Adaptive target association method and system based on multi-source data

Abstract

The invention provides a self-adaptive target association method and system based on multi-source data. The method comprises the steps of S1, preprocessing to obtain a boundary threshold value; S2, inputting a current target detection result and filtering repeated position information; S3, accumulating and storing data of the same source; s5, judging whether an associated track exists or not; S6, if not, taking the target point as a new track starting point; s7, if yes, calculating a search threshold value to determine a search area; S8, calculating a matching measurement matrix of a trajectory and a target position; S9, carrying out two-dimensional distribution to obtain an optimal matching result; S10, updating a trajectory state, S11, judging unassociated target points, S12, forming a new trajectory, S13, updating a trajectory set, and S14, carrying out iterative processing. The method is suitable for marine target trajectory association and tracking under the multi-load observation background, effectively solves the problem of inconsistent positioning error and time interval, and realizes continuous and stable tracking of the target.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition and tracking, and in particular, relates to an adaptive target association method and system based on multi-source data. Background Art

[0002] Due to the limitation of the movement cycle, the satellite visits the monitoring area intermittently, and the time interval between two visits is long, usually in the order of hours, so the revisit rate of satellite information is low. At the same time, affected by the original observation and processing algorithm, the information obtained by the satellite usually has a certain degree of uncertainty and incompleteness. Under the above background conditions, it is difficult to stably and effectively track and identify each target in the monitoring area based on the information obtained by the satellite.

[0003] The Chinese patent document with publication number CN105654133A discloses a ship trajectory fusion system based on multi-source data and its implementation method, including: step 1: the ship position coordinate conversion module converts the polar coordinates of the ship position detected by sonar and radar to the WGS-84 coordinate system, that is, unifies the ship positions detected by sonar, radar and AIS into the same coordinate system, and transmits the data to the ship position interpolation module; step 2: when any group of ship position data processed by sonar, radar or AIS in step 1 is received, the ship position interpolation module performs interpolation on the other two groups of ship positions that have not been received. Interpolation processing is performed to synchronize the sonar, radar or AIS detection data that are asynchronous in time, and the data is transmitted to the ship track association module; Step 3: When two or three of the three devices, the sonar, radar or AIS, detect the same ship at the same time, the device outputs the track of the ship as being associated; Step 4: The underwater and surface ship judgment module makes a judgment based on the results of each ship track association analysis transmitted by the ship track association module; Step 5: The underwater and surface ship judgment module transmits the ship fusion track results transmitted by the association track fusion module and the ship judgment results to the ship track display module. However, this technology projects the position information from different sources into the same coordinate system, and does not perform position deduplication of the same target. The position information of the same target will interfere with the track association, thereby affecting the association accuracy.

[0004] The Chinese patent document with publication number CN106950956A discloses a vehicle trajectory prediction system that integrates a kinematic model and a behavioral cognition model, including: an interactive mixing module interactively mixes the prediction results of each prediction module at the previous moment and outputs a mixed result, and the mixed result is used for prediction at the next moment; the prediction module includes a behavioral cognition trajectory prediction module based on behavioral cognition and a motion trajectory prediction module based on kinematics, and the behavioral cognition trajectory prediction module and the motion trajectory prediction module predict according to the mixed results output by the interactive mixing module, and output the prediction results of each prediction module, and the prediction results include vehicle position and covariance matrix; the fusion update module fuses the final prediction results according to the prediction results, and updates the weight coefficient, and outputs the vehicle position and covariance matrix at a certain moment in the future. However, this technology is suitable for continuous and stable motion target observation forms, and cannot effectively handle the situation where there is a random precision error in the target position and the observation sequence is discontinuous under different load observation modes.

[0005] In actual situations, for the observation of moving targets in the real sea environment, the satellite-borne sensors are affected by the satellite orbit period, so their access to the monitoring area is intermittent, and the time interval between two visits is relatively long. At the same time, sensor observations are uncertain, there are random positioning accuracy errors, and the errors of each sensor are not completely consistent, making it difficult to observe a continuous and stable position sequence, which greatly increases the difficulty of marine target recognition and trajectory tracking. Traditional target recognition and tracking algorithms have great limitations when dealing with situations where target position information is sparse and position accuracy errors are large, and cannot effectively achieve trajectory association and continuation. Summary of the invention

[0006] In view of the defects in the prior art, an object of the present invention is to provide an adaptive target association method and system based on multi-source data.

[0007] According to the present invention, an adaptive target association method based on multi-source data includes:

[0008] Step S1: Perform preprocessing operations to obtain a grid division boundary threshold;

[0009] Step S2: input the target detection result at the current moment, and perform filtering to remove the repeated position information provided by different payloads of the same target;

[0010] Step S3: Determine whether the target detection result at the current moment has the same load source as the data at the previous moment, and accumulate and store the data with the same load source;

[0011] Step S4: If the target detection result obtained by the current load is empty, the accumulated data obtained by the next load is read for subsequent processing;

[0012] Step S5: Determine whether there is currently an associated track;

[0013] Step S6: If it does not exist, each target point is used as the starting point of the trajectory, assembled into a defined format and stored;

[0014] Step S7: otherwise, the adaptive search threshold is calculated based on the current load acquisition data and the last position point of the trajectory to determine the sector search area;

[0015] Step S8: further calculating the matching metric matrix between the associated trajectory and the target detection position information of the current frame;

[0016] Step S9: performing two-dimensional distribution on the calculated matching metric matrix to obtain the optimal matching result;

[0017] Step S10: According to the matching result, the prediction method is applied to update the current associated trajectory state;

[0018] Step S11: determining whether there is an unassociated target point in the current frame;

[0019] Step S12: for the associated target, compare its corresponding value in the matching metric matrix with the threshold value to determine whether to add a new track, otherwise use the unassociated target point as the starting point of the new track;

[0020] Step S13: updating the trajectory set according to the last moment status of each trajectory;

[0021] Step S14: Wait for the next frame of target detection results to perform iterative processing and update the trajectory set.

