Multi-target resolution method and system based on trace feature condensation

Abstract

This invention provides a multi-target discrimination method and system based on point feature aggregation, comprising: acquiring point data output from a radar signal processor and preprocessing the point data to obtain effective point data; extracting multi-dimensional point features from the effective point data and standardizing the multi-dimensional point features to obtain a standardized feature set; calculating target state parameters based on real-time sensor data, and dynamically calculating the aggregation weight of each feature in the standardized feature set based on the target state parameters; calculating the correlation between the current point and each candidate target through feature aggregation based on the aggregation weight and the standardized feature set, and completing the target discrimination of the current point based on the correlation, obtaining the discrimination result; verifying the discrimination result, and performing reverse optimization of the aggregation weight based on the verification result, thereby achieving accurate discrimination and target identity association of overlapping, adjacent, and interfered point traces, improving the accuracy and adaptability of multi-target discrimination.

Description

Technical Field

[0001] This invention relates to the field of radar data processing technology, and in particular to a multi-target resolution method and system based on point feature aggregation. Background Technology

[0002] In multi-target monitoring scenarios, radar converts the signals reflected or radiated by targets into discrete point data. When multiple targets are close together or are subject to environmental interference, the point data is prone to overlap, proximity, or dispersion, leading to difficulties in multi-target resolution.

[0003] Existing multi-target discrimination methods have significant shortcomings: First, they rely on single spatial features, resulting in poor robustness and susceptibility to misjudgment when the target spacing is less than the sensor's resolution limit or when strong clutter interference exists. Second, point association lacks dynamic adaptability, employing fixed association rules that cannot adapt to dynamic changes in target motion states. Third, feature fusion methods are simple and fail to fully exploit the correlation and complementarity between features, leading to low information utilization. Fourth, they suffer from high computational complexity and poor real-time performance in complex scenarios, making it difficult to meet the millisecond-level data update requirements of radar systems.

[0004] These problems result in existing methods having insufficient resolution accuracy and weak adaptability in complex scenarios, failing to meet the practical application needs of multi-target monitoring systems. Summary of the Invention

[0005] This invention provides a multi-target discrimination method and system based on point feature aggregation, which solves the problems of insufficient discrimination accuracy and weak adaptability in complex scenes, and cannot meet the actual application needs of multi-target monitoring systems.

[0006] On one hand, the present invention provides a multi-target discrimination method based on point feature aggregation, which includes: Collect point data output from the radar signal processor and preprocess the point data to obtain valid point data; Multidimensional trace features are extracted from the valid trace data and the multidimensional trace features are standardized to obtain a standardized feature set; The target state parameters are calculated based on real-time sensor data, and the agglomeration weight of each feature in the standardized feature set is dynamically calculated based on the target state parameters. Based on the agglomeration weight and the standardized feature set, the correlation degree between the current point and each candidate target is calculated through feature agglomeration, and the target discrimination of the current point is completed according to the correlation degree to obtain the discrimination result. The resolution results are verified, and the agglomeration weights are optimized in reverse based on the verification results.

[0007] On the other hand, the present invention also provides a multi-target resolution system based on dot feature aggregation, which includes: The preprocessing module is used to collect the point data output by the radar signal processor and preprocess the point data to obtain valid point data. The standardization processing module is used to extract multi-dimensional trace features from the valid trace data and perform standardization processing on the multi-dimensional trace features to obtain a standardized feature set. The determination module is used to calculate the target state parameters based on real-time sensor data, and dynamically calculate the agglomeration weight of each feature in the standardized feature set based on the target state parameters. The discrimination module is used to calculate the correlation between the current point and each candidate target through feature aggregation based on the aggregation weight and the standardized feature set, and to perform target discrimination of the current point according to the correlation, so as to obtain the discrimination result. An optimization module is used to verify the resolution results and perform reverse optimization on the agglomeration weights based on the verification results.

[0008] The present invention provides a multi-target discrimination method and system based on point feature aggregation. By collecting radar point data and preprocessing it, extracting multi-dimensional point features and standardizing them, dynamically calculating feature aggregation weights based on target state parameters, calculating the correlation between point and candidate target through feature aggregation to complete target discrimination, verifying the discrimination results and optimizing the weights in reverse, it achieves accurate discrimination and target identity association for overlapping, adjacent and interfered point patterns, and improves the accuracy and adaptability of multi-target discrimination. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the multi-target resolution method based on dot feature aggregation provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a multi-target resolution system based on dot feature aggregation provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0012] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0013] Figure 1 This is a flowchart illustrating the multi-target resolution method based on point feature aggregation provided in an embodiment of the present invention.

[0014] like Figure 1 As shown, the multi-target discrimination method based on point feature aggregation provided in this embodiment of the invention mainly includes the following steps: 101. Collect the point data output by the radar signal processor and preprocess the point data to obtain valid point data; In a specific implementation, point data output from the radar signal processor can be acquired. This point data can include basic feature data such as spatial location, timestamp, and motion parameters, as well as signal feature data such as signal strength, signal-to-noise ratio (SNR), and signal feature vectors. Preprocessing is crucial for ensuring data quality. First, Kalman filtering or wavelet transform algorithms are used to filter out random noise such as sensor measurement errors and environmental clutter. Then, point data from different times are adjusted to a unified time window based on timestamp interpolation to ensure time consistency. Finally, a SNR threshold is set, and invalid spurious points below the threshold are removed to obtain clean and valid point data.

