By introducing time frame constraints, motion consistency verification, and three-point validity verification mechanisms, robustness is improved in the clustering stage, and continuity and credibility are guaranteed in the trajectory-level fusion stage. At the same time, some clutter points should be allowed to be re-incorporated into the real trajectory under the premise of satisfying dynamic constraints. After the abnormal clustering and re-clustering stage is completed, the remaining clutter points are re-clustered to observe whether new effective tracks can be formed, so as to achieve global optimal fusion and robust identification of multi-target trajectory reconstruction.

[0013] The technical solution to achieve the purpose of this invention is as follows:

[0014] A multi-stage point clustering and trajectory reconstruction method is disclosed. This method can process point data from radar, sonar, or multi-sensor systems. It achieves multi-target trajectory reconstruction in high-clutter environments through multi-stage, layer-by-layer optimization. The method includes the following steps:

[0015] Step 1, Initial Clustering Stage: Based on the adaptive clustering model of spatiotemporal neighborhood density and motion consistency, the points are initially aggregated to obtain the initial target cluster set;

[0016] Step 2, Anomaly Clustering Screening Stage: Perform time continuity, velocity consistency and heading smoothness analysis on the initial target clusters to identify low-confidence clusters and classify them into anomaly cluster sets;

[0017] Step 3, Anomaly Clustering Re-clustering Stage: Construct triplet combinations for points within anomaly clusters and verify their effectiveness. Recover potential true trajectories based on reachable path generation algorithms. Unused points are used as clutter candidate points.

[0018] Step 4, Clutter Point Re-clustering Stage: Perform neighborhood space search and motion constraint verification on clutter candidate points, and re-cluster them or merge them into existing trajectory clusters;

[0019] Step 5, Trajectory-level merging and clutter re-fusion stage: Perform cross-time fusion on all trajectory clusters, determine merging feasibility based on time frame continuity, velocity consistency and triplet validity, and verify trajectory fusion point by point for the remaining clutter points, outputting the final trajectory set and clutter set.

[0020] Furthermore, the initial clustering stage in step 1 adopts an improved density-based clustering algorithm. Its density calculation considers the following factors simultaneously: spatial distance between points; time frame difference between adjacent points; velocity vector difference; heading angle change, etc., and fuses different neighborhood features through an exponential decay weight function to improve the sensitivity of clustering to motion trends.

[0021] Furthermore, in step 1, the points are modeled as vectors containing multi-dimensional features such as spatial location, velocity, heading, time frame, and RCS intensity, and each feature dimension is normalized to ensure the comparability of density nearest neighbor calculations.

[0022] Furthermore, step 2 performs a cluster-by-cluster analysis on the initial clusters based on the following three indicators:

[0023] (1) Time frame continuity index: If the maximum frame interval within a cluster exceeds the threshold, it is determined to be an abnormal cluster;

[0024] (2) Speed ​​consistency index: If the speed change exceeds the threshold, it is judged as an abnormal cluster;

[0025] (3) Heading smoothness index: If the second difference of the heading angle exceeds the threshold, it is determined to be an abnormal cluster.

[0026] Furthermore, step 3 involves constructing and validating the point triples, specifically including:

[0027] (1) The time difference between adjacent points satisfies:

[0028]

[0029] (2) Velocity continuity satisfies:

[0030]

[0031] (3) The rationality of acceleration is satisfied:

[0032]

[0033] (4) The smoothness of the heading angle satisfies:

[0034]

[0035] in, , , , These are the time difference threshold, velocity difference threshold, acceleration difference threshold, and heading angle difference threshold, respectively. , This represents the time corresponding to the i-th and i+1-th points. , This represents the velocity corresponding to the i-th and i+1-th points. , This represents the heading angle corresponding to the i-th and i+1-th points. , , This represents the acceleration corresponding to the (i-1), i, and i+1th points;

[0036] Triples that meet all the conditions are used to construct feasible trajectory segments.

[0037] Furthermore, by constructing a triple reachability graph and employing dynamic programming or depth-first search algorithms, the longest feasible path in the graph is found to recover the potential trajectory structure in the anomalous cluster.

