A Real-time and Efficient Method for Fusing Heterogeneous Target Track Data

Abstract

The present invention discloses a real-time and efficient method for fusing heterogeneous target track data, including: obtaining the target track data of all detection devices and putting them into the first message queue; subscribing to the target track data to obtain the corresponding destination number; performing batch processing on the target track data based on a preset time window: grouping according to the destination number to form a target list; matching the historical target track data of the targets corresponding to each target list; identifying the targets with lost tracks for the targets that do not match; determining the identity of any two targets and performing fusion matching; performing error correction on the target track data corresponding to the same target based on Kalman filter smoothing processing; evaluating the target threat level. This method can improve the operation speed and accuracy, better adapt to the quantity uncertainty and uneven time distribution of the target track data, accurately identify the true quantity and true movement trajectory of the targets, improve the accuracy of target evaluation, facilitate maintenance, and has good versatility.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a real-time and efficient fusion method for heterogeneous target track data. Background Art

[0002] In a key area prevention and control system, generally, security levels and core areas are divided for the entire area. And intrusion target detection and tracking devices and target disposal devices corresponding to different levels of defense areas are equipped. Usually, in order to ensure the reliability of the prevention and control system and avoid monitoring blind spots, a large number of target detection and tracking devices are deployed, such as optoelectronic devices and radar devices. These devices detect and track targets independently or cooperatively and report the track data of the detected targets in real time. In this context, a key problem that the upper-layer information processing system needs to handle is the fusion problem of target track data. Specifically, for a target in the same intrusion prevention area, multiple different types of devices that independently observe and track the target will continuously report the real-time position information of this target. And during this process, the batch numbering of the target by different devices is not unified, resulting in multiple pieces of track data of this target received by the information processing system. If not processed, there will be a large error in the number of threat targets in the current prevention area and the threat assessment of the entire prevention area by the system. Thus, it affects the reliability of the prevention and control system and ultimately affects the security of the entire prevention area. In addition, there are errors in the observation of a target by a single device. If the observation data of a single device is used, there will also be a large error between the position of the target and its true spatial position.

[0003] Regarding the track fusion problem, the following solutions exist in the prior art:

[0004] Chinese Patent with Application No. 201910987019.2 discloses a radar optoelectronic information fusion technology, and proposes a scheme for fusing radar and optoelectronic information by matching pixel coordinates and geographic coordinates. This scheme first uses a deep learning method to identify the position of a target in the field of view of an optoelectronic device, and then performs conversion and comparison with all target positions detected by the radar and then conducts cooperative tracking control. However, the fusion method combined with image processing and an artificial intelligence inference system has problems such as complex operations and high hardware requirements. The system construction requires a high hardware configuration. At the same time, there are also large uncertainties and low reliability in the matching between target track data and the targets identified from audio and video data by AI algorithms.

[0005] In the prior art, a fusion method based on spatio-temporal alignment of target tracks is also proposed. The approach is to group a large amount of acquired target track data by target batch number, and perform position point calculation on the target data reported by devices with a relatively high data reporting frequency, so that the final point track data is aligned in time to the time point reported by devices with less target data. Then, fusion calculation is performed on the multi-device reported data with aligned time points. However, the track matching method based on spatio-temporal alignment has a large dependence on the data volume of target track data. In this solution, if the track data reported by the devices for spatio-temporal alignment modeling is unevenly distributed or scarce, it will seriously affect the accuracy and reliability of the fusion result.

[0006] In addition, a target track fusion method based on device linkage is also proposed. Its implementation scheme is to perform fusion calculation on the data reported by radar devices and optoelectronic devices with linkage tracking relationships. In this scenario, first, the radar discovers a suspicious target, and then sends the target information to the optoelectronic device to participate in linkage tracking. However, the track fusion based on device linkage has business limitations and poor generality. It can only perform fusion calculation on the data reported by devices with established linkage relationships; the method for fusing the observations of the same target by multiple radars also has business limitations. The prerequisite is that it is known that multiple radars observe the same target and the identity of its target track is known. It is not applicable to the scenario of a large amount of data reported in real time by multiple devices in a complex environment. Summary of the Invention

[0007] The object of the present invention is to propose a real-time and efficient fusion method for heterogeneous target track data in view of the above problems. This method helps to ensure the real-time nature and high availability of data, improve the operation speed and accuracy, and can better adapt to the quantity uncertainty and uneven time distribution of target track data, accurately identify the true quantity of targets in the defense area and restore the true movement track of the targets, improve the accuracy of target threat level assessment, and is easy to maintain with good generality.