[0015] 102. Extract multidimensional trace features from the valid trace data and standardize the multidimensional trace features to obtain a standardized feature set; In a specific implementation process, spatial clustering features, motion consistency features, signal correlation features, temporal continuity features, and signal strength stability features can be extracted from effective point data to form multidimensional point features. Then, the Min-Max standardization method is used to transform each feature value to the [0,1] interval, eliminating the dimensional differences between different features and forming a standardized feature set, which lays the foundation for subsequent weight calculation and correlation analysis.

[0016] The standardized feature set includes at least one of spatial clustering features, motion consistency features, signal correlation features, temporal continuity features, and signal strength stability features.

[0017] Specifically, multidimensional features are extracted from the preprocessed trace data to construct a standardized trace feature set. The features are defined as follows: f1: Spatial clustering feature, represented by the Euclidean distance between the point trace and the center of the candidate target, as shown in equation (1): (1); in, For the spatial location of the dot, The initial center positions of the candidate targets (obtained through initial clustering). The smaller the value, the stronger the spatial correlation between the dot and the candidate target; The motion consistency feature is characterized by the matching degree between the motion parameters of the point and the motion model of the candidate target. It is achieved by calculating the cosine similarity between the velocity vector of the point and the velocity vector of the candidate target, as shown in equation (2). (2); in, For the velocity vector of the point, This represents the velocity vector of the candidate target. , The closer it is to 1, the stronger the motion consistency.

[0018] The signal association feature is represented by the cosine similarity between the feature vector of the trace signal and the candidate target signal template, as shown in equation (3). (3); in It is the feature vector of the trace signal (such as Doppler frequency shift, spectral features). The signal template for candidate targets (obtained through historical data learning). The closer it is to 1, the stronger the correlation of the signal features.

[0019] The time continuity characteristic is calculated based on the continuity of the time series of the trace timestamp and the historical traces of the candidate target, and the formula is (4): (4); Where t is the current timestamp. Let k be the timestamp of the most recent point of the candidate target, and k be the decay coefficient (k>0). The smaller, The closer it is to 1, the stronger the temporal continuity.

[0020] Signal strength stability characteristics, calculating the signal strength of a point trace. Compared with the historical signal strength average of candidate targets The deviation rate is given by formula (5): (5); The smaller the deviation rate, The closer it is to 1, the stronger the signal strength stability.

[0021] The extracted multidimensional features are standardized by using Min-Max standardization to transform each feature value to the [0,1] interval, eliminating dimensional differences. The formula is (6): (6); in, Let be the minimum value of the i-th feature. Let be the maximum value of the i-th feature. These are the standardized feature values.

[0022] 103. Calculate the target state parameters based on real-time sensor data, and dynamically calculate the agglomeration weight of each feature in the standardized feature set based on the target state parameters; In a specific implementation process, target state parameters include at least one of target density, target motion complexity, and interference intensity. Target density (number of targets per unit space), target motion complexity (ratio of target acceleration standard deviation to average velocity), and interference intensity (ratio of clutter spot counts to effective spot counts) can be obtained based on real-time sensor data. An improved fuzzy hierarchical analysis method is used to assign agglomeration weights to each feature based on the comparison results between the values ​​of the target state parameters and the corresponding preset parameter thresholds, ensuring optimal weight allocation under different scenarios. The preset parameter thresholds are determined through experimental calibration based on typical scenarios of radar multi-target monitoring (such as formation flying, dense vehicle traffic, and strong clutter interference).

[0023] Specifically, if only the target density is higher than the preset density threshold, the agglomeration weight of the spatial aggregation feature is increased; in this case, it indicates that the target is dense in the current scene and spatial location differentiation is crucial, and the weight of the spatial aggregation feature is adjusted to the range of [0.3, 0.4].

[0024] If the target motion complexity is higher than the preset complexity threshold, the cohesion weight of the motion consistency feature is increased. In this case, it indicates that the target motion state is variable and the role of motion features needs to be strengthened. The motion consistency feature weight is adjusted to the range of [0.3, 0.4].

[0025] If the interference intensity is higher than the preset interference threshold, the cohesion weight of the signal correlation feature is increased; in this case, the clutter interference is severe and the signal feature has strong anti-interference ability, so the weight of the signal correlation feature is adjusted to the range of [0.3, 0.4].

[0026] If at least two of the target state parameters, namely the target density, the target motion complexity, and the interference intensity, are higher than the corresponding thresholds, the dominant feature of the current scene is determined, and the cohesion weight of the dominant feature is increased.