[0038] Furthermore, step 4 performs the following judgment for each clutter point:

[0039] (1) Whether the time difference with the candidate trajectory satisfies the time window constraint;

[0040] (2) Whether the triples in the trajectory after addition still satisfy the validity verification.

[0041] If the conditions are met, the clutter point is included in the trajectory cluster; otherwise, it is retained as a clutter candidate point.

[0042] Furthermore, if multiple clutter points satisfy the conditions of no temporal conflict, spatial proximity, and the ability to form a continuous triplet, a new short track cluster is automatically generated.

[0043] Furthermore, in step 5, the merging of the two trajectory clusters must simultaneously satisfy the following:

[0044] (1) The time frames do not conflict with each other;

[0045] (2) Any three consecutive points after merging satisfy the triplet validity;

[0046] (3) The deviation between the predicted speed and the actual speed does not exceed the allowable threshold.

[0047] After merging, each remaining clutter point is checked to see if it meets the trajectory absorption condition. All points that cannot be integrated into the trajectory constitute the final real clutter set.

[0048] A multi-stage point clustering and trajectory reconstruction system includes:

[0049] The initial clustering module performs preliminary aggregation of points based on an adaptive clustering model that considers spatiotemporal neighborhood density and motion consistency, resulting in an initial set of target clusters.

[0050] The anomaly clustering screening module is used to perform time continuity, velocity consistency and heading smoothness analysis on the initial target clusters, identify low-confidence clusters and classify them into anomaly cluster sets;

[0051] The anomaly clustering and re-clustering module is used to construct triplet combinations of points within anomaly clusters and verify their validity. It recovers potential true trajectories based on the reachable path generation algorithm, and unused points are used as clutter candidate points.

[0052] The clutter point re-clustering module is used to perform neighborhood space search and motion constraint verification on clutter candidate points, and re-cluster them or merge them into existing trajectory clusters;

[0053] The trajectory-level merging and clutter re-fusion module is used to perform cross-time fusion of all trajectory clusters. It determines the merging feasibility based on time frame continuity, velocity consistency, and triplet validity, and verifies the trajectory fusion of the remaining clutter points one by one, outputting the final trajectory set and clutter set.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] (1) This invention significantly improves the stability of point clustering and the temporal consistency of trajectory reconstruction through a multi-stage adaptive clustering and recursive correction mechanism. It does not rely on the number of prior targets or fixed model parameters and has strong versatility and transferability. This method can be widely applied to target detection and trajectory recognition tasks in radar, sonar and multi-sensor fusion environments, realize high-precision multi-target trajectory reconstruction in complex interference environments, and effectively improve track continuity, clustering robustness and clutter suppression capability.

[0056] (2) The multi-stage point clustering and trajectory reconstruction method proposed in this invention achieves stable target identification and continuous trajectory generation in environments with high clutter, weak points, and complex interference by constructing a hierarchical processing mechanism at the point, cluster, and trajectory levels. The method introduces multi-dimensional features such as time continuity constraints, triplet validity judgment, and motion trend consistency, making point clustering more consistent with the actual movement of the target and effectively avoiding common problems in traditional algorithms such as mis-clustering, false trajectories, and trajectory breakage. Simultaneously, through abnormal clustering reconstruction and clutter point re-clustering mechanisms, this invention can adaptively correct erroneous clustering and fully utilize weak points or points temporarily detached from the trajectory to achieve dynamic trajectory repair and enhancement. In the trajectory fusion stage, robust splicing of cross-segment trajectories is achieved through velocity continuity, prediction bias, and temporal constraints, making the reconstructed trajectory smoother in space and more consistent in time. Compared with existing methods, this method improves trajectory integrity and has universality and broad engineering application value. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the technical invention of a multi-stage point clustering and trajectory reconstruction method. Detailed Implementation