[0022] Preferably, step S1 comprises:

[0023] Step S1.1: Input target motion and load observation characteristic parameters;

[0024] Step S1.2: Determine the segmentation period according to the load observation characteristics;

[0025] Step S1.3: Calculate the moving range of the target within a fixed period according to its motion characteristics;

[0026] Step S1.4: Grid the observation area according to the motion characteristics;

[0027] Step S1.5: Calculate the boundary coordinate information of each grid area;

[0028] Step S1.6: Calculate and store the coordinate coverage threshold of each grid.

[0029] Preferably, step S2 comprises:

[0030] Step S2.1: Input the single-frame target detection result and find the corresponding positioning longitude error according to the load source of the target point;

[0031] Step S2.2: traverse the two targets in turn to determine whether they are from the same load source;

[0032] Step S2.3: If the load sources are different, the positions of the two target points to be compared are projected into the plane coordinate system;

[0033] Step S2.4: Calculate the position error between the two target points projected into the plane coordinate system;

[0034] Step S2.5: Determine whether the position error is greater than a threshold;

[0035] Step S2.6: For target point pairs whose position error does not exceed the threshold, draw circles with the error range as the radius and update the position of the target point;

[0036] Step S2.7: Continue to process the next frame of target detection results for iterative update.

[0037] Preferably, step S2.3 comprises:

[0038] Step S2.3.1: Project the longitude and latitude lines on the earth's surface onto a cylindrical surface, and then unfold the cylindrical surface into a flat map, where the circumference is regarded as the X-axis;

[0039] Step S2.3.2: Convert longitude from degrees to radians according to the following formula:

[0040] x_radian = lon*PI / 180

[0041] Among them, x_radian represents the corresponding radian value, lon represents the longitude value, and PI represents pi;

[0042] Step S2.3.3: Convert the latitude from degrees to radians according to the following formula:

[0043] y_radian = lat*PI / 180

[0044] Among them, y_radian represents the corresponding radian value, lat represents the latitude value, and PI represents pi;

[0045] Step S2.3.4: Perform the Miller projection transformation according to the following formula:

[0046] y_miller=1.25*log(tan(0.25*PI+0.4*y_radian))

[0047] Among them, y millerRepresents the conversion value of radians after Miller projection, PI represents pi, and y_radian represents the radian value corresponding to the latitude;

[0048] Step S2.3.5: Convert radians to actual distances according to the following formula:

[0049] x=(W / 2)+(W / (2*PI))*x_radian

[0050] y=(H / 2)+(H / (2*mill))*y_radian

[0051] Among them, x represents the coordinate of the converted x-axis, y represents the coordinate of the converted y-axis, W represents the X-axis, which is equal to the circumference of the earth, PI represents pi, x_radian represents the radian value corresponding to the longitude, H represents the Y-axis, which is approximately equal to half of the circumference of the earth, mill represents a constant in the Miller projection, which ranges approximately between plus and minus 2.3, and y_radian represents the radian value corresponding to the latitude.

[0052] Preferably, step S2.6 comprises:

[0053] Case 1: The error ranges of the two target points have no overlapping area;

[0054] Case 2: The error ranges of the two target points have overlapping areas but no intersection;

[0055] Case 3: The error ranges of the two target points overlap and the intersection is on the same side of the two circle centers;

[0056] Case 4: The error ranges of the two target points overlap and the intersection is on the opposite sides of the two circle centers;

[0057] In step S2.6, when the target position error obtained by different loads at the same time is case 1, the target point position coordinates are not updated; if it is case 2, the target position is updated to the center position coordinates of the small radius circle; if it is case 3 or case 4, the target position is updated to the midpoint of the line segment passing through the intersection of the two circles;

[0058] In step S2.6, when the target position errors obtained by different loads at the same time are within the set identity recognition range, the target position is updated to the center position of the overlapping area of ​​the positioning accuracy error ranges corresponding to the different loads.

[0059] Preferably, the step S7 includes:

[0060] Step S7.1: Calculate the number of lost frames;

[0061] Step S7.2: Calculate the upper and lower thresholds of the target point correlation error;

[0062] Step S7.3: Calculate the upper and lower thresholds of the position coordinate search error along the coordinate system.

[0063] Preferably, the step S8 comprises:

[0064] Step S8.1: input a single associated trajectory, and predict the next moment position based on the last moment position information of the trajectory;

[0065] Step S8.2: Calculate the difference between the detected target position of the current processing frame and the trajectory prediction position along each direction;

[0066] Step S8.3: Determine whether the rough determination condition of non-association is met;

[0067] Step S8.4: for the detected target points that do not meet the unrelated rough determination condition, calculate the error between them and the trajectory prediction position;

[0068] Step S8.5: Compare the position error with the threshold value, and modify the matching metric matrix corresponding to the correlation degree;

[0069] Step S8.6: Continue processing the next formed trajectory for iterative update.

[0070] Preferably, in step S9, two-dimensional allocation is performed according to the calculated matching metric matrix according to the formula:

[0071]

[0072] The optimal matching solution is

[0073] Among them, F * represents the two-dimensional allocation function, m ij Represents the matching metric matrix The (i, j)th element of ij Represents the allocation matrix The (i, j)th element of , and is a binary variable; C * represents the optimal matching solution that satisfies the two-dimensional allocation function, n1 represents the number of trajectories; n2 represents the number of detected target points; (z 1i , z 2j ) represents the matching pair of successfully associated trajectories and targets.