[0027] Specifically, the process of determining the dominant features of the current scene may include: a1. Calculate the over-threshold ratio of the state parameters of the over-standard target respectively; the over-standard target state parameters include at least two of the target density, the target motion complexity, and the interference intensity; For each target state parameter that exceeds the standard (such as both target density and interference intensity exceeding the standard), its over-threshold ratio can be calculated separately. The formula can be understood as (actual measured value of parameter - preset threshold) / preset threshold. This ratio intuitively reflects the degree to which the parameter exceeds the threshold.

[0028] a2. Normalize each threshold ratio to obtain the scene influence factor of the out-of-standard target state parameters; Each threshold ratio can be normalized by dividing each threshold ratio by the sum of the threshold ratios of all exceeding parameters, thus obtaining the scene impact factor for each exceeding parameter. The larger the impact factor, the more significant the impact of the parameter on the current scene.

[0029] a3. If there is a unique maximum scene influence factor, the feature corresponding to the excess target state parameter corresponding to the unique maximum scene influence factor shall be taken as the dominant feature; a4. If there are multiple maximum scenario impact factors, obtain the historical scenario impact factors of the target state parameters exceeding the standard corresponding to each maximum scenario impact factor in the most recent N processing cycles. a5. Calculate the average historical scenario impact factor of the target state parameters exceeding the standard corresponding to each maximum scenario impact factor; a6. The feature corresponding to the target state parameter exceeding the standard corresponding to the maximum average value is taken as the dominant feature.

[0030] Specifically, if a unique maximum scene influence factor exists after normalization, for example, the scene influence factor for target density is 0.6 and the interference intensity is 0.4, then the spatial clustering feature corresponding to the target density is directly used as the dominant feature. If multiple maximum scene influence factors exist, for example, the scene influence factors for both target density and interference intensity are 0.5, then the historical scene influence factors of both are obtained within the most recent N processing cycles (the value of N can be determined according to the radar data update frequency, such as 10 cycles), and their average values ​​are calculated. If the average historical influence factor of target density is larger, then the spatial clustering feature is used as the dominant feature; otherwise, the signal correlation feature is selected.

[0031] In a specific implementation, the cohesion weights of the time continuity feature and the signal strength stability feature can be kept unchanged, and the sum of the cohesion weights of all features can be 1, with each cohesion weight value located within its respective preset interval. That is, the time continuity feature and the signal strength stability feature are used as auxiliary features, with their weights kept constant in the interval [0.1, 0.2], and the sum of all feature weights is always 1, ensuring the rationality and effectiveness of the weight allocation.

[0032] 104. Based on the agglomeration weight and the standardized feature set, calculate the correlation degree between the current point and each candidate target through feature agglomeration, and complete the target discrimination of the current point according to the correlation degree to obtain the discrimination result; In a specific implementation process, this step can be implemented in the following way: b1. Based on the effective point data, generate a candidate target set through initial clustering; Specifically, the effective point data can be clustered using a density peak clustering algorithm to obtain an initial clustering result containing at least one candidate cluster center; based on the spatial clustering characteristics and signal correlation characteristics of the effective point data, at least one cluster center is determined from the initial clustering result; for each cluster center, a corresponding candidate target is generated based on the point cluster it represents, to form the candidate target set.

[0033] Specifically, the effective point data is first processed using a density peak clustering algorithm. The core of this algorithm is to find points in the dataset with high local density that are far from other high-density points as candidate cluster centers. In radar point data processing, local density is determined by calculating the number of other points within a preset distance range around each point. The distance between points is calculated using Euclidean distance, a spatial clustering feature. These two metrics are used to select initial candidate cluster centers, forming the initial clustering results. This process requires no iterative calculations and is computationally efficient.

[0034] Valid cluster centers are then selected from the initial clustering results, focusing on a combination of spatial clustering characteristics and signal correlation characteristics: For each candidate cluster center, the average spatial distance between it and other points within its cluster is calculated; the smaller the average distance, the higher the spatial clustering. Simultaneously, the average similarity between the signal feature vector of the candidate cluster center and the signal feature vectors of other points within the cluster is calculated; the higher the average similarity, the stronger the signal representativeness. A comprehensive confidence score is determined based on the weighted sum of these two indicators (the weights can be preset according to the scenario, such as 0.6 for spatial clustering and 0.4 for signal representativeness). If the comprehensive confidence score is not lower than a preset confidence threshold (e.g., 0.7), it is determined to be a valid cluster center.

[0035] Finally, each cluster of points represented by an effective cluster center corresponds to a candidate target. These candidate targets are integrated to form a set of candidate targets. Each candidate target contains basic information such as the average spatial location, comprehensive motion parameters, and signal feature template of the cluster.

[0036] Furthermore, the process of determining at least one cluster center from the initial clustering results based on the spatial clustering characteristics and signal correlation characteristics of the effective point data may include: c1. Calculate the average spatial distance between any candidate cluster center and other points within its cluster, and determine the spatial clustering score of any candidate cluster center based on the average spatial distance; Specifically, for any candidate cluster center in the initial clustering results, the Euclidean distance between the center and all other points in the corresponding cluster can be calculated based on the spatial clustering characteristics. The average of these distances is then taken, and the average distance is converted into a spatial clustering score in the range of [0,1] through reverse mapping. The smaller the average distance, the higher the spatial clustering score. For example, the score is 1 when the average distance is 0, and the score is 0 when the average distance exceeds the preset maximum distance.

