Track association method based on local and global similarity fusion

Abstract

A flight path association method based on local and global similarity fusion belongs to the technical field of flight path association, and comprises the following steps: S1, preprocessing obtained flight path pair data, including abnormal value elimination and time coincidence part interpolation alignment; s2, respectively extracting local similarity features and global similarity features of the track pairs; s3, performing weighted fusion on the local and global similarity features, inputting the local and global similarity features into a random forest classifier model for training, and determining an optimal weight combination of different similarity features; and S4, carrying out weighted fusion on the local and global similarity features of the unknown track pair according to the optimal weight combination, and inputting the local and global similarity features into the trained random forest classifier model to carry out association judgment. According to the method, the influence caused by misjudgment of local similarity features can be reduced under the condition that the overlap ratio of the track to time is low, and the method still has good association accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of track association, and in particular relates to a track association method based on local and global similarity fusion. Background Art

[0002] In the field of track data analysis, track association is an important task, which plays an important role in both civil and military fields. By analyzing the position information of objects at different time points, tracks can be effectively matched and associated, thereby achieving the understanding and prediction of object behavior. However, as the amount of track data increases and the track forms become more and more complex, the track association task is facing more and more challenges.

[0003] Traditional track similarity discrimination methods usually focus on the local similarity features of tracks, and most of them only target one-dimensional features. This results in the extracted features being unable to fully represent the overall similarity relationship of tracks when faced with complex multi-dimensional track scenarios, resulting in misjudgment. Therefore, existing track association methods often show inadequacies when dealing with complex scenarios, especially when the temporal overlap between tracks is low. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present invention proposes an association method based on the fusion of local and global similarities. The method first extracts local similarity features by interpolating and time-aligning the tracks, and extracts global similarity features from the original tracks at the same time, and determines the optimal weight combination of all features through an optimization scheme, and finally achieves accurate association and matching of the tracks. This scheme that integrates local and global similarities still has a high association accuracy in scenarios where the temporal overlap between tracks is low.

[0005] The solution adopted by the present invention to solve the above problems is: A track association method based on local and global similarity fusion includes the following steps: S1. Preprocessing the obtained track data, including removing outliers from the track data and aligning the time-coinciding parts by interpolation; S2, extracting local similarity features and global similarity features of the track pairs respectively; S3, weighted fusion of local and global similarity features, input into the random forest classifier model for training, and determine the optimal weight combination of different similarity features; S4. The local and global similarity features of the unknown track pairs are weightedly fused according to the optimal weight combination and input into the trained random forest classifier model for association judgment.

[0006] Furthermore, in step S1, outliers are eliminated by using a method based on median sliding window filtering, which can effectively process sampling noise and outlier data, enhance the robustness of the model, and ensure the stability of association judgment. The process is as follows: S111, determine the size of the sliding window, the window is usually an odd number of data points, such as 3, 5, 7, etc.; S112, starting from the starting position of the signal, slide the window to the right one data point at a time; S113. For each window, sort all data points in the window and take the sorted middle value as the output value of the current window position. Compare the current point with the median value. If the difference exceeds the threshold, it is determined to be an abnormal point, and then the abnormal point is replaced by the median value of the two adjacent points.

[0007] S114, slide the window to the next position, and repeat step S113 until all data points of the signal are processed.

[0008] Furthermore, in step S1, the track is interpolated and aligned for the time-coinciding part, using a cubic spline interpolation method. This method makes the track generated by interpolation smoother, can more accurately reflect the real motion state, and reduce the error caused by data loss. The process is as follows: S121. Construct a cubic polynomial on each interval; S122, determining the interpolation condition, continuity condition and boundary condition of the polynomial; S123, solving the correlation coefficient; Assume that the two tracks from two sources are processed by outlier replacement, respectively: ; ; in and Respectively represent the sequence and No. The location information at a certain moment, and is the corresponding timestamp, and Respectively and The number of track points, track and track After interpolation, the time-aligned track parts are expressed as and , ; ; Among them, after interpolation and Already aligned on the timeline, and Represents the interpolation sequence and No. The location information at a certain moment, is the corresponding timestamp, Indicates the number of interpolated partial track points.