[0074] Preferably, step S10 includes:

[0075] Step S10.1: input a single associated trajectory and determine whether the trajectory is successfully matched in the two-dimensional allocation;

[0076] Step S10.2: For the successfully matched trajectory, find its corresponding matching value in the matching metric matrix;

[0077] Step S10.3: Determine whether the corresponding matching value is greater than a threshold;

[0078] Step S10.4: Perform Kalman filtering on the target position information whose matching value does not exceed the threshold to obtain the predicted position information according to the formula:

[0079]

[0080]

[0081] in, Represents the state prediction vector at time k-1 for time k; represents the state vector at time k-1; i represents the target number at time k-1; TN k-1 Indicates the number of targets at time k-1; Represents the estimated error covariance matrix from time k-1 to time k; represents the posterior estimation error covariance matrix at time k-1; F is the state transfer matrix; F T represents the transpose of the state transfer matrix; Q is the error covariance matrix corresponding to the process noise; c i，j represents the target state prediction value; represents the state vector at time k; represents the inverse of the error covariance matrix; σ T is the tracking prediction gate; j represents the target number at time k; represents the number of targets at time k;

[0082] Step S10.5: Update according to the last position state of the trajectory;

[0083] Step S10.6: Continue processing the next formed trajectory for iterative update.

[0084] An adaptive target association system based on multi-source data, comprising:

[0085] Module M1: used to perform preprocessing operations to obtain the grid division boundary threshold;

[0086] Module M2: used to input the target detection results at the current moment, and perform filtering to remove the repeated position information provided by different payloads of the same target;

[0087] Module M3: used to determine whether the target detection result at the current moment has the same load source as the data at the previous moment, and to accumulate and store the data with the same load source;

[0088] Module M4: used to read the accumulated data obtained by the next load for subsequent processing if the target detection result obtained by the current load is empty;

[0089] Module M5: used to determine whether there is an associated trajectory at present;

[0090] Module M6: used to assemble each target point into a defined format and store it if it does not exist;

[0091] Module M7: used to calculate the adaptive search threshold according to the current load acquisition data and the last position point of the trajectory, and determine the sector search area;

[0092] Module M8: used to further calculate the matching metric matrix between the associated trajectory and the target detection position information of the current frame;

[0093] Module M9: used to perform two-dimensional distribution on the calculated matching metric matrix to obtain the optimal matching result;

[0094] Module M10: used to update the current associated trajectory state by applying a prediction method according to the matching result;

[0095] Module M11: used to determine whether there is an unassociated target point in the current frame;

[0096] Module M12: used to compare the corresponding value of the associated target in the matching metric matrix with the threshold value to determine whether to add a new trajectory, otherwise the unassociated target point is used as the starting point of the new trajectory;

[0097] Module M13: used to update the trajectory set according to the last moment status of each trajectory;

[0098] Module M14: used to wait for the next frame of target detection results to perform iterative processing and update the trajectory set.

[0099] Compared with the prior art, the present invention has the following beneficial effects:

[0100] 1. By projecting the position information obtained by the satellite-borne sensor into the plane coordinate system, and then fusing and filtering the position information of the same target observed by different sensors, the influence of the sensor's own positioning accuracy error is effectively reduced;

[0101] 2. By adaptively generating the search threshold according to the number of lost frames, a fan-shaped search area is formed to further improve the accuracy of target trajectory association;

[0102] 3. Through the access cycle characteristics of satellite-borne sensors, the position information obtained by the same sensor is accumulated, and then the observation data of different sensors are associated to achieve trajectory tracking of the same target, which effectively reduces the interference of sensor positioning accuracy errors, thereby further improving the accuracy of target trajectory association. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0104] Figure 1 It is a flowchart of the adaptive target association method for multi-source data;

[0105] Figure 2 A schematic diagram of a process for obtaining a meshing boundary threshold for preprocessing operations;

[0106] Figure 3 A flowchart of a method for filtering and removing duplicate location information for target detection results;

[0107] FIG4 is a schematic diagram showing the update of the coordinates of the same target position obtained by the medium and low load;

[0108] Figure 5 A flowchart for calculating the matching metric matrix;

[0109] Figure 6 It is a flowchart of the trajectory updating method based on Kalman filtering;

[0110] Figure 7 Corresponding characteristic diagrams of 9 simulated loads;

[0111] Figure 8 It is a set of simulated distribution effect diagrams of the positions of 30 moving targets at sea;

[0112] Fig. 9 for Figure 8 The real motion trajectory effect diagram of the simulated marine moving target;

[0113] Fig.10 for Figure 8 The effect diagram of trajectory tracking processing of simulated moving targets at sea. DETAILED DESCRIPTION

[0114] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0115] According to the characteristics of the access cycle of the onboard sensors, the position information obtained by the same sensor is accumulated, and then the observation data of different sensors are associated to achieve trajectory tracking of the same target, which effectively reduces the interference of sensor positioning accuracy errors and further improves the accuracy of target trajectory association.

[0116] An adaptive target association method based on multi-source data, such as Figure 1 As shown, including:

[0117] Step S1: Perform preprocessing operations to obtain a grid division boundary threshold;

[0118] Among them, in step S1, a preprocessing operation is required to obtain the grid division boundary threshold. The method flow is as follows: Figure 2 As shown, including:

[0119] Step S1.1: Input target motion and load observation characteristic parameters;

[0120] Step S1.2: Determine the segmentation period according to the load observation characteristics;

[0121] Step S1.3: Calculate the moving range of the target within a fixed period according to its motion characteristics;

[0122] Step S1.4: Grid the observation area according to the motion characteristics;

[0123] Step S1.5: Calculate the boundary coordinate information of each grid area;

[0124] Step S1.6: Calculate and store the coordinate coverage threshold of each grid.