[0008] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A real-time and efficient fusion method for heterogeneous target track data proposed by the present invention includes the following steps:

[0010] S1. Obtain the target track data of all detection devices and put them into the first message queue. The target track data includes the detection device number, the original target batch number, the target position, and the timestamp;

[0011] S2. Subscribe to the target track data in the first message queue, and calculate the MD5 value according to the detection device number and the original target batch number as the destination number of the corresponding target track data;

[0012] S3. Batch process the target track data based on a preset time window. The processing procedure for each batch of target track data is as follows:

[0013] S31. Group according to the destination number, and each group of target track data forms a target list;

[0014] S32. Match the targets corresponding to each target list with historical target track data, and perform the following operations:

[0015] S321. Obtain all cached historical destination numbers and put them into the first data set;

[0016] S322. Determine whether the destination number corresponding to the target in the current time window exists in the cache. If so, it is considered that there is a historical track. Remove the destination number corresponding to the target from the first data set, cache the target track data corresponding to the target in the current time window, and insert the historical target track data corresponding to the target into the target list corresponding to the target in the current time window to form a new target list. Otherwise, it is considered that there is no historical track, and put the destination number and target track data corresponding to the target in the current time window into the second data set and the cache respectively;

[0017] S323. Retrieve the fusion batch number in the cache for the targets existing in the current time window. If a target has obtained a fusion batch number, use this fusion batch number as the latest fusion batch number corresponding to the target and put it into the third data set;

[0018] S33. Identify the targets with lost tracks for the targets corresponding to the remaining destination numbers in the first data set, and perform the following operations:

[0019] S331. Traverse the first data set to determine whether there is a track model for each remaining target. If so, load the corresponding track model. Otherwise, create and train a track model based on the historical target track data of the corresponding target, and store the trained track model in the model set and the cache;

[0020] S332. Traverse the second data set to obtain the target track data of each target in the new target list;

[0021] S333. Use the loaded track model or the trained track model to predict the target position aligned with the time point of the target track data corresponding to the target, and calculate the point track average error ε:

[0022]

[0023] where n is the data volume of the new target list where the target is located, is the predicted longitude of the target at time point i, is the actual longitude of the target at time point i, is the predicted latitude of the target at time point i, is the actual latitude of the target at time point i, is the predicted altitude of the target at time point i, is the actual altitude of the target at time point i;

[0024] S334. Determine whether the average error ε of the track points is less than the target threshold. If so, consider the corresponding target in the current time window as a historical target with lost track, and perform batch number fusion processing on the corresponding target in the current time window and the historical target. Otherwise, consider the corresponding target in the current time window as a newly emerged target;

[0025] S34. Determine the identity of any two targets and perform fusion matching, specifically as follows:

[0026] S341. Denote one of the targets as the model reference target and the other target as the alignment target, and select a model algorithm for the model reference target;

[0027] S342. Based on the selected model algorithm, establish a prediction model for the model reference target, and train it using the target track data in the new target list where the model reference target is located to obtain the final prediction model;

[0028] S343. Based on interpolation or extrapolation, predict the target position aligned with the target track data time point of the alignment target on the final prediction model;

[0029] S344. Determine whether the average distance error between the predicted target position and the target position of the alignment target is less than the preset distance threshold. If so, the target meets the identity, and it is determined as the same target. Check whether the target has been assigned a fusion batch number. If any target has a fusion batch number, apply the fusion batch number transfer logic to share this fusion batch number. If both targets have not been assigned a fusion batch number, use the purpose number of the earlier emerged target as the shared fusion batch number, and save the assigned fusion batch number to the third data set and the cache; otherwise, the target does not meet the identity, and add non-fusion identifiers for the two targets to the cache;

[0030] S35. Group the target track data of the same target in the current time window into a group, and perform error correction on the target track data corresponding to each target after this grouping based on Kalman filter smoothing processing;

[0031] S36. Evaluate the threat level of the corresponding target based on the error-corrected target track data, and output the threat level score to the second message queue for subscription.