[0009] Furthermore, in step S2, the local similarity features are extracted from the interpolated aligned part of the track pair, and the global similarity features are extracted from the whole part of the original track pair without interpolation. The local similarity features include: average Euclidean distance, Hausdorff distance and average Frechet distance; the global similarity features include: centroid distance, bounding box difference and normalized DTW distance.

[0010] Preferably, the average Euclidean distance is used to measure the average spatial distance between two tracks at the same time point, reflecting the local position characteristics; the Hausdorff distance is used to measure the distance between the farthest points in the two tracks, capturing the extreme differences and reflecting the local difference characteristics; the average Frechet distance is used to reflect the overall similarity of the track shapes and reflecting the local shape characteristics; Local similarity feature extraction in interpolated tracks and The local similarity feature set It is expressed as, ; in, represents the average Euclidean distance, represents the Hausdorff distance, represents the average Frechet distance; The centroid distance is used to calculate the distance between the centroids (center points of the tracks), reflecting the global position characteristics; the bounding box difference is used to compare the minimum bounding boxes of the tracks, measure the difference in track ranges, and reflect the global difference characteristics; the standardized DTW distance uses an algorithm based on dynamic time warping to calculate the shape similarity of the tracks, and after normalization, it reflects the overall shape similarity of the tracks, reflecting the global shape characteristics; Global similarity feature extraction in the original track and The global similarity feature set It is expressed as, ; in, represents the track center of gravity, represents the bounding box difference, represents the normalized DTW distance.

[0011] In step S3, the method for determining the optimal weight combination of different similarity features adopts a method based on Bayesian optimization; this method approximates the objective function through a proxy model and selects evaluation points based on the acquisition function. Compared with traditional grid search or random search, the computational cost required for feature weight optimization is greatly reduced, while still being able to approximate the optimal weight combination and reducing the model tuning time. The process is as follows: S31. Define the optimization goal and use the model accuracy as the objective function of Bayesian optimization, the goal is to maximize the accuracy; S32, determining variables to be optimized, including weights of local and global similarity features; S33, construct an optimization problem, use feature weights as optimization variables, and define upper and lower limits of parameters; S34, performing Bayesian optimization, gradually selecting parameter combinations through the Bayesian optimization framework, evaluating the effect of each set of parameters, and updating the search strategy to approach the optimal solution; S35, verification and evaluation, based on the optimal parameter combination obtained by Bayesian optimization, retrain the model and evaluate the model performance through cross-validation.

[0012] In S32, the local similarity feature weight obtained by Bayesian optimization is assumed to be , the global similarity feature weight is , the weighted fusion features, ; ; ; in, is the local similarity weighted fusion feature, is the global similarity weighted fusion feature, is the final input feature vector.

[0013] In step S4, the random forest classifier model is input with local and global similarity features of unknown track pairs weighted and fused according to the optimal weight, and outputs the association result, 0 for no association and 1 for association.

[0014] The beneficial effects of the present invention are mainly manifested in: (1) High association accuracy: This method uses local and global similarity features to jointly determine whether a track pair is associated, and determines the importance of different features through Bayesian optimization and assigns corresponding weights, thereby avoiding the situation where the local similarity features are insufficient to determine whether the tracks are associated due to the low temporal overlap of the track pairs, thus causing misjudgment of the association results. The track association method has a high association accuracy, and can maintain a high association accuracy even when the temporal overlap of the track pairs is low.

[0015] (2) High association versatility: The local and global similarity features used in this method are mainly extracted from three dimensions: position features, difference features, and shape features. They do not rely on motion features based on speed, acceleration, etc., avoiding the situation where the association scheme is not universal due to large differences in the motion modes of the associated targets. This makes the method highly adaptable to different types of object association tasks (such as aircraft, ships, vehicles, etc.), and is especially suitable for scenarios with variable track motion modes. In addition, the importance of different similarity features may be different in different scenarios. The present invention automatically adjusts the weight combination of local and global features through Bayesian optimization technology to maximize the accuracy of the association. Compared with the fixed weight scheme, this method can adapt to different scenarios and different track complexities, improving the generalization ability and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a track association method based on local and global similarity fusion provided by an example of the present invention.