[0125] Step S2: Input the target detection result at the current moment, that is, the multi-load target detection information obtained at non-uniform time intervals, and perform filtering to remove the repeated position information provided by different loads of the same target; the method flow is as follows Figure 3 As shown, it includes the following sub-steps:

[0126] Step S2.1: Input the single-frame target detection result and find the corresponding positioning longitude error according to the load source of the target point;

[0127] Step S2.2: traverse the two targets in turn to determine whether they are from the same load source;

[0128] Step S2.3: If the load sources are different, the positions of the two target points to be compared are projected into the plane coordinate system;

[0129] Wherein, the step S2.3 includes the following sub-steps:

[0130] Step S2.3.1: Project the longitude and latitude lines on the earth's surface onto a cylindrical surface, and then unfold the cylindrical surface into a flat map, where the circumference is regarded as the X-axis;

[0131] Step S2.3.2: Convert longitude from degrees to radians according to the following formula:

[0132] x_radian = lon*PI / 180

[0133] Among them, x_radian represents the corresponding radian value, lon represents the longitude value, and PI represents pi;

[0134] Step S2.3.3: Convert the latitude from degrees to radians according to the following formula:

[0135] y_radian = lat*PI / 180

[0136] Among them, y_radian represents the corresponding radian value, lat represents the latitude value, and PI represents pi;

[0137] Step S2.3.4: Perform the Miller projection transformation according to the following formula:

[0138] y_miller=1.25*log(tan(0.25*PI+0.4*y_radian))

[0139] Among them, y miller Represents the conversion value of radians after Miller projection, PI represents pi, and y_radian represents the radian value corresponding to the latitude;

[0140] Step S2.3.5: Convert radians to actual distances according to the following formula:

[0141] x=(W / 2)+(W / (2*PI))*x_radian

[0142] y=(H / 2)+(H / (2*mill))*y_radian

[0143] Among them, x represents the coordinate of the converted x-axis, y represents the coordinate of the converted y-axis, W represents the X-axis, which is equal to the circumference of the earth, PI represents pi, x_radian represents the radian value corresponding to the longitude, H represents the Y-axis, which is approximately equal to half of the circumference of the earth, mill represents a constant in the Miller projection, which ranges approximately between plus and minus 2.3, and y_radian represents the radian value corresponding to the latitude.

[0144] Step S2.4: Calculate the position error between the two target points projected into the plane coordinate system;

[0145] Step S2.5: Determine whether the position error is greater than a threshold;

[0146] Step S2.6: For the target point pairs whose position error does not exceed the threshold, draw circles with the error range as the radius, and update the position of the target point. As shown in Figure 4, it is a schematic diagram of the update of the coordinates of the same target position obtained by different loads. The two points in the figure correspond to the positions of the same target observed by the two loads, and a circle is drawn with each point as the center and the error accuracy of the corresponding load as the radius, where (a) means that the two circles have no overlapping area, (b) means that the two circles have overlapping areas but no intersection, (c) means that the two circles have overlapping areas and the intersection is on the same side of the two circles, and (d) means that the two circles have overlapping areas and the intersection is on the opposite side of the two circles.

[0147] Step S2.6 includes the following cases:

[0148] Case 1: The error ranges of the two target points have no overlapping area;

[0149] Case 2: The error ranges of the two target points have overlapping areas but no intersection;

[0150] Case 3: The error ranges of the two target points overlap and the intersection is on the same side of the two circle centers;

[0151] Case 4: The error ranges of the two target points overlap and the intersection is on the opposite sides of the two circle centers.

[0152] In step S2.6, when the target position error obtained by different loads at the same time is case one, the target point position coordinates are not updated; if it is case two, the target position is updated to the center position coordinates of the small radius circle; if it is case three or case four, the target position is updated to the midpoint of the line segment passing through the intersection of the two circles.

[0153] In step S2.6, when the target position errors obtained by different loads at the same time are within the set identity recognition range, the target position is updated to the center position of the overlapping area of ​​the positioning accuracy error ranges corresponding to the different loads.

[0154] Step S2.7: Continue to process the next frame of target detection results for iterative update.

[0155] Step S3: Determine whether the target detection result at the current moment has the same load source as the data at the previous moment, and accumulate and store the data with the same load source;

[0156] Step S4: If the target detection result obtained by the current load is empty, the accumulated data obtained by the next load is read for subsequent processing;

[0157] Step S5: Determine whether there is currently an associated track;

[0158] Step S6: If it does not exist, each target point is used as the starting point of the trajectory, assembled into a defined format and stored;

[0159] Step S7: otherwise, the adaptive search threshold is calculated based on the current load acquisition data and the last position point of the trajectory to determine the sector search area;

[0160] The search threshold calculation in step S7 includes the following sub-steps:

[0161] Step S7.1: Calculate the number of lost frames, that is, the number of frames without data between two non-empty frames;

[0162] Step S7.2: Calculate the upper and lower thresholds of the target point correlation error;

[0163] Step S7.3: Calculate the upper and lower thresholds of the position coordinate search error along the coordinate system.

[0164] Step S8: further calculating the matching metric matrix between the associated trajectory and the target detection position information of the current frame;

[0165] Among them, the calculation step of the matching metric matrix in step S8 is as follows: Figure 5 As shown, including:

[0166] Step S8.1: input a single associated trajectory, and predict the next moment position based on the last moment position information of the trajectory;

[0167] Step S8.2: Calculate the difference between the detected target position of the current processing frame and the trajectory prediction position along each direction;

[0168] Step S8.3: Determine whether the rough determination condition of non-association is met;

[0169] Step S8.4: for the detected target points that do not meet the unrelated rough determination condition, calculate the error between them and the trajectory prediction position;

[0170] Step S8.5: Compare the position error with the threshold value, and modify the matching metric matrix corresponding to the correlation degree;

[0171] Step S8.6: Continue processing the next formed trajectory for iterative update.