[0058] This invention provides a multi-stage point clustering and trajectory reconstruction method. Addressing the issues of insufficient robustness and poor target trajectory continuity in traditional clustering methods caused by sparse target points, dense clutter, and complex interference in radar detection environments, this invention proposes a point clustering and trajectory reconstruction framework that integrates a multi-stage hierarchical optimization mechanism. Based on the spatiotemporal feature constraints of points, this invention constructs a hierarchical optimization structure at the point, cluster, and trajectory levels, introducing an automatic outlier screening and re-clustering mechanism to achieve dynamic optimization and adaptive association throughout the entire process from point to trajectory. This significantly improves target detection accuracy and trajectory reconstruction integrity in cluttered environments. The specific process is as follows: Figure 1 The process mainly includes five stages: initial clustering, anomaly clustering screening, anomaly clustering re-clustering, clutter point re-clustering, trajectory-level fusion, and clutter re-fusion. This embodiment breaks down and explains the processing flow, mathematical basis, algorithm logic, and key parameters of each stage layer by layer, so as to fully demonstrate the technical essence and feasibility of the invention.

[0059] The following are the detailed implementation steps.

[0060] 1. Initial clustering stage

[0061] This stage employs the adaptive clustering algorithm based on spatiotemporal density and motion consistency proposed in this invention. Compared to the traditional clustering method that only uses Euclidean distance, this method integrates multi-dimensional features of spatial neighborhood, temporal constraints, and motion parameters (velocity, heading angle), enabling the initial clustering to take into account both temporal continuity and physical motion rationality.

[0062] (1) Multidimensional modeling of point data

[0063] The original point trace not only contains the target's spatial coordinate information, but also contains multi-dimensional parameters that reflect the laws of physical motion, such as velocity, heading angle, acceleration, and RCS signal strength. These dimensions all have physical meaning under a specific radar system.

[0064] This invention models point traces as multi-dimensional vectors:

[0065]

[0066] The functions of each feature include:

[0067] Spatial coordinates Determining the target's position in the three-dimensional observation space is the core of distance calculation.

[0068] Time frame The temporal structure of the points is an important basis for the continuity of the trajectory.

[0069] Speed With heading angle : Reflects the instantaneous motion state of the target and is used to judge motion consistency.

[0070] Acceleration It can enhance the modeling accuracy of maneuvering targets.

[0071] RCS and other attributes: can provide additional dimensions of target separability.

[0072] To avoid inconsistencies in dimensions caused by differences in dimensionality, this invention employs normalization during the clustering process to make features such as velocity and location comparable, while also facilitating the expression of the exponential decay term of the density function.

[0073] (2) Spatiotemporal neighborhood construction

[0074] In order to construct physically meaningful clustered neighborhoods, this invention considers three dimensions simultaneously: spatial distance, temporal proximity, and velocity difference.

[0075] Spatial distance function:

[0076]

[0077] This distance reflects the actual physical distance between two points in three-dimensional space and directly determines whether the points may belong to the same target.

[0078] Time neighborhood constraints:

[0079] Considering the sampling period of the detection system, for the same target, its trajectory points appear in consecutive time frames. Therefore, this invention defines:

[0080]

[0081] This constraint can effectively filter out points that are too far apart across frames and cannot form a trajectory relationship.

[0082] • Speed ​​consistency weight:

[0083] In complex and noisy environments, relying solely on spatial proximity may lead to a large number of noisy points being incorrectly clustered. This invention introduces a velocity similarity enhancement term in the initial clustering stage:

[0084]

[0085] This is to reduce the contribution of points with large speed differences to clustering.

[0086] (3) Improved DBSCAN clustering algorithm based on multiple feature densities

[0087] The density index used in this embodiment considers both spatial and velocity dimensions, and the comprehensive density is defined as:

[0088]

[0089] in, This is a spatial scale parameter used to control distance sensitivity; This is a velocity scaling parameter used to smooth out velocity differences; A set of points that satisfy spatiotemporal neighborhood constraints. Compared to the fixed... The invention achieves neighborhood flexibility through an exponentially decaying density function, which makes the points closer together and with more consistent speeds contribute more to the density and improves the stability of clustering decisions.