[0017] Figure 2 It is a test flow diagram of a track association method based on local and global similarity fusion provided by an example of the present invention.

[0018] Figure 3 It is a schematic diagram of track scenarios with different time overlaps provided by an example of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0020] Reference Figure 1~Figure 3 , a track association method based on local and global similarity fusion, comprising the following steps: S1. Preprocessing the obtained track data, including removing outliers from the track data and aligning the time-coinciding parts by interpolation; In step S1, outliers are eliminated by using a method based on median sliding window filtering; this method can effectively process sampling noise and outlier data, enhance the robustness of the model, and ensure the stability of association judgment. The process is as follows: Set track sequence ,in Indicates the location information of a track point.

[0021] S111. Determine the window size (like etc.), the window moves through each track point; S112. For each window , sort the track points in the window by value and take the middle value .

[0022] S113. For each track point , and the median of the corresponding window For comparison: like , then determine For abnormal points, replace them with the median of two adjacent points. ; S114, repeat the above steps until all points are processed.

[0023] In step S1, the track is interpolated and aligned for the time-coinciding part, using a cubic spline interpolation method; this method makes the track generated by interpolation smoother, can more accurately reflect the real motion state, and reduce the error caused by data loss; suppose the track sequence The corresponding time series is ,in Indicates the timestamp of the track point. The process is as follows:

[0024] S121, in each interval [ t i , t i +1 ] superior Construct a cubic polynomial, ; in, , , and is the coefficient to be determined.

[0025] S122. Determine the undetermined coefficients by the following conditions: Interpolation conditions: The spline curve must pass through each data point, , ; Continuity conditions: For each node, the first and second derivatives of the spline are continuous. ; Boundary conditions: for example, natural boundary conditions (the second derivative is 0 at the boundary).

[0026] S123, solve the corresponding coefficient , , and ;

[0027] Consider two tracks from two sources after outlier replacement. Expressed as ,track Expressed as , ; ; in and Respectively represent the sequence and No. The location information at a certain moment, and is the corresponding timestamp, and Respectively and The number of track points, track and track After interpolation, the time-aligned track parts are expressed as and , ; ; Among them, the interpolated and Already aligned on the timeline, and Represents the interpolation sequence and No. The location information at a certain moment, is the corresponding timestamp, Indicates the number of interpolated partial track points.

[0028] S2, extracting local similarity features and global similarity features of the track pairs respectively;

[0029] In step S2, the local similarity features are extracted from the interpolated aligned part of the track pair, and the global similarity features are extracted from the whole part of the original track pair without interpolation. The local similarity features include: average Euclidean distance, Hausdorff distance and average Frechet distance. The global similarity features include: centroid distance, bounding box difference and normalized DTW distance.

[0030] The average Euclidean distance is used to measure the average spatial distance between two tracks at the same time point, reflecting the local position characteristics; the Hausdorff distance is used to measure the distance between the farthest points in the two tracks, capturing extreme differences and reflecting local difference characteristics; the average Frechet distance is used to reflect the overall similarity of track shapes and reflecting local shape characteristics.

[0031] Local similarity feature extraction in interpolated tracks and The process is as follows: (1) Extracting the average Euclidean distance , ; (2) Extracting Hausdorff distance , ; (3) Extracting the average Frechet distance : Number of track points It may be quite large. In order to speed up the calculation and facilitate normalization, the method of first segmenting the track, then calculating the Frechet distance for each segment, and finally calculating the weighted average Frechet distance is adopted.