[0037] c2. Calculate the average similarity between the signal feature vector of any candidate cluster center and the signal feature vectors of other points within the same cluster, and use this as the signal representativeness score of any candidate cluster center. Specifically, based on signal correlation features, signal feature vectors (such as Doppler frequency shift, spectral features, etc.) of candidate cluster centers and other points within the cluster can be extracted. The cosine similarity between the center and the signal feature vector of each other point can be calculated. The average value of these similarities is taken as the signal representativeness score. The higher the average value, the more representative the center is of the signal features of the points within the cluster. The score range is [0,1].

[0038] c3. Obtain the comprehensive confidence score of any candidate cluster center by weighted sum of the spatial clustering score and the signal representativeness score; c4. If the overall confidence score is not lower than the preset confidence threshold, then any candidate cluster center shall be taken as the cluster center. c5. Traverse all candidate cluster centers, only to find that there is one less cluster center.

[0039] Specifically, a weighting coefficient can be preset for the spatial clustering score and the signal representativeness score (e.g., spatial clustering coefficient 0.6, signal representativeness coefficient 0.4). The two scores are multiplied by their corresponding coefficients and then summed to obtain the overall confidence score. A preset confidence threshold is calibrated based on experimental data, typically set to 0.7. If the overall confidence score of a candidate cluster center is ≥0.7, it is considered a valid cluster center; if it is below this threshold, the candidate cluster center is removed to avoid interference with subsequent resolution. This calculation and judgment process is repeated for all candidate cluster centers in the initial clustering results until all valid cluster centers are obtained.

[0040] b2. For the current point, calculate the matching degree between the current point and each candidate target in the candidate target set on each standardized feature dimension; Specifically, for the current point and each candidate target, a matching degree analysis is performed on each standardized feature dimension: the matching degree of spatial clustering feature is determined based on the complementary distance value between the point and the center of the candidate target, and the closer the distance, the higher the matching degree; the matching degree of motion consistency feature is determined by the similarity of motion parameters between the point and the candidate target; the matching degree of signal association feature is calculated based on the similarity of signal feature vectors; the matching degree of temporal continuity and signal strength stability features are determined based on timestamp continuity and signal strength deviation rate, respectively, and the matching degree of each dimension is in the range of [0,1].

[0041] b3. The matching degree and the corresponding cohesion weight are weighted and summed to obtain the comprehensive correlation degree between the current point and each candidate target; Specifically, the matching degree of each dimension can be multiplied one-to-one with the dynamically calculated cohesion weight and then summed to obtain the comprehensive correlation degree between the current point and each candidate target. This calculation method is simple and efficient, ensuring real-time performance. The preset correlation degree threshold is dynamically adjusted according to the target state parameters. For example, the threshold can be appropriately increased to [0.7, 0.8] in dense target scenarios and adjusted to [0.6, 0.7] in strong interference scenarios. The threshold adjustment logic is connected with the aforementioned target state evaluation.

[0042] b4. If the highest comprehensive correlation degree exceeds the preset correlation degree threshold, the current point is associated with the candidate target corresponding to the highest comprehensive correlation degree to obtain the resolution result; b5. If the highest comprehensive correlation degree does not exceed the preset correlation degree threshold, the current point is initialized as a new candidate target, and the resolution result is obtained.

[0043] Specifically, if the highest comprehensive correlation exceeds the preset threshold, it indicates that the current point is strongly correlated with the candidate target, and it is associated with the candidate target; if it does not exceed the threshold, the current point is determined to belong to a new target, a new candidate target is initialized and added to the candidate target set, and the initial feature template of the new target is updated to provide a reference for subsequent point association.

[0044] Furthermore, each candidate target in the candidate target set has a dynamically updated feature template; The method further includes: After the current point is assigned, the feature template of the candidate target to which the current point belongs is updated: the center position of the candidate target to which the current point belongs is updated using the position information of the current point; the motion vector of the candidate target to which the current point belongs is updated by fusing the motion information of the current point; the signal feature template of the candidate target to which the current point belongs is updated by fusing the signal features of the current point; and the timestamp and feature information of the current point are added to the historical point sequence of the candidate target to which the current point belongs.

[0045] Specifically, center position update: a weighted average method can be used to fuse the current point position information with the original center position of the candidate target. The weight of the current point can be determined according to its signal-to-noise ratio. The higher the signal-to-noise ratio, the greater the weight (e.g., weight range [0.3, 0.5]). The new center position is obtained through fusion calculation, making the center position more consistent with the actual position of the target.

[0046] Motion vector update: The motion vector of the candidate target includes parameters such as velocity, acceleration, and heading angle. The motion information of the current point is fused with the original motion vector. The fusion process takes into account the continuity of the motion state and adopts the idea of ​​first-order low-pass filtering to smooth the sudden changes in motion parameters, so that the updated motion vector can better reflect the motion trend of the target.