[0032] Preferably, in step S2, subscribing to the target track data in the first message queue is implemented based on the Flink computing framework.

[0033] Preferably, in step S31, anomaly detection and verification processing are also performed on each target list, and abnormal target track data is deleted, specifically as follows:

[0034] S311. Determine whether the target position of the target track data exceeds the preset range. If so, delete the target track data; otherwise, execute step S312;

[0035] S312. Obtain the current system time after time synchronization and read the timestamp of the target track data. Determine whether the difference between the time of the target track data and the current system time exceeds the preset time delay threshold or whether the time of the target track data exceeds the average time of the target track data in the target list. If so, delete the target track data; otherwise, execute step S313;

[0036] S313. Calculate the average speed of the target track data in the target list, and calculate the instantaneous speed of each target track data based on the position difference and time difference between adjacent points. Determine whether the instantaneous speed is higher than the average speed. If so, delete the target track data; otherwise, retain it.

[0037] Preferably, in step S322, the following operations are also performed on the new target list:

[0038] When the data volume in the new target list is greater than the preset retention value, delete the historical target track data at the front until the data volume reaches half of the preset retention value.

[0039] Preferably, in step S331, a track model is created and trained based on the historical target track data of the corresponding target, specifically as follows:

[0040] S3311. Establish models for the three dimensions of longitude, latitude, and altitude of the target based on the least squares method respectively. The formulas are as follows:

[0041]

[0042] where t is the time, f(t) is the target function, a x is the constant term coefficient of the x-th term of the polynomial, and M is the maximum order;

[0043] S3312. Determine whether pointNum > max-degree + 1 is satisfied, where pointNum is the data volume of the new target list where the target is located, and max-degree is the preset maximum order. If so, the maximum order M = max-degree; otherwise, the maximum order M = pointNum - 1;

[0044] S3313. Train the models for the three dimensions of the target based on the corresponding historical target track data using the fit method, and regard the trained three models as the track model.

[0045] Preferably, in step S334, the batch number fusion process is as follows:

[0046] Retrieve and determine from the third data set and the cache whether the corresponding historical target has been assigned a fusion batch number. If so, store the mapping information between the original target batch number of the corresponding target and the fusion batch number assigned to the historical target in the third data set and the cache. Otherwise, reassign the fusion batch number.

[0047] Preferably, in step S34, determine the identity of any two targets and perform fusion matching, and also perform the following operations:

[0048] SA1. Determine whether the two targets are from the same detection device. If so, consider that the targets do not meet the identity. Otherwise, perform step SA2;

[0049] SA2. Determine whether the two targets have been respectively assigned fusion batch numbers. If so, consider that the targets do not meet the identity. Otherwise, perform step SA3;

[0050] SA3. Determine whether there is a non-fusion flag for the two targets. If so, consider that the targets do not meet the identity. Otherwise, perform step S341.

[0051] Preferably, in step S341, mark one of the targets as the model reference target and the other target as the alignment target, and select a model algorithm for the model reference target, as follows:

[0052] When the data volume in the new target list where the two targets are located is equal to 2 for both, the model algorithm selects a linear function model, and uses the target with a larger time span as the model reference target;

[0053] When the data volume in the new target list where the two targets are located is equal to 2 and 3 respectively, the model algorithm selects a parabola function model, and uses the target with a larger data volume as the model reference target;

[0054] When the data volume in the new target list where the two targets are located is equal to 3 for both, the model algorithm selects a parabola function model and selects any target as the model reference target;

[0055] When the data volume in the new target list where the two targets are located is greater than 3 for both, obtain the data volume of the two targets that fall within each other's time range. If the data volume that falls within each other's time range is greater than the first preset threshold and the data volume of the other party is greater than the second preset threshold, then the model algorithm selects a cubic spline interpolation model and uses the other party as the model reference target. Otherwise, the model algorithm selects a least squares method model and uses the target with a larger data volume as the model reference target.