[0032] The Frechet distance of each segment is defined as, F ( T i ( k ) , T j ( k ) )= inf α , β sup t ∈[0,1] P iα ( t ) ' - P jβ ( t ) ' ; in, Indicates part, and is a parameterized function representing the track segment and The point sequence on is matched, Indicates the time parameter, the value range is [0,1] , and Represents the track segment and In time The traces of time;

[0033] Assume that there are segment, the final average Frechet distance is, ; in, For the The length of the segment; (4) Normalize each feature and represent the local similarity feature set It is expressed as, ; The centroid distance is used to calculate the distance between the centroids (center points of the tracks), reflecting the global position characteristics; the bounding box difference is used to compare the minimum bounding boxes of the tracks, measure the difference in track ranges, and reflect the global difference characteristics; the standardized DTW distance uses an algorithm based on dynamic time warping to calculate the shape similarity of the tracks, and after normalization, it reflects the overall shape similarity of the tracks, reflecting the global shape characteristics.

[0034] Global similarity feature extraction in the original track and The process is as follows: (1) Extracting the centroid distance , ; (2) Extracting bounding box differences , ; in, Indicates the track The bounding box, i.e. the minimum enclosing rectangle, is represented by the minimum and maximum values ​​in each coordinate dimension; (3) Extracting the standardized DTW distance , ; in, is the best matching path found by the DTW dynamic time warping algorithm, consisting of a series of index pairs composition, is the number of matching point pairs in the path; (4) Normalize each feature to represent the global similarity feature set It is expressed as, ; S3. Perform weighted fusion of local and global similarity features and input them into the random forest classifier model for training to determine the optimal weight combination of different similarity features.

[0035] In step S3, the method for determining the optimal weight combination of different similarity features adopts a method based on Bayesian optimization. This method approximates the objective function through a proxy model and selects evaluation points based on the acquisition function. Compared with traditional grid search or random search, it greatly reduces the computational cost required for feature weight optimization, while still being able to approximate the optimal weight combination and reducing the model tuning time. The process is as follows:

[0036] S31. Define the optimization goal and use the model accuracy as the objective function of Bayesian optimization. The optimization objective function can be written as: ; in, and Respectively represent the weighted combination of local similarity features and global similarity features, is the prediction accuracy of the model, and the goal is to maximize the accuracy.

[0037] S32, determining the variables to be optimized, including the weights of local and global similarity features, ; ; in, Indicates The weight of the local similarity feature, Take 1, 2 and 3, Indicates The weight of the global similarity feature, Take 1, 2, 3.

[0038] S33, construct an optimization problem, use feature weights as optimization variables, and define upper and lower limits of parameters. , ; , ; These constraints ensure that the weights sum to 1, representing the contribution of each feature.

[0039] S34. Perform Bayesian optimization, gradually select parameter combinations through the Bayesian optimization framework, evaluate the effect of each set of parameters, and update the search strategy to approach the optimal solution. The core of Bayesian optimization is to build a proxy model To approximate the objective function .

[0040] Select the starting point , calculate the corresponding objective function value .

[0041] Then, we fit the Gaussian process to generate an estimated mean function. As The expected value of and update the proxy model , ; The next evaluation point is selected by the acquisition function, ; Calculate the new objective function value , and update the model.

[0042] The above process is repeated until the maximum number of iterations is reached or the error is less than the threshold.

[0043] S35, verification and evaluation, based on the optimal parameter combination obtained by Bayesian optimization, retrain the model and evaluate the model performance through cross-validation.

[0044] Assume that the local similarity feature weight obtained by Bayesian optimization is , the global similarity feature weight is , the weighted fusion features, ; ; ; in, is the local similarity weighted fusion feature, is the global similarity weighted fusion feature, is the final input feature vector.

[0045] S4. The local and global similarity features of the unknown track pairs are weightedly fused according to the optimal weight combination and input into the trained random forest classifier model for association judgment.

[0046] The test process of this embodiment is as follows Figure 2 As shown, the random forest classifier model described in step S4, the input is the local and global similarity features of the unknown track pair weighted fused according to the optimal weight, and the output is the association result, 0 is not associated, and 1 is associated.

[0047] Suppose the trained random forest classifier is , input the weighted similarity feature of the test track pair , the classifier outputs the predicted label , ; in, Indicates the correlation judgment result of the test track pair, output Indicates that the track pair is not associated, output Indicates track pair association.