[0090] (4) Initial clustering output

[0091] The algorithm outputs the initial cluster set:

[0092]

[0093] 2. Anomaly Clustering Screening Stage

[0094] The initial clustering results may contain anomalous clusters caused by clutter, missing probes, or abrupt noise changes. This stage aims to conduct a systematic quality assessment of the clusters, using multiple indicators such as temporal continuity, velocity smoothness, and heading change trends to identify low-confidence clusters and provide a target set for subsequent re-clustering stages.

[0095] (1) Time continuity analysis

[0096] The essential characteristics of trajectory data are its temporal order and gradual progression. Therefore, this invention calculates the frame interval between adjacent points after sorting each cluster by time:

[0097]

[0098] If the maximum interval satisfies the following condition:

[0099]

[0100] Clusters are then considered to have temporal breaks. For example, when the detection environment is obstructed or the radar SNR is low, the actual trajectory may be broken, but the original initial cluster may still group it into a cluster, so it must be stripped and reprocessed at this stage.

[0101] (2) Speed ​​consistency detection

[0102] The velocity changes of a real target usually have a certain degree of continuity. This invention defines a velocity jump variable:

[0103]

[0104] If the variable exceeds the threshold, it is considered that there are noise points mixed in or different targets are incorrectly clustered.

[0105] (3) Heading continuity detection (based on second-order difference)

[0106] Changes in heading angle reflect the target's turning and maneuvering characteristics. Since first-order differences may not be sensitive enough to detect sudden changes, this invention uses second-order differences (for acceleration):

[0107]

[0108] If this amount is greater than the threshold This indicates that the cluster's motion pattern has undergone a drastic change.

[0109] (4) Output of abnormal cluster set

[0110] All clusters that satisfy any one of the anomalous conditions are added to the anomalous set:

[0111]

[0112] The remaining clusters are considered high-confidence clusters and directly proceed to the subsequent trajectory stitching process.

[0113] 3. Abnormal clustering and re-clustering stage

[0114] The abnormal re-clustering stage primarily addresses the problem of chaotic intra-cluster structure caused by insufficient robustness of the initial clustering algorithm. This stage focuses on the "rationality of triplet motion," recovering the potential true target trajectory by reconstructing time-feasible and smooth trajectory segments.

[0115] (1) Construction and significance of triples

[0116] For all traces in the outlier cluster, sort them by time and construct non-overlapping trace triples:

[0117]

[0118] The functions of a triplet include:

[0119]  Modeling local motion trends (first-order and second-order changes in position).

[0120] Verification of the continuity of speed and heading changes.

[0121]  As the basic constraint unit for trajectory construction.

[0122] If the triple satisfies all constraints, then the three points are considered to belong to a small segment of the true trajectory.

[0123] (2) Formula for validating triplet validity

[0124] This invention determines the rationality of triplet motion from the following perspectives:

[0125] Time frame continuity constraint

[0126] Define the triplet time frame:

[0127]

[0128] It also requires that the interval between adjacent frames does not exceed a threshold. :

[0129]

[0130] in, This constraint can be set according to the radar or sensor sampling period, typically 1 to 2 frames. This constraint ensures that the triples are continuous or nearly continuous in time, preventing points that are too far apart across frames from being incorrectly combined into trajectory segments.

[0131]  Velocity continuity constraint

[0132] The actual motion speed should not exhibit abnormal jumps between adjacent frames. Definition:

[0133]

[0134] in To prevent small constants divided by zero, a speed jump threshold is set to... ,like:

[0135]

[0136] The speed change is then reasonable.

[0137] • Acceleration rationality constraints (based on second-order position difference)

[0138] Using second-order differences to measure the magnitude of local acceleration:

[0139]

[0140] In kinematics, the target's acceleration does not fluctuate drastically, therefore it is required that:

[0141]

[0142] Smoothness constraint of heading change

[0143] Sudden changes in heading angle often originate from clutter interference:

[0144]

[0145] set up Let the heading angle jump threshold be:

[0146]

[0147] Therefore, the change in direction is considered reasonable.

[0148] (3) Trajectory reconstruction of heterogeneous clusters (greedy longest trajectory search based on graph theory)

[0149] Based on the above triplet construction rules, exhaustively enumerate the point triplets that meet the requirements in the heterogeneous clusters, and use them as the three-point shortest path construction rules to further construct the triplet reachable path map.