[0056] Preferably, in step S36, the threat level score final-threat-score is calculated as follows:

[0057]

[0058]

[0059] where F y (FAC y ) is the threat assessment score of the threat factor FAC y , α y is the target threat weight coefficient, and N is the total number of threat factors.

[0060] Preferably, the threat factor FAC y includes the position of the target from the center point of the defense area, the real-time speed of the target, the target type, the working state of the target, the course angle of the target, and the security level of the center point of the defense area.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] 1) The unified batch numbering and fusion batch number allocation of the target batch numbers reported by the detection device have better understandability and system maintainability compared with the method that strongly relies on device identification information;

[0063] 2) Based on the dynamic time window, the reported target track data is pre-grouped. More target track data of the same target can be obtained within a processing time window, which helps to improve the accuracy of the fusion calculation result. At the same time, the method based on the association of historical target track data is used, which can well solve the problem of the fluctuation of the real-time result accuracy caused by insufficient target data volume within the current time window, and can avoid the accidental error caused by calculating each target track one by one and reduce the performance overhead of the information processing system, ensuring the real-time and high availability of the data;

[0064] 3) By caching and retrieving the historical target track data of the target track, the historical target track data is dynamically allocated to participate in the track fusion calculation. On the one hand, it supplements the data volume in the current time window. On the other hand, for the case of a large amount of historical data, the historical data is excluded and discarded based on the quantity and time, avoiding the overfitting effect of the long-term historical track data on the result, supporting the identification and association of lost target tracks, and being able to well make up for the target statistics and analysis capabilities in the case of hardware deployment and abnormal device working states at the software level;

[0065] 4) By dynamically adding and expanding the abnormal point checking and verification process to exclude abnormal data points, the interference is reduced, which helps to improve the operation speed and accuracy;

[0066] 5) Compared with the existing fusion methods based on device linkage and the single-type device data fusion methods, this method has stronger generality, is not restricted by the device type and the number of devices. As long as the devices can stably report the target track data, they can be uniformly processed, and there are no special requirements for the frequency of the device-reported data;

[0067] 6) Compared with the target track identity recognition in the traditional fusion methods, this method proposes the concept of target fusion batch number transfer through data cache matching, greatly reducing the amount of identity calculation and improving the system data processing ability;

[0068] 7) Compared with the existing fusion methods based on target track spatio-temporal alignment, this method can also realize the recognition of the same target track based on non-spatio-temporally aligned data. At this stage, all the original target data is not selected or discarded, and the respective real reported data is used during the same fusion, which can make more full use of the original target track data of each detection device and improve the authenticity of the track;

[0069] 8) A perfect fusion data retrieval system. During the process of lost target recognition and fusion matching of the track, there is no need to frequently and repeatedly calculate data matching and selection, reducing unnecessary complex core calculation logics;

[0070] 9) The dynamic allocation logic of the track model during the target identity determination process, and at the same time, there is a dynamic model reference target selection strategy. Compared with a single matching logic, it can better adapt to the quantity uncertainty and uneven time distribution of the target track data, improve the accuracy, and make the same determination result more reliable;

[0071] 10) Based on the Kalman filter smoothing processing method, the full amount of target tracks after identity determination is subjected to track smoothing processing and position error correction, and the targets with the same identity are fused and matched to realize the associated recognition calculation of multiple tracks of the same target, which can accurately identify the real quantity of the targets in the defense area and restore the real movement track of the targets, which is beneficial to improving the evaluation accuracy when realizing the all-round dynamic and expandable target threat level evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is the flow chart of the real-time and efficient fusion method for heterogeneous target track data of the present invention;

[0073] Figure 2 It is the flow chart of the abnormal point picking and verification processing of the present invention;

[0074] Figure 3 It is the flow chart of the matching of historical target track data of the present invention;

[0075] Figure 4 It is the flow chart of the lost target recognition of the present invention;

[0076] Figure 5 It is a flow chart for target identity determination and fusion matching of the present invention;

[0077] Figure 6 It is a schematic diagram for threat level assessment of the present invention;

[0078] Figure 7 It is a configuration interface diagram for threat factor assessment of the present invention;

[0079] Figure 8 It is a configuration interface diagram for factor weight configuration of the present invention. Detailed implementation manners

[0080] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0081] It should be noted that unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used in the specification of this application herein are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0082] The corresponding term interpretations in this application are as follows:

[0083] Track: The trajectory of target movement. It is formed by a series of target track data with timestamps and target batch numbers.