[0048] In summary, the method proposed in the present invention extracts the local and global similarity features of the track pair, and determines the optimal weight combination of all features through the Bayesian optimization scheme, and finally realizes accurate association and matching of the tracks. This scheme that integrates local and global similarities has high association accuracy and is applicable to scenarios with low temporal overlap between tracks. The schematic diagrams of scenarios with different temporal overlaps are shown in the figure. Figure 3 shown.

[0049] The above description is only a preferred example of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A track association method based on local and global similarity fusion, characterized in that: The method comprises the following steps: S1. Preprocessing the obtained track data, including removing outliers from the track data and aligning the time-coinciding parts by interpolation; S2, extracting local similarity features and global similarity features of the track pairs respectively; S3, weighted fusion of local and global similarity features, input into the random forest classifier model for training, and determine the optimal weight combination of different similarity features; S4. The local and global similarity features of the unknown track pairs are weightedly fused according to the optimal weight combination and input into the trained random forest classifier model for association judgment.

2. A track association method based on local and global similarity fusion according to claim 1, characterized in that: In step S1, outliers are eliminated by using a method based on median sliding window filtering, and the process is as follows: S111, determine the size of the sliding window, the window is an odd number of data points; S112, starting from the starting position of the signal, slide the window to the right one data point at a time; S113, for each window, sort all data points in the window, take the sorted middle value as the output value of the current window position; compare the current point position with the median value, if the difference exceeds the threshold, it is determined to be an abnormal point, and then replace the abnormal point with the median value of two adjacent points; S114, slide the window to the next position, and repeat step S113 until all data points of the signal are processed.

3. A track association method based on local and global similarity fusion according to claim 1 or 2, characterized in that: In step S1, the track is interpolated and aligned for the time-coinciding part, using a cubic spline interpolation method, and the process is as follows: S121. Construct a cubic polynomial on each interval; S122, determining the interpolation condition, continuity condition and boundary condition of the polynomial; S123. Solve for the correlation coefficient.

4. A track association method based on local and global similarity fusion according to claim 1 or 2, characterized in that: In step S2, the local similarity features are extracted from the interpolated aligned part of the track pair, and the global similarity features are extracted from the entire part of the original track pair without interpolation. The local similarity features include average Euclidean distance, Hausdorff distance and average Frechet distance; the global similarity features include center of gravity distance, bounding box difference and standardized DTW distance.

5. A track association method based on local and global similarity fusion according to claim 4, characterized in that: The average Euclidean distance is used to measure the average spatial distance between two tracks at the same time point, reflecting the local position characteristics; The Hausdorff distance is used to measure the distance between the farthest points in two tracks, captures extreme differences, and reflects local difference characteristics; the average Frechet distance is used to reflect the overall similarity of track shapes and reflects local shape characteristics.

6. A track association method based on local and global similarity fusion according to claim 4, characterized in that: The centroid distance is used to calculate the distance between the centroids of the tracks, reflecting the global position characteristics; The bounding box difference is used to compare the minimum bounding box of the track, measure the difference in track range, and reflect the global difference characteristics; the standardized DTW distance uses an algorithm based on dynamic time warping to calculate the shape similarity of the track, and after normalization, it reflects the overall shape similarity of the track and reflects the global shape characteristics.

7. A track association method based on local and global similarity fusion according to claim 1 or 2, characterized in that: In step S3, the method for determining the optimal weight combination of different similarity features adopts a Bayesian optimization method, and the process is as follows: S31. Define the optimization goal and use the model accuracy as the objective function of Bayesian optimization; S32, determining variables to be optimized, including weights of local and global similarity features; S33, construct an optimization problem, use feature weights as optimization variables, and define upper and lower limits of parameters; S34, performing Bayesian optimization, gradually selecting parameter combinations through the Bayesian optimization framework, evaluating the effect of each set of parameters, and updating the search strategy to approach the optimal solution; S35, verification and evaluation, based on the optimal parameter combination obtained by Bayesian optimization, retrain the model and evaluate the model performance through cross-validation.

8. A track association method based on local and global similarity fusion according to claim 1 or 2, characterized in that: In step S4, the random forest classifier model is input with local and global similarity features of unknown track pairs weighted and fused according to the optimal weight, and outputs the association result, 0 for no association and 1 for association.

Images