[0047] Signal feature template update: Extract signal features such as signal intensity, signal-to-noise ratio, and signal feature vector of the current point, and perform weighted fusion with the original signal feature template of the candidate target. The fusion weight is determined according to the signal quality of the current point. The better the signal quality, the greater the weight, ensuring that the signal feature template can adapt to subtle changes in the target signal features while maintaining stability.

[0048] Historical point sequence update: The complete information such as the timestamp, location, motion parameters, and signal characteristics of the current point are added to the historical point sequence of the candidate target. The length of the historical sequence can be set to a fixed value (such as the most recent 100 points). The oldest points that exceed the length are automatically removed to ensure that the historical sequence can provide sufficient reference data without consuming too much storage resources.

[0049] 105. Verify the resolution results and perform reverse optimization on the agglomeration weights based on the verification results.

[0050] In a specific implementation process, a preliminary tracking trajectory can be formed based on the resolution results. The effectiveness of the resolution can be judged by analyzing the trajectory smoothness, etc. If errors exist, the gradient descent algorithm is used to adjust the agglomeration weights in reverse based on the error information, thereby realizing online adaptive iterative optimization of the weights. This step can be implemented in the following way: d1. Based on the target identity association table generated by the resolution result, obtain the point sequence of each candidate target within a continuous time window to form a preliminary tracking trajectory; Specifically, based on the target identity association table obtained from the resolution results, all associated points of each candidate target within a continuous time window (the length of the time window can be set according to the radar data update cycle, such as 50ms) can be extracted, sorted by timestamp to form a preliminary tracking trajectory. The trajectory data includes information such as position and motion parameters at each time.

[0051] d2. Perform smoothness analysis on the preliminary tracking trajectory. If the trajectory curvature change rate is lower than the preset curvature threshold and the standard deviation of the trajectory point interval time is lower than the preset time stability threshold, then the preliminary tracking trajectory is determined to be continuous and effective, and the corresponding resolution result is reliable. Specifically, the rate of change of curvature of the trajectory can be calculated. The rate of change of curvature reflects the smoothness of the trajectory. If the rate of change is lower than the preset curvature threshold (e.g., 0.1 rad / m), it indicates that there is no obvious abrupt change in the trajectory. At the same time, the standard deviation of the time interval between trajectory points can be calculated. If the standard deviation is lower than the preset time stability threshold (e.g., 5 ms), it indicates that the distribution of the points is uniform, and the tracking trajectory is determined to be continuous and effective, and the resolution result is reliable.

[0052] d3. If the missing point rate in the preliminary tracking trajectory is higher than the preset missing threshold, or if multiple candidate targets compete for the same new point and their identities conflict within the same time window, then a resolution error is determined, and the error type and the candidate target identifiers involved are recorded. Specifically, if the trajectory point missing rate (the ratio of the number of missing points to the total number of points that should be there) is higher than the preset missing threshold (e.g., 0.2), or if multiple candidate targets compete for the identity of the same new point, it is determined that there is a discrimination error, and the error type (e.g., spatial confusion, motion mismatch) and the candidate target identifiers involved are recorded.

[0053] d4. Generate a weight adjustment signal based on the error type and historical weight adjustment records; Specifically, by combining error types with historical weight adjustment records, the causes of errors can be analyzed. For example, abrupt trajectory changes may be caused by insufficient weight of motion consistency features, and missing points caused by clutter interference may be caused by insufficient weight of signal correlation features. Based on this, targeted weight adjustment signals can be generated to clarify the features that need to be adjusted and the direction of adjustment.

[0054] d5. Based on the weight adjustment signal, a gradient direction optimization strategy is used to adjust the agglomeration weights of each feature. The adjustment magnitude is positively correlated with the severity of the error, and after each adjustment, the weights need to be renormalized so that the sum of the weights is 1.

[0055] Specifically, a gradient direction optimization strategy can be adopted, aiming to minimize the resolution error, and adjusting the agglomeration weights of each feature according to the weight adjustment signal. The adjustment magnitude is positively correlated with the severity of the error; the more severe the error, the larger the adjustment magnitude (the adjustment magnitude range is controlled within ±0.1). For example, severe spatial confusion error can increase the weight of spatial agglomeration features by 0.1. After each adjustment, all feature weights are renormalized to ensure that the sum of the weights is always 1, and that each weight remains within the preset range.

[0056] In this embodiment, based on the smoothness and stability analysis of the tracking trajectory, an objective standard for verifying the resolution results is provided to ensure the accurate identification of the resolution error. Targeted weight adjustment signals are generated for different error types, making weight optimization more directional and effective. The gradient direction optimization strategy is linked to the severity of the error, achieving precise adjustment of the weights and avoiding blind optimization. The closed-loop optimization mechanism enables the weight parameters to gradually adapt to the feature distribution of specific application scenarios, and the resolution accuracy continues to improve with increasing running time.