[0150] Find the longest feasible path in the triple reachable path graph using depth-first search (DFS) or dynamic programming methods:

[0151]

[0152] in A sequence of points along the path. This is the path length. The path represents an anomaly cluster. The most reliable trajectory segment. After finding the longest path, remove the points on the path from the anomaly cluster. Removed from:

[0153]

[0154] Repeat the above process of constructing a triplet reachable graph and searching for paths until no longer a length-based reachable graph can be formed. The trajectory. Ultimately, all unused points are considered as clutter candidates, forming the clutter set:

[0155]

[0156] These clutter points will be further processed in the subsequent clutter re-clustering stage.

[0157] 4. Clutter point re-clustering stage

[0158] The goal of the clutter point re-clustering stage is to reuse isolated points while satisfying the motion laws, thereby further improving the integrity and robustness of the trajectory.

[0159] If any clutter point forms a track, the following condition must be met:

[0160] The time frames of any two points are not repeated;

[0161] A short track consisting of any three adjacent points in a track satisfies the triplet validity verification formula.

[0162] These clutter points can then form new small trajectory clusters:

[0163]

[0164] 5. Trajectory-level merging and clutter re-fusion stage

[0165] This stage involves a global merging of all trajectory clusters across time and motion, which is a key step in achieving the final continuous trajectory output.

[0166] (1) Logic for merging two clusters (with multiple constraints)

[0167] Given two track clusters and This invention designs the following constraints to determine whether two track clusters can be merged:

[0168] Time frames do not conflict

[0169]

[0170] Triple validity verification

[0171] Assumption ,Will After arranging the dots in the time frame sequence, randomly select three consecutive dots:

[0172]

[0173] It satisfies the triplet validity verification formula.

[0174] If track cluster and If the above constraints are met, then the track cluster and They can be merged into a single track cluster. .

[0175] (2) Global Iterative Merging Process

[0176] A bottom-up trajectory cluster merging strategy is adopted:

[0177] a) Perform a mergeability check on all cluster pairs.

[0178] b) Perform a merge on clusters that meet the conditions.

[0179] c) Update the cluster set.

[0180] d) Repeat steps a) to c) until the system converges.

[0181] The final output is a complete set of trajectories:

[0182]

[0183] (3) Clutter points in fusion and final processing

[0184] For the remaining clutter points, each point is assessed to determine whether it can be fused with an existing track cluster. The criteria for this assessment include time frame non-repetition and a three-point validity verification condition. Clutter points that meet these fusion conditions are dynamically inserted into their corresponding track clusters, enhancing track integrity. Points that remain unfused after this process are considered genuine clutter, ultimately forming a complete set of clutter points. .

[0185] Based on the above method, this embodiment also provides a multi-stage point clustering and trajectory reconstruction system, including five core technology modules:

[0186] 1. Initial clustering module

[0187] An adaptive clustering algorithm based on spatiotemporal neighborhood density is used to initially aggregate the points, forming the first layer structure of the target cluster. Temporal continuity constraints and spatial proximity constraints are introduced at this stage to ensure that the initial clustering results reflect both the spatial distribution characteristics of the targets and maintain consistency in motion direction and velocity, providing a high-quality clustering foundation for subsequent stages.

[0188] 2. Anomaly Clustering Filtering Module

[0189] The initial clustering results are used to identify potentially outlier clusters. By employing metrics such as statistical bias detection, trajectory smoothness analysis, and silhouette coefficient evaluation, the stability of the internal structure of each target cluster is determined. Outlier segments are then labeled or removed, providing data support for subsequent dynamic correction and trajectory reconstruction. This stage effectively reduces the impact of mis-clustering on overall track quality.

[0190] 3. Anomaly Clustering and Re-clustering Module

[0191] For low-confidence target clusters, an anomalous trajectory reconstruction algorithm based on triplet rules is used for secondary decomposition and reconstruction to recover the potential target trajectory structure. Specific methods include:

[0192] Within the anomalous cluster, based on the combination of triplet points, all possible point triplets are enumerated, and each triplet is checked for spatiotemporal rationality and motion characteristic constraints to filter out combinations that do not conform to the laws of physical motion.