[0172] Step S9: performing two-dimensional distribution on the calculated matching metric matrix to obtain the optimal matching result;

[0173] The two-dimensional allocation is performed according to the calculated matching metric matrix in step S9 according to the following formula:

[0174]

[0175] The optimal matching solution is

[0176] Among them, F * represents the two-dimensional allocation function, m ij Represents the matching metric matrix The (i, j)th element of ij Represents the allocation matrix The (i, j)th element of , and is a binary variable; C * represents the optimal matching solution that satisfies the two-dimensional allocation function, n1 represents the number of trajectories; n2 represents the number of detected target points; (z 1i , z 2j ) represents the matching pair of successfully associated trajectories and targets.

[0177] Step S10: According to the matching result, a prediction method based on Kalman filtering is applied to update the currently associated trajectory state;

[0178] The method steps of step S10 are as follows: Figure 6 As shown, including:

[0179] Step S10.1: input a single associated trajectory and determine whether the trajectory is successfully matched in the two-dimensional allocation;

[0180] Step S10.2: For the successfully matched trajectory, find its corresponding matching value in the matching metric matrix;

[0181] Step S10.3: Determine whether the corresponding matching value is greater than a threshold;

[0182] Step S10.4: Perform Kalman filtering on the target position information whose matching value does not exceed the threshold to obtain the predicted position information according to the following formula:

[0183]

[0184] in, Represents the state prediction vector at time k-1 for time k; represents the state vector at time k-1; i represents the target number at time k-1; TN k-1 Indicates the number of targets at time k-1; Represents the estimated error covariance matrix from time k-1 to time k; represents the posterior estimation error covariance matrix at time k-1; F is the state transfer matrix; F T represents the transpose of the state transfer matrix; Q is the error covariance matrix corresponding to the process noise; c i，j represents the target state prediction value; represents the state vector at time k; represents the inverse of the error covariance matrix; σT is the tracking prediction gate; j represents the target number at time k; represents the number of targets at time k;

[0185] Step S10.5: Update according to the last position state of the trajectory;

[0186] Step S10.6: Continue processing the next formed trajectory for iterative update.

[0187] Step S11: determining whether there is an unassociated target point in the current frame;

[0188] Step S12: for the associated target, compare its corresponding value in the matching metric matrix with the threshold value to determine whether to add a new track, otherwise use the unassociated target point as the starting point of the new track;

[0189] Step S13: updating the trajectory set according to the last moment status of each trajectory;

[0190] Step S14: Wait for the next frame of target detection results to perform iterative processing and update the trajectory set.

[0191] The effects of the present invention can be further illustrated by the following simulation results.

[0192] 1. Simulation data:

[0193] The position information sequence of 30 marine moving targets observed within 23 minutes using 9 types of payloads simulated by STK, and the data processing was completed through PyCharm 2017 platform simulation.

[0194] 2. Simulation content and results:

[0195] In order to verify the effectiveness of the adaptive target association method based on multi-source data provided by the present invention, the position information sequence of 30 marine moving targets observed by 9 types of payloads within 23 minutes was selected for experimental testing. The following is a set of scenes, adding random missed detection and random errors within three kilometers, and comparing the representative experimental original images and processing effect images.

[0196] Figure 4 is a schematic diagram of the update of the coordinates of the same target position obtained by different loads. The two points in the figure correspond to the positions of the same target observed by the two loads respectively. A circle is drawn with each point as the center and the error accuracy of the corresponding load as the radius. (a) indicates that the two circles have no overlapping area, (b) indicates that the two circles have overlapping area but no intersection, (c) indicates that the two circles have overlapping area and the intersection is on the same side of the two circles, and (d) indicates that the two circles have overlapping area and the intersection is on the opposite side of the two circles.

[0197] Figure 7Data characteristics obtained for the nine observed payloads simulated by STK, including payload ID number, positioning accuracy error range in kilometers, and detection rate.

[0198] Figure 8 Figure 2 is a set of simulated distribution diagrams of the positions of 30 moving targets at sea, where (a) is a schematic diagram of the distribution of all targets, and (b), (c) and (d) are enlarged schematic diagrams of the three densely populated areas in (a).

[0199] Fig. 9 for Figure 8 The real motion trajectory renderings of simulated moving targets at sea, where (a) represents the real trajectory rendering of all targets, (b), (c) and (d) respectively represent the enlarged schematic diagrams of the three dense areas in (a).

[0200] Fig.10 for Figure 8 The trajectory tracking processing effect diagram of the simulated moving target at sea, where (a) represents the trajectory effect diagram formed by all associations, (b), (c) and (d) respectively represent the enlarged schematic diagrams of the three dense areas in (a).

[0201] The present invention also provides an adaptive target association system based on multi-source data, which can be implemented by executing the process steps of the adaptive target association method based on multi-source data, that is, those skilled in the art can understand the adaptive target association method based on multi-source data as a preferred implementation of the adaptive target association system based on multi-source data. The system specifically includes the following contents:

[0202] Module M1: Perform preprocessing operations to obtain the grid division boundary threshold.

[0203] Among them, module M1 includes:

[0204] Module M1.1: Input target motion and load observation characteristic parameters;

[0205] Module M1.2: Determine the segmentation period based on the load observation characteristics;

[0206] Module M1.3: Calculate the moving range of the target within a fixed period based on its motion characteristics;

[0207] Module M1.4: Gridding the observation area according to motion characteristics;

[0208] Module M1.5: Calculate the boundary coordinate information of each grid area;

[0209] Module M1.6: Calculate the coordinate coverage threshold of each grid and store it.