[0057] Further, based on the error type and historical weight adjustment records, a weight adjustment signal is generated, including: Based on the error type, a candidate feature set to be adjusted is determined; wherein, if the error type is spatial confusion error, the candidate feature set includes spatial clustering features, motion consistency features, and signal correlation features; if the error type is motion mismatch error, the candidate feature set includes motion consistency features and temporal continuity features; if the error type is signal interference error, the candidate feature set includes signal correlation features and signal strength stability features. Obtain the current cohesion weight, historical adjustment range, and corresponding error severity factor for each feature in the candidate feature set; Based on the error severity factor, the weight adjustment step size for each feature in the candidate feature set is calculated, wherein the weight adjustment step size is positively correlated with the error severity factor and negatively correlated with the historical adjustment magnitude. Based on the weight adjustment step size and the preset weight adjustment direction, a weight adjustment signal is generated for each feature in the candidate feature set; The error severity factor is quantified based on at least one of the following: missing point rate, trajectory curvature change rate, or number of identity determination conflicts.

[0058] Specifically, firstly, based on the identified error type, a candidate feature set to be adjusted is determined. Error types include spatial confusion error, motion mismatch error, and signal interference error, each corresponding to a different candidate feature set. When the error type is spatial confusion error, the candidate feature set includes spatial clustering features as mandatory adjustments, and optionally includes motion consistency features and signal correlation features as auxiliary adjustments. When the error type is motion mismatch error, the candidate feature set includes motion consistency features as mandatory adjustments, and optionally includes temporal continuity features as auxiliary adjustments. When the error type is signal interference error, the candidate feature set includes signal correlation features and signal strength stability features as mandatory adjustments, and optionally includes spatial clustering features as auxiliary adjustments.

[0059] After determining the candidate feature set, the current cohesion weight, historical adjustment range, and corresponding error severity factor for each feature in the candidate feature set are obtained. The current cohesion weight is read from the system's weight configuration table. The historical adjustment range records the change in the feature weight over several past adjustments. The error severity factor is quantified based on at least one of the following: point missing rate, trajectory curvature change rate, or number of identity determination conflicts. For example, the higher the point missing rate, the larger the error severity factor; the greater the degree to which the trajectory curvature change rate exceeds the threshold, the larger the error severity factor; the more identity determination conflicts, the larger the error severity factor.

[0060] Based on the error severity factor, the weight adjustment step size for each feature in the candidate feature set is calculated. The weight adjustment step size is positively correlated with the error severity factor, meaning the more severe the error, the larger the adjustment step size. Simultaneously, the weight adjustment step size is negatively correlated with the historical adjustment magnitude; that is, if a feature has already undergone significant adjustments in the past few adjustments, the current adjustment step size should be appropriately reduced to avoid over-adjustment leading to weight oscillations. The adjustment step size can be calculated using a linear function or a piecewise function to ensure the smoothness of the adjustment process.

[0061] Based on the calculated weight adjustment step size and the preset weight adjustment direction, a weight adjustment signal is generated for each feature in the candidate feature set. The weight adjustment direction is preset according to the error type; for example, spatial confusion error corresponds to the direction of increasing the weight of spatial clustering features, motion mismatch error corresponds to the direction of increasing the weight of motion consistency features, and signal interference error corresponds to the direction of increasing the weight of signal correlation features. The adjustment direction of auxiliary adjustment features is consistent with or opposite to that of the features that must be adjusted, depending on the specific circumstances of the error type.

[0062] After generating the weight adjustment signal, it is applied to the actual weight adjustment process. Once the adjustment is complete, the aggregate weights of all features are renormalized to ensure the weight sum is 1. Simultaneously, the magnitude of this adjustment is recorded in the historical adjustment record to provide a reference for subsequent weight optimization.

[0063] Through the aforementioned refined weight adjustment signal generation mechanism, this invention can dynamically calculate the optimal adjustment step size for each feature based on the specific type, severity, and historical adjustment of the error, avoiding blind and over-adjustment, making the weight optimization process more stable and efficient, and ultimately achieving a continuous improvement in multi-target resolution accuracy.

[0064] This embodiment solves the problems of insufficient resolution accuracy and weak adaptability of existing multi-target resolution methods in complex scenarios, meets the practical application requirements of multi-target monitoring systems, and has important theoretical significance and engineering application value.

[0065] Based on the same general inventive concept, this invention also protects a multi-target resolution system based on point feature aggregation. The multi-target resolution system based on point feature aggregation provided by this invention will be described below. The multi-target resolution system based on point feature aggregation described below can be referred to in correspondence with the multi-target resolution method based on point feature aggregation described above.

[0066] Figure 2 This is a schematic diagram of the structure of a multi-target resolution system based on dot feature aggregation provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the multi-target resolution system based on dot feature aggregation in this embodiment includes a preprocessing module 21, a standardization processing module 22, a determination module 23, a determination module 24, and an optimization module 25.