[0193] • Based on a greedy algorithm, prioritize searching for the longest track to ensure the continuity and integrity of the reconstructed track;

[0194] For isolated points that cannot form tracks, they are marked as clutter points to provide data for the next stage of re-clustering.

[0195] This mechanism can self-correct initial clustering errors under high clutter conditions, improving the robustness and interpretability of clustering.

[0196] 4. Clutter Point Re-clustering Module

[0197] After re-clustering the anomaly clusters, the remaining clutter points are further clustered. Through historical trajectory prediction, neighborhood search, and motion consistency constraints, isolated points are attempted to be aggregated into new tracks.

[0198] • Check whether clutter points can be merged with existing tracks, conditions include that the time frames are not repeated, the interval between adjacent time frames does not exceed a set threshold (e.g., the maximum interval is 2 frames), and the three-point trajectory verification function returns a reasonable result;

[0199] For clutter points that cannot be merged into existing tracks, attempt to generate new tracks to ensure that potential target points are not missed;

[0200] This mechanism enables dynamic utilization of clutter points, improving trajectory continuity while reducing the false positive rate.

[0201] 5. Trajectory-level merging and clutter re-fusion module

[0202] At the trajectory level, short tracks generated by multi-stage clustering are correlated and globally fused across time. The system comprehensively considers time frame order, velocity consistency, trajectory topology, and the dynamic effectiveness of the three points, employing a trajectory stitching algorithm based on time constraints to merge segmented trajectories into a complete target track. Simultaneously, clutter points are allowed to be re-incorporated under the conditions of motion consistency and inter-frame continuity, achieving globally optimal reconstruction of the target trajectory and intelligent clutter suppression.

[0203] This invention achieves self-correction and dynamic optimization of initial clustering errors under high clutter and complex interference environments by introducing an anomaly clustering two-level reconstruction mechanism, a clutter point re-clustering and fusion judgment mechanism, and a multi-stage trajectory-level fusion algorithm. The anomaly clustering two-level reconstruction mechanism can perform secondary decomposition and re-clustering of low-confidence clusters under high clutter conditions, improving the interpretability and robustness of clustering; the clutter point re-clustering and fusion judgment mechanism effectively suppresses false trajectories and isolated point interference through dual constraints of motion features and neighborhood structure; the multi-stage trajectory-level fusion algorithm achieves temporal continuity and dynamic consistency reconstruction at the target cluster level, making trajectory connections smoother, target recognition more accurate, and ensuring the integrity and physical rationality of the final trajectory.

[0204] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-stage point clustering and trajectory reconstruction method, characterized in that: include: Step 1, Initial Clustering Stage: Based on the adaptive clustering model of spatiotemporal neighborhood density and motion consistency, the points are initially aggregated to obtain the initial target cluster set; Step 2, Anomaly Clustering Screening Stage: Perform time continuity, velocity consistency and heading smoothness analysis on the initial target clusters to identify low-confidence clusters and classify them into anomaly cluster sets; Step 3, Anomaly Clustering Re-clustering Stage: Construct triplet combinations for points within anomaly clusters and verify their effectiveness. Recover potential true trajectories based on reachable path generation algorithms. Unused points are used as clutter candidate points. Step 4, Clutter Point Re-clustering Stage: Perform neighborhood space search and motion constraint verification on clutter candidate points, and re-cluster them or merge them into existing trajectory clusters; Step 5, Trajectory-level merging and clutter re-fusion stage: Perform cross-time fusion on all trajectory clusters, determine merging feasibility based on time frame continuity, velocity consistency and triplet validity, and verify trajectory fusion point by point for the remaining clutter points, outputting the final trajectory set and clutter set.

2. The multi-stage point clustering and trajectory reconstruction method according to claim 1, characterized in that: The initial clustering stage employs a density-based clustering algorithm, whose density calculation considers the following factors: spatial distance between points, time frame difference between adjacent points, velocity vector difference, and heading angle change, and fuses different neighborhood features through an exponential decay weight function.