[0210] Module M2: Input the target detection result at the current moment, that is, the multi-load target detection information obtained at non-uniform time intervals, and perform filtering to remove the repeated position information provided by different loads of the same target.

[0211] Among them, module M2 includes:

[0212] Module M2.1: Input the single-frame target detection result and find the corresponding positioning longitude error according to the load source of the target point;

[0213] Module M2.2: traverse each pair of targets in turn to determine whether they are from the same load source;

[0214] Module M2.3: If the load sources are different, the positions of the two target points to be compared are projected into the plane coordinate system;

[0215] Among them, the calculation steps of module M2.3 include:

[0216] Module M2.3.1: Project the longitude and latitude lines of the Earth's surface onto a cylindrical surface, and then unfold the cylindrical surface into a flat map, with the circumference considered as the X-axis;

[0217] Module M2.3.2: Convert longitude from degrees to radians according to the following formula:

[0218] x_radian = lon*PI / 180

[0219] Among them, x_radian represents the corresponding radian value, lon represents the longitude value, and PI represents pi;

[0220] Module M2.3.3: Convert latitude from degrees to radians according to the following formula:

[0221] y_radian = lat*PI / 180

[0222] Among them, y_radian represents the corresponding radian value, lat represents the latitude value, and PI represents pi;

[0223] Module M2.3.4: Perform the Miller projection conversion according to the following formula:

[0224] y_miller=1.25*log(tan(0.25*PI+0.4*y_radian))

[0225] Among them, y miller Represents the conversion value of radians after Miller projection, PI represents pi, and y_radian represents the radian value corresponding to the latitude;

[0226] Module M2.3.5: Convert radians to actual distances using the following formula:

[0227] x=(W / 2)+(W / (2*PI))*x_radian

[0228] y=(H / 2)+(H / (2*mill))*y_radian

[0229] Among them, x represents the coordinate of the converted x-axis, y represents the coordinate of the converted y-axis, W represents the X-axis, which is equal to the circumference of the earth, PI represents pi, x_radian represents the radian value corresponding to the longitude, H represents the Y-axis, which is approximately equal to half of the circumference of the earth, mill represents a constant in the Miller projection, which ranges approximately between plus and minus 2.3, and y_radian represents the radian value corresponding to the latitude;

[0230] Module M2.4: Calculate the position error between two target points projected into a plane coordinate system;

[0231] Module M2.5: Determine whether the position error is greater than the threshold;

[0232] Module M2.6: For target point pairs whose position error does not exceed the threshold, draw circles with the error range as the radius and update the position of the target point;

[0233] Module M2.6 includes the following cases:

[0234] Case 1: The error ranges of the two target points have no overlapping area;

[0235] Case 2: The error ranges of the two target points have overlapping areas but no intersection;

[0236] Case 3: The error ranges of the two target points overlap and the intersection is on the same side of the two circle centers;

[0237] Case 4: The error ranges of the two target points overlap and the intersection is on the opposite sides of the two circle centers.

[0238] In module M2.6, when the target position error obtained by different loads at the same time is case 1, the target point position coordinates are not updated; if it is case 2, the target position is updated to the center position coordinates of the small radius circle; if it is case 3 or case 4, the target position is updated to the midpoint of the line segment passing through the intersection of the two circles.

[0239] In module M2.6, when the target position error obtained by different loads at the same time is within the set identity recognition range, the target position is updated to the center position of the overlapping area of ​​the positioning accuracy error range corresponding to the different loads.

[0240] Module M2.7: Continue to process the next frame of target detection results for iterative update.

[0241] Module M3: Determine whether the target detection result at the current moment has the same load source as the data at the previous moment, and accumulate and store the data with the same load source.

[0242] Module M4: If the target detection result obtained by the current load is empty, the accumulated data obtained by the next load is read for subsequent processing.

[0243] Module M5: Determine whether there is an associated trajectory at present.

[0244] Module M6: If it does not exist, each target point is used as the starting point of the trajectory, assembled into a defined format and stored.

[0245] Module M7: Otherwise, the adaptive search threshold is calculated based on the current load acquisition data and the last position point of the trajectory to determine the sector search area.

[0246] The search threshold calculation in module M7 includes the following submodules:

[0247] Module M7.1: Calculate the number of lost frames, that is, the number of frames without data between two non-empty frames;

[0248] Module M7.2: Calculate the upper and lower thresholds of the target point correlation error;

[0249] Module M7.3: Calculate the upper and lower thresholds of the position coordinate search error along the coordinate system.

[0250] Module M8: further calculates the matching metric matrix between the associated trajectory and the target detection position information of the current frame.

[0251] The calculation steps of the matching metric matrix in module M8 include:

[0252] Module M8.1: Input a single associated trajectory and predict the next moment’s position based on the last moment’s position information of the trajectory;

[0253] Module M8.2: Calculate the difference between the detected target position of the current processing frame and the trajectory prediction position along each direction;

[0254] Module M8.3: Determine whether the rough determination condition of non-association is met;

[0255] Module M8.4: For the detected target points that do not meet the unrelated rough judgment conditions, calculate the error between them and the trajectory prediction position;

[0256] Module M8.5: Compare the position error with the threshold value and modify the matching metric matrix corresponding to the correlation degree;

[0257] Module M8.6: Continue to process the next formed trajectory for iterative update.

[0258] Module M9: Perform two-dimensional allocation on the calculated matching metric matrix to obtain the optimal matching result.

[0259] The two-dimensional allocation formula in module M9 based on the calculated matching metric matrix is ​​as follows:

[0260]

[0261] The optimal matching solution is

[0262] Among them, F * represents the two-dimensional allocation function, m ij Represents the matching metric matrix The (i, j)th element of ij Represents the allocation matrix The (i, j)th element of , and is a binary variable; C * represents the optimal matching solution that satisfies the two-dimensional allocation function, n1 represents the number of trajectories; n2 represents the number of detected target points; (z 1i , z 2j ) represents the matching pair of successfully associated trajectories and targets.