[0067] The preprocessing module 21 is used to collect the point data output by the radar signal processor and preprocess the point data to obtain valid point data. The standardization processing module 22 is used to extract multi-dimensional spot features from the effective spot data and perform standardization processing on the multi-dimensional spot features to obtain a standardized feature set. The determination module 24 is used to calculate the target state parameters based on real-time sensor data, and dynamically calculate the cohesion weight of each feature in the standardized feature set based on the target state parameters. The discrimination module 24 is used to calculate the correlation degree between the current point and each candidate target through feature aggregation based on the aggregation weight and the standardized feature set, and to complete the target discrimination of the current point according to the correlation degree to obtain the discrimination result. The optimization module 25 is used to verify the resolution result and perform reverse optimization on the agglomeration weight based on the verification result.

[0068] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, communication interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a multi-target resolution method based on point feature aggregation.

[0069] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0070] It should be noted that all relevant information that may be involved in the various embodiments of the present invention is processed in strict accordance with the requirements of laws and regulations, following the principles of legality, legitimacy, and necessity, based on the reasonable purpose of the business scenario, and is information that users actively provide or generate during the use of the product / service, as well as information obtained with user authorization.

[0071] The information processed by this invention may vary depending on the specific product / service scenario and should be based on the specific scenario in which the user uses the product / service. This may involve user account information, device information, or other related information. This invention will treat the relevant information and its processing with the utmost diligence.

[0072] This invention places great emphasis on the security of relevant information and has adopted reasonable and feasible security protection measures that comply with industry standards to protect user information and prevent unauthorized access, public disclosure, use, modification, damage or loss of relevant information.

[0073] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0074] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-target discrimination method based on point feature aggregation, characterized in that, include: Collect point data output from the radar signal processor and preprocess the point data to obtain valid point data; Multidimensional trace features are extracted from the valid trace data and the multidimensional trace features are standardized to obtain a standardized feature set; The target state parameters are calculated based on real-time sensor data, and the agglomeration weight of each feature in the standardized feature set is dynamically calculated based on the target state parameters. Based on the agglomeration weight and the standardized feature set, the correlation degree between the current point and each candidate target is calculated through feature agglomeration, and the target discrimination of the current point is completed according to the correlation degree to obtain the discrimination result. The resolution results are verified, and the agglomeration weights are optimized in reverse based on the verification results.

2. The multi-target discrimination method based on point feature aggregation according to claim 1, characterized in that, The target state parameters include at least one of target density, target motion complexity, and interference intensity; The standardized feature set includes at least one of spatial clustering features, motion consistency features, signal correlation features, temporal continuity features, and signal strength stability features; Based on the target state parameters, the agglomeration weight of each feature in the multidimensional point trace feature is dynamically calculated, including: The fuzzy hierarchical analysis method is used to assign agglomeration weights to each feature based on the comparison results between the values ​​of the target state parameters and the corresponding preset parameter thresholds. If only the target density is higher than the preset density threshold, then the agglomeration weight of the spatial aggregation feature is increased. If only the target motion complexity is higher than the preset complexity threshold, then increase the cohesion weight of the motion consistency feature; If only the interference intensity is higher than the preset interference threshold, then the cohesion weight of the signal correlation feature is increased; If at least two of the target state parameters, namely the target density, the target motion complexity, and the interference intensity, are higher than the corresponding threshold, the dominant feature of the current scene is determined, and the cohesion weight of the dominant feature is increased. The cohesion weights of the time continuity feature and the signal strength stability feature remain unchanged, and the sum of the cohesion weights of all features is 1, with each cohesion weight value located within its respective preset range.

3. The multi-target discrimination method based on point feature aggregation according to claim 2, characterized in that, Identify the dominant features of the current scene, including: Calculate the threshold ratio of the out-of-limit target state parameters respectively; the out-of-limit target state parameters include at least two of the target density, the target motion complexity, and the interference intensity; Normalize each threshold ratio to obtain the scene influence factor of the state parameters of the exceeding target; If there exists a unique maximum scene impact factor, the feature corresponding to the exceedance target state parameter corresponding to the unique maximum scene impact factor shall be taken as the dominant feature; If there are multiple maximum scenario impact factors, obtain the historical scenario impact factors of the target state parameters exceeding the standard corresponding to each maximum scenario impact factor in the most recent N processing cycles; Calculate the average historical scenario impact factor of the target state parameter exceeding the standard corresponding to each maximum scenario impact factor; The feature corresponding to the out-of-target state parameter corresponding to the maximum average value is taken as the dominant feature.

4. The multi-target discrimination method based on point feature aggregation according to claim 1, characterized in that, Based on the agglomeration weights and the standardized feature set, the correlation between the current point and each candidate target is calculated through feature agglomeration, and the target discrimination of the current point is completed according to the correlation, resulting in a discrimination result, including: Based on the effective point data, a candidate target set is generated through initial clustering; For the current point, calculate the matching degree between the current point and each candidate target in the candidate target set on each standardized feature dimension; The matching degree is weighted and summed with the corresponding cohesion weight to obtain the comprehensive correlation degree between the current point and each candidate target; If the highest comprehensive correlation score exceeds the preset correlation score threshold, the current point is associated with the candidate target corresponding to the highest comprehensive correlation score to obtain the resolution result. If the highest overall correlation score does not exceed the preset correlation score threshold, the current point is initialized as a new candidate target, and the resolution result is obtained.