3. The multi-stage point clustering and trajectory reconstruction method according to claim 1, characterized in that: In step 1, the point trace is modeled as a multi-dimensional feature vector containing spatial position, velocity, heading, time frame, and RCS intensity, and each feature dimension is normalized.

4. The multi-stage point clustering and trajectory reconstruction method according to claim 1, characterized in that: Step 2 involves analyzing the time continuity, velocity consistency, and heading smoothness of the initial target cluster, specifically as follows: Temporal frame continuity index: If the maximum frame interval within a cluster exceeds the threshold, it is determined to be an abnormal cluster; Speed ​​consistency index: If the speed change exceeds the threshold, it is judged as an abnormal cluster; Heading smoothness index: If the second difference of the heading angle exceeds the threshold, it is identified as an abnormal cluster.

5. The multi-stage point clustering and trajectory reconstruction method according to claim 1, characterized in that: Step 3 includes validity verification, which includes: The time difference between adjacent points satisfies: ； Velocity continuity satisfies: ； The acceleration rationality is satisfied: ； The heading angle smoothness satisfies: ； in, , , , These are the time difference threshold, velocity difference threshold, acceleration difference threshold, and heading angle difference threshold, respectively. , This represents the time corresponding to the i-th and i+1-th points. , This represents the velocity corresponding to the i-th and i+1-th points. , This represents the heading angle corresponding to the i-th and i+1-th points. , , This represents the acceleration corresponding to the (i-1), i, and i+1th points; Triples that meet all the conditions are considered valid data and are used to construct feasible trajectory segments.

6. The multi-stage point clustering and trajectory reconstruction method according to claim 5, characterized in that: Step 3, which involves restoring the potential true trajectory based on the reachable path generation algorithm, specifically involves constructing a triplet reachable graph and using dynamic programming or depth-first search algorithms to find the longest feasible path in the graph in order to restore the potential trajectory structure in the anomaly cluster.

7. The multi-stage point clustering and trajectory reconstruction method according to claim 1, characterized in that: Step 4, motion constraint verification, specifically includes the following judgments: Does the time difference with the candidate trajectory satisfy the time window constraint? Does the triplet in the trajectory still satisfy the validity verification after addition? If the above conditions are met, the clutter point is included in the trajectory cluster; otherwise, it is retained as a clutter candidate point.

8. The multi-stage point clustering and trajectory reconstruction method according to claim 7, characterized in that: If multiple clutter points satisfy the conditions of no temporal conflict, spatial proximity, and the ability to form a continuous triplet, a new short track cluster is automatically generated.

9. The multi-stage point clustering and trajectory reconstruction method according to claim 1, characterized in that: Step 5, the determination of merging feasibility, includes: The time frames do not conflict; After merging, any three consecutive points satisfy the triplet validity; The deviation between the predicted speed and the actual speed does not exceed the allowable threshold. After merging, each remaining clutter point is checked to see if it meets the trajectory absorption condition. All points that cannot be integrated into the trajectory constitute the final real clutter set.

10. A multi-stage point clustering and trajectory reconstruction system implementing the method of any one of claims 1-9, characterized in that: include: The initial clustering module performs preliminary aggregation of points based on an adaptive clustering model that considers spatiotemporal neighborhood density and motion consistency, resulting in an initial set of target clusters. The anomaly clustering screening module is used to perform time continuity, velocity consistency and heading smoothness analysis on the initial target clusters, identify low-confidence clusters and classify them into anomaly cluster sets; The anomaly clustering and re-clustering module is used to construct triplet combinations of points within anomaly clusters and verify their validity. It recovers potential true trajectories based on the reachable path generation algorithm, and unused points are used as clutter candidate points. The clutter point re-clustering module is used to perform neighborhood space search and motion constraint verification on clutter candidate points, and re-cluster them or merge them into existing trajectory clusters; The trajectory-level merging and clutter re-fusion module is used to perform cross-time fusion of all trajectory clusters. It determines the merging feasibility based on time frame continuity, velocity consistency, and triplet validity, and verifies the trajectory fusion of the remaining clutter points one by one, outputting the final trajectory set and clutter set.

Images