[0263] Module M10: According to the matching results, the prediction method based on Kalman filtering is applied to update the current associated trajectory state.

[0264] Among them, the algorithm steps of module M10 include:

[0265] Module M10.1: Input a single associated trajectory and determine whether the trajectory is successfully matched in the two-dimensional allocation;

[0266] Module M10.2: For the successfully matched trajectory, find its corresponding matching value in the matching metric matrix;

[0267] Module M10.3: Determine whether the corresponding matching value is greater than a threshold;

[0268] Module M10.4: Perform Kalman filtering on the target position information whose matching value does not exceed the threshold to obtain the predicted position information according to the following formula:

[0269]

[0270] in, Represents the state prediction vector at time k-1 for time k; represents the state vector at time k-1; i represents the target number at time k-1; TN k-1 Indicates the number of targets at time k-1; Represents the estimated error covariance matrix from time k-1 to time k; represents the posterior estimation error covariance matrix at time k-1; F is the state transfer matrix; F T represents the transpose of the state transfer matrix; Q is the error covariance matrix corresponding to the process noise; c i，j represents the target state prediction value; represents the state vector at time k; represents the inverse of the error covariance matrix; σ T is the tracking prediction gate; j represents the target number at time k; represents the number of targets at time k;

[0271] Module M10.5: Update according to the last position state of the trajectory;

[0272] Module M10.6: Continue to process the next formed trajectory for iterative update.

[0273] Module M11: Determine whether there is an unassociated target point in the current frame;

[0274] Module M12: For the associated target, compare its corresponding value in the matching metric matrix with the threshold value to determine whether to add a new trajectory. Otherwise, take the unassociated target point as the starting point of the new trajectory.

[0275] Module M13: Update the trajectory set according to the last moment status of each trajectory;

[0276] Module M14: Wait for the next frame of target detection results to perform iterative processing and update the trajectory set.

[0277] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0278] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An adaptive target association method based on multi-source data, characterized in that: include: Step S1: Perform preprocessing operations to obtain a grid division boundary threshold; Step S2: input the target detection result at the current moment, and perform filtering to remove the repeated position information provided by different payloads of the same target; Step S3: Determine whether the target detection result at the current moment has the same load source as the data at the previous moment, and accumulate and store the data with the same load source; Step S4: If the target detection result obtained by the current load is empty, the accumulated data obtained by the next load is read for subsequent processing; Step S5: Determine whether there is currently an associated track; Step S6: If it does not exist, each target point is used as the starting point of the trajectory, assembled into a defined format and stored; Step S7: otherwise, the adaptive search threshold is calculated based on the current load acquisition data and the last position point of the trajectory to determine the sector search area; Step S8: further calculating the matching metric matrix between the associated trajectory and the target detection position information of the current frame; Step S9: performing two-dimensional distribution on the calculated matching metric matrix to obtain the optimal matching result; Step S10: According to the matching result, the prediction method is applied to update the current associated trajectory state; Step S11: determining whether there is an unassociated target point in the current frame; Step S12: for the associated target, compare its corresponding value in the matching metric matrix with the threshold value to determine whether to add a new track, otherwise use the unassociated target point as the starting point of the new track; Step S13: updating the trajectory set according to the last moment status of each trajectory; Step S14: Wait for the next frame of target detection results to perform iterative processing and update the trajectory set.

2. The adaptive target association method based on multi-source data according to claim 1, characterized in that: The step S1 comprises: Step S1.1: Input target motion and load observation characteristic parameters; Step S1.2: Determine the segmentation period according to the load observation characteristics; Step S1.3: Calculate the moving range of the target within a fixed period according to its motion characteristics; Step S1.4: Grid the observation area according to the motion characteristics; Step S1.5: Calculate the boundary coordinate information of each grid area; Step S1.6: Calculate and store the coordinate coverage threshold of each grid.

3. The adaptive target association method based on multi-source data according to claim 1, characterized in that: The step S2 comprises: Step S2.1: Input the single-frame target detection result and find the corresponding positioning longitude error according to the load source of the target point; Step S2.2: traverse the two targets in turn to determine whether they are from the same load source; Step S2.3: If the load sources are different, the positions of the two target points to be compared are projected into the plane coordinate system; Step S2.4: Calculate the position error between the two target points projected into the plane coordinate system; Step S2.5: Determine whether the position error is greater than a threshold; Step S2.6: For target point pairs whose position error does not exceed the threshold, draw circles with the error range as the radius and update the position of the target point; Step S2.7: Continue to process the next frame of target detection results for iterative update.

4. The adaptive target association method based on multi-source data according to claim 1, characterized in that: The step S2.3 comprises: Step S2.3.1: Project the longitude and latitude lines on the earth's surface onto a cylindrical surface, and then unfold the cylindrical surface into a flat map, where the circumference is regarded as the X-axis; Step S2.3.2: Convert longitude from degrees to radians according to the following formula: x_radian = lon*PI / 180 Among them, x_radian represents the corresponding radian value, lon represents the longitude value, and PI represents pi; Step S2.3.3: Convert the latitude from degrees to radians according to the following formula: y_radian = lat*PI / 180 Among them, y_radian represents the corresponding radian value, lat represents the latitude value, and PI represents pi; Step S2.3.4: Perform the Miller projection transformation according to the following formula: y_miller=1.25*log tan(0.25*PI+0.4*y_radian)) Among them, y miller Represents the conversion value of radians after Miller projection, PI represents pi, and y_radian represents the radian value corresponding to the latitude; Step S2.3.5: Convert radians to actual distances according to the following formula: x=(W / 2)+W / (2*I))*x_radian y=(H / 2)+H / (2*ill))*y_radian Among them, x represents the coordinate of the converted x-axis, y represents the coordinate of the converted y-axis, W represents the X-axis, which is equal to the circumference of the earth, PI represents pi, x_radian represents the radian value corresponding to the longitude, H represents the Y-axis, which is approximately equal to half of the circumference of the earth, mill represents a constant in the Miller projection, which ranges approximately between plus and minus 2.3, and y_radian represents the radian value corresponding to the latitude.