5. The multi-target discrimination method based on point feature aggregation according to claim 4, characterized in that, Based on the effective point data, a candidate target set is generated through initial clustering, including: The effective point data is clustered using the density peak clustering algorithm to obtain an initial clustering result containing at least one candidate cluster center; Based on the spatial clustering characteristics and signal correlation characteristics of the effective point data, at least one cluster center is determined from the initial clustering results; For each cluster center, a corresponding candidate target is generated based on the point cluster it represents, thus forming the candidate target set.

6. The multi-target discrimination method based on point feature aggregation according to claim 5, characterized in that, Based on the spatial clustering characteristics and signal correlation characteristics of the effective point data, at least one cluster center is determined from the initial clustering results, including: Calculate the average spatial distance between any candidate cluster center and other points within its cluster, and determine the spatial clustering score of any candidate cluster center based on the average spatial distance; Calculate the average similarity between the signal feature vector of any candidate cluster center and the signal feature vectors of other points within the same cluster, and use this as the signal representativeness score of the candidate cluster center. The comprehensive confidence score of any candidate cluster center is obtained by weighting the spatial clustering score and the signal representativeness score. If the overall confidence score is not lower than the preset confidence threshold, then any candidate cluster center will be taken as the cluster center. By iterating through all candidate cluster centers, we find that we are missing just one cluster center.

7. The multi-target resolution method based on point feature aggregation according to claim 4, characterized in that, Each candidate target in the candidate target set has a dynamically updated feature template; The method further includes: After completing the assignment of the current point, update the feature template of the candidate target to which the current point belongs: Among them, the center position of the candidate target to which the current point belongs is updated using the position information of the current point. The motion vector of the candidate target to which the current point belongs is updated by fusing the motion information of the current point. The signal feature template of the candidate target to which the current point belongs is updated by fusing the signal features of the current point. The timestamp and feature information of the current point are added to the historical point sequence of the candidate target to which it belongs.

8. The multi-target discrimination method based on point feature aggregation according to claim 1, characterized in that, The resolution results are verified, and the agglomeration weights are inversely optimized based on the verification results, including: Based on the target identity association table generated by the resolution results, the point sequence of each candidate target in a continuous time window is obtained to form a preliminary tracking trajectory. A smoothness analysis is performed on the preliminary tracking trajectory. If the rate of change of trajectory curvature is lower than a preset curvature threshold and the standard deviation of trajectory point interval time is lower than a preset time stability threshold, then the preliminary tracking trajectory is determined to be continuous and effective, and the corresponding resolution result is reliable. If the missing point rate in the initial tracking trajectory is higher than the preset missing threshold, or if multiple candidate targets compete for the same new point and their identities conflict within the same time window, then a discrimination error is determined, and the error type and the candidate target identifiers involved are recorded. Based on the error type and historical weight adjustment records, a weight adjustment signal is generated; Based on the weight adjustment signal, a gradient direction optimization strategy is used to adjust the agglomeration weights of each feature. The adjustment magnitude is positively correlated with the severity of the error, and after each adjustment, the weights need to be renormalized so that the sum of the weights is 1.

9. The multi-target discrimination method based on point feature aggregation according to claim 8, characterized in that, Based on the error type and historical weight adjustment records, a weight adjustment signal is generated, including: Based on the error type, a candidate feature set to be adjusted is determined; wherein, if the error type is spatial confusion error, the candidate feature set includes spatial clustering features, motion consistency features, and signal correlation features; if the error type is motion mismatch error, the candidate feature set includes motion consistency features and temporal continuity features; if the error type is signal interference error, the candidate feature set includes signal correlation features and signal strength stability features. Obtain the current cohesion weight, historical adjustment range, and corresponding error severity factor for each feature in the candidate feature set; Based on the error severity factor, the weight adjustment step size for each feature in the candidate feature set is calculated, wherein the weight adjustment step size is positively correlated with the error severity factor and negatively correlated with the historical adjustment magnitude. Based on the weight adjustment step size and the preset weight adjustment direction, a weight adjustment signal is generated for each feature in the candidate feature set; The error severity factor is quantified based on at least one of the following: missing point rate, trajectory curvature change rate, or number of identity determination conflicts.

10. A multi-target resolution system based on point feature aggregation, characterized in that, include: The preprocessing module is used to collect the point data output by the radar signal processor and preprocess the point data to obtain valid point data. The standardization processing module is used to extract multi-dimensional trace features from the valid trace data and perform standardization processing on the multi-dimensional trace features to obtain a standardized feature set. The determination module is used to calculate the target state parameters based on real-time sensor data, and dynamically calculate the agglomeration weight of each feature in the standardized feature set based on the target state parameters. The discrimination module is used to calculate the correlation between the current point and each candidate target through feature aggregation based on the aggregation weight and the standardized feature set, and to perform target discrimination of the current point according to the correlation, so as to obtain the discrimination result. An optimization module is used to verify the resolution results and perform reverse optimization on the agglomeration weights based on the verification results.

Images