[0084] Time window: The time interval for each batch data processing operation.

[0085] Target batching: The batch number arrangement performed by the detection device on the targets it discovers itself. It ensures the unique identification of the targets on the detection device itself.

[0086] As Figure 1-8 shown, a real-time and efficient fusion method for heterogeneous target track data includes the following steps:

[0087] S1. Obtain the target track data of all detection devices and put them into the first message queue. The target track data includes the detection device number, the original target batch number, the target position, and the timestamp.

[0088] Among them, the target track data reported by all detection devices (such as optoelectronic devices, radar devices, etc.) within the defense area enters the first message queue (in). Through real-time subscription and consumption of the target track data from the first message queue (in), a large amount of heterogeneous target track data is processed in real time and efficiently. The processed data will be published to the second message queue (out) for downstream systems (such as command and display control systems, etc.) to subscribe and consume. The original target batch number carried by the target track data is the batch number arranged by the detection device for the targets it discovers itself, ensuring the unique identification of the target on the detection device itself. The target position carried by the target track data, such as including the longitude, latitude, altitude, speed, heading angle, type, etc. of the target, can be determined according to actual requirements.

[0089] S2. Subscribe to the target track data in the first message queue, and calculate the MD5 value based on the detection device number and the original target batch number as the target number corresponding to the target track data.

[0090] In one embodiment, in step S2, subscribing to the target track data in the first message queue is implemented based on the Flink computing framework.

[0091] Among them, since the target batch numbers assigned by different detection devices to different targets may be repeated, the original target batch numbers in the target track data reported by the detection devices do not have unique identification. By considering combining the detection device number (deviceId) with the original target batch number (oldtargetNo) to generate a target batch number that uniquely identifies the target within the system. To avoid the generated target batch number showing strong semantic association with a specific device number in form, this application calculates the MD5 value based on the detection device number (deviceId) and the original target batch number (oldtargetNo) as the target unique batch number of the corresponding target track data, that is, the target number targetNo = MD5(deviceId_targetNo) as the target unique batch number. For example, take the first eight digits as the target number, or adjust according to actual requirements.

[0092] S3. Perform batch processing on the target track data based on a preset time window. The processing process for each batch of target track data is as follows:

[0093] S31. Group according to the target number, and each group of target track data forms a target list.

[0094] In one embodiment, in step S31, anomaly detection and verification processing are also performed on each target list, and abnormal target track data is deleted. Specifically as follows:

[0095] S311. Determine whether the target position of the target track data exceeds the preset range. If so, delete the target track data. Otherwise, execute step S312;

[0096] S312. Obtain the current system time through the timekeeping system and read the timestamp of the target track data. Determine whether the difference between the time of the target track data and the current system time exceeds the preset time delay threshold or whether the time of the target track data exceeds the average time of the target track data in the target list. If so, delete the target track data; otherwise, execute step S313.

[0097] S313. Calculate the average speed of the target track data in the target list, and calculate the instantaneous speed of each target track data based on the position difference and time difference between adjacent points. Determine whether the instantaneous speed is higher than the average speed. If so, delete the target track data; otherwise, retain it.

[0098] Among them, the target tracks are uniformly batch-processed using a time window, and a dynamic time window mechanism is adopted to process the data. The default time window size is 500 ms, that is, a batch of target track data is processed every 500 ms. The dynamic adjustment of the time window can also be implemented according to actual requirements. Generally, the reporting frequency of the detection device for target track data is 20 HZ - 50 HZ, that is, once every 50 ms - 20 ms. And the allowable range of the data refresh frequency of the downstream command and display control system is generally reasonable at the order of 1 hz.

[0099] Within a time window, the target track data is grouped according to the destination number targetNo. Each destination number corresponds to multiple target track data to form a target list TargetList. All target lists TargetList within the time window form a target map targetMap.

[0100] Perform abnormal point selection and verification processing on the target track data in the target map targetMap (that is, including all target lists TargetList). Specifically as follows:

[0101] 1) Target position (longitude, latitude, altitude) verification: The longitude range around the current point in the defense area is [lon-west, lon-east], the latitude range is [la-south, la-north], and the altitude range is [hei-min, hei-max]. The above configuration items can be read from the system configuration. When at least one of the longitude, latitude, and altitude exceeds the preset range, the corresponding target track data is directly excluded.

[0102] 2) Time verification: Obtain the current system time through the timekeeping system, and read the timestamp of the target track data. If the difference between the time of the target track data and the current system time exceeds the preset time delay threshold or the time of the target track data exceeds the average time of the target track data in the target list, the corresponding target track data is directly excluded.

[0103] 3) Speed verification: Calculate the average speed of the target track data in the target list, and calculate the instantaneous speed of each target track data based on the position difference and time difference between adjacent points. If the instantaneous speed is higher than the average speed, directly remove the corresponding target track data.

[0104] It should be noted that the order of target position verification, time verification, and speed verification can be determined according to actual needs. Preferably, target position verification, time verification, and speed verification are performed in sequence.

[0105] S32. Match the historical target track data for the targets corresponding to each target list, and perform the following operations:

[0106] S321. Obtain all the cached historical destination numbers and put them into the first data set.

[0107] S322. Determine whether there is a destination number corresponding to the target in the current time window in the cache. If so, it is considered that there is a historical track. Remove the destination number corresponding to the target from the first data set, cache the target track data corresponding to the target in the current time window, and at the same time insert the historical target track data corresponding to the target into the target list corresponding to the target in the current time window to form a new target list. Otherwise, it is considered that there is no historical track, and the destination number and target track data corresponding to the target in the current time window are respectively put into the second data set and the cache.

[0108] S323. Retrieve the fusion batch number for the targets existing in the current time window in the cache. If a target has obtained a fusion batch number, use this fusion batch number as the latest fusion batch number for the corresponding target and put it into the third data set.

[0109] In one embodiment, in step S322, the following operations are further performed on the new target list:

[0110] When the data volume in the new target list is greater than the preset retention value, delete the prior historical target track data until the data volume reaches half of the preset retention value.

[0111] Among them, the historical target track data matching is to obtain more sufficient historical target track data and the existing fusion batch numbers of the targets. Specifically, first, all historical target numbers existing in the cache are obtained and put into the first data set cacheedTarNoSet. Subsequently, the target number targetNo of the target in the current time window is used to retrieve the cache to check whether the historical track exists. If the historical track does not exist, the target number is put into the second data set newTagNoSet. The data to be stored in this set are all the newly emerged targets in the current window. And the target track data corresponding to the current time window is put into the cache as the historical target track data for subsequent time windows. If the historical track exists, the target number corresponding to the current time window in the first data set cacheedTarNoSet is removed, and the newly obtained target track data of this target in the current time window is appended to the cache. At the same time, the historical target track data is inserted into the target list targetList of the corresponding target in the current time window to form a new target list. Further, the historical target track data can be the data after the abnormal point picking and verification process for the corresponding target list.

[0112] Since there may be many situations for the historical target track data, and the historical target track data that is too old has little reference significance for the current new movement trend and even has a negative interference effect. Therefore, it is necessary to set a preset retention value MAX_TARGET_HISTORY_POINT_NUM. If the data volume in the new target list is greater than the preset retention value, the historical target track data of the corresponding target is removed. And here it is set that the retained data volume after removal is MAX_TARGET_HISTORY_POINT_NUM / 2, which can avoid frequent interception and update of the cached data when the data volume reaches the upper limit. After removal, the new target list and the cache are updated synchronously. Then, a fusion batch number retrieval is performed on the target number targetNo of the target in the current time window. If this target number has obtained a fusion batch number, the obtained fusion batch number is directly read as the latest fusion batch number of the target, and the obtained fusion batch number is put into the third data set tagFuseNoMap.

[0113] In this stage, the first data set cacheedTarNoSet is used to store the target numbers of the targets that no longer appear in the current time window among the historically appeared target numbers; the second data set newTagNoSet is used to store the newly emerged target numbers in the current time window; the target map targetMap is used to store the targets that appear in the current time window and their associated historical target track data.

[0114] S33. Identify the targets with lost tracks for the targets corresponding to the remaining target numbers in the first data set, and perform the following operations:

[0115] S331. Traverse the first data set, and determine whether there is a trajectory model for each remaining target. If so, load the corresponding trajectory model; otherwise, create and train a trajectory model based on the historical target trajectory data of the corresponding target, and store the trained trajectory model in the model set and cache.

[0116] In one embodiment, in step S331, creating and training a trajectory model based on the historical target trajectory data of the corresponding target is specifically as follows:

[0117] S3311. Establish models for the three dimensions of longitude, latitude, and altitude of the target based on the least squares method respectively. The formulas are as follows:

[0118]

[0119] where t is time, f(t) is the objective function, and a x is the constant term coefficient of the x-th term of the polynomial, and M is the maximum order;

[0120] S3312. Determine whether pointNum > max-degree + 1 is satisfied, where pointNum is the amount of data in the new target list where the target is located, and max-degree is the preset maximum order. If so, the maximum order M = max-degree; otherwise, the maximum order M = pointNum - 1;

[0121] S3313. Train the models for the three dimensions of the target based on the corresponding historical target trajectory data using the fit method, and regard the trained three models as the trajectory model.

[0122] S332. Traverse the second data set, and obtain the target trajectory data of each target in the new target list where the target is located;

[0123] S333. Use the loaded trajectory model or the trained trajectory model to predict the target position aligned with the time point of the target trajectory data of the corresponding target, and calculate the point trace average error ε:

[0124]

[0125] where n is the amount of data in the new target list where the target is located, is the predicted longitude of the target at time point i, is the actual longitude of the target at time point i, is the predicted latitude of the target at time point i, is the actual latitude of the target at time point i, is the predicted altitude of the target at time point i, is the actual altitude of the target at time point i;

[0126] S334. Determine whether the average error ε of the track points is less than the target threshold. If so, consider the corresponding target in the current time window as a historical target with lost track, and perform batch number fusion processing on the corresponding target in the current time window and the historical target. Otherwise, consider the corresponding target in the current time window as a newly emerged target.

[0127] In one embodiment, in step S334, the batch number fusion processing is specifically as follows:

[0128] Retrieve from the third data set and the cache to determine whether the corresponding historical target has been assigned a fusion batch number. If so, store the mapping information between the original target batch number of the corresponding target and the fusion batch number assigned to the historical target in the third data set and the cache. Otherwise, reassign a fusion batch number.

[0129] Among them, for a target that continuously moves within the defense area and has been observed and tracked by the detection device at a previous moment, but later, due to reasons such as the blind area or failure of the detection device, the tracking and observation of this target are lost, that is, there is no data of this target for a period of time, but in the current time window, the observation data of this target by the detection device reappears. Since its purpose number has changed, it is impossible to directly perform historical target track data association and fusion batch number association. Therefore, it is necessary to associate this target with its historical track through this track loss target recognition algorithm.

[0130] Traverse the first data set cacheedTarNoSet and the second data set newTagNoSet, retrieve the track model corresponding to each target number targetNo in the first data set cacheedTarNoSet. The retrieval order is to first retrieve from the model set modelMap. If not, load from the cache. If it does not exist in the cache either, use the historical target track data corresponding to the target number targetNo to create and train a track model, store it in the model set modelMap and put it into the cache. The training method of the track model is to train a two-dimensional least squares model in three dimensions of (time, latitude), (time, longitude), and (time, height) for the sequence formed by each target track data in the target track in time. Among them, a x is the constant term coefficient of the x-th term of the polynomial. For example, when the maximum order M = 1, it corresponds to the well-known linear function in the art. When the maximum order M = 2, it corresponds to the well-known parabola function in the art.

[0131] Three models related to longitude, latitude, altitude and time obtained through calculation. The specific method for training the three models uses the fit