5. The adaptive target association method based on multi-source data according to claim 1, characterized in that: Step S2.6 includes: Case 1: The error ranges of the two target points have no overlapping area; Case 2: The error ranges of the two target points have overlapping areas but no intersection; Case 3: The error ranges of the two target points overlap and the intersection is on the same side of the two circle centers; Case 4: The error ranges of the two target points overlap and the intersection is on the opposite sides of the two circle centers; In step S2.6, when the target position error obtained by different loads at the same time is case 1, the target point position coordinates are not updated; if it is case 2, the target position is updated to the center position coordinates of the small radius circle; if it is case 3 or case 4, the target position is updated to the midpoint of the line segment passing through the intersection of the two circles; In step S2.6, when the target position errors obtained by different loads at the same time are within the set identity recognition range, the target position is updated to the center position of the overlapping area of ​​the positioning accuracy error ranges corresponding to the different loads.

6. The adaptive target association method based on multi-source data according to claim 1, characterized in that: The step S7 includes: Step S7.1: Calculate the number of lost frames; Step S7.2: Calculate the upper and lower thresholds of the target point correlation error; Step S7.3: Calculate the upper and lower thresholds of the position coordinate search error along the coordinate system.

7. The adaptive target association method based on multi-source data according to claim 1, characterized in that: The step S8 comprises: Step S8.1: input a single associated trajectory, and predict the next moment position based on the last moment position information of the trajectory; Step S8.2: Calculate the difference between the detected target position of the current processing frame and the trajectory prediction position along each direction; Step S8.3: Determine whether the rough determination condition of non-association is met; Step S8.4: for the detected target points that do not meet the unrelated rough determination condition, calculate the error between them and the trajectory prediction position; Step S8.5: Compare the position error with the threshold value, and modify the matching metric matrix corresponding to the correlation degree; Step S8.6: Continue processing the next formed trajectory for iterative update.

8. The adaptive target association method based on multi-source data according to claim 1, characterized in that: In step S9, two-dimensional allocation is performed according to the calculated matching metric matrix according to the formula: The optimal matching solution is Among them, F * represents the two-dimensional allocation function, m ij Represents the matching metric matrix The (i, j)th element of ij Represents the allocation matrix The (i, j)th element of , and is a binary variable; C * represents the optimal matching solution that satisfies the two-dimensional allocation function, n1 represents the number of trajectories; n2 represents the number of detected target points; (z 1i , z 2j ) represents the matching pair of successfully associated trajectories and targets.

9. The adaptive target association method based on multi-source data according to claim 1, characterized in that: The step S10 comprises: Step S10.1: input a single associated trajectory and determine whether the trajectory is successfully matched in the two-dimensional allocation; Step S10.2: For the successfully matched trajectory, find its corresponding matching value in the matching metric matrix; Step S10.3: Determine whether the corresponding matching value is greater than a threshold; Step S10.4: Perform Kalman filtering on the target position information whose matching value does not exceed the threshold to obtain the predicted position information according to the formula: in, Represents the state prediction vector at time k-1 for time k; represents the state vector at time k-1; i represents the target number at time k-1; TN k-1 Indicates the number of targets at time k-1; Represents the estimated error covariance matrix from time k-1 to time k; represents the posterior estimation error covariance matrix at time k-1; F is the state transfer matrix; F T represents the transpose of the state transfer matrix; Q is the error covariance matrix corresponding to the process noise; c i，j represents the target state prediction value; represents the state vector at time k; represents the inverse of the error covariance matrix; σ T is the tracking prediction gate; j represents the target number at time k; represents the number of targets at time k; Step S10.5: Update according to the last position state of the trajectory; Step S10.6: Continue processing the next formed trajectory for iterative update.

10. An adaptive target association system based on multi-source data, characterized in that: include: Module M1: used to perform preprocessing operations to obtain the grid division boundary threshold; Module M2: used to input the target detection results at the current moment, and perform filtering to remove the repeated position information provided by different payloads of the same target; Module M3: used to determine whether the target detection result at the current moment has the same load source as the data at the previous moment, and to accumulate and store the data with the same load source; Module M4: used to read the accumulated data obtained by the next load for subsequent processing if the target detection result obtained by the current load is empty; Module M5: used to determine whether there is an associated trajectory at present; Module M6: used to assemble each target point into a defined format and store it if it does not exist; Module M7: used to calculate the adaptive search threshold according to the current load acquisition data and the last position point of the trajectory, and determine the sector search area; Module M8: used to further calculate the matching metric matrix between the associated trajectory and the target detection position information of the current frame; Module M9: used to perform two-dimensional distribution on the calculated matching metric matrix to obtain the optimal matching result; Module M10: used to update the current associated trajectory state by applying a prediction method according to the matching result; Module M11: used to determine whether there is an unassociated target point in the current frame; Module M12: used to compare the corresponding value of the associated target in the matching metric matrix with the threshold value to determine whether to add a new trajectory, otherwise the unassociated target point is used as the starting point of the new trajectory; Module M13: used to update the trajectory set according to the last moment status of each trajectory; Module M14: used to wait for the next frame of target detection results to perform iterative processing and update the trajectory set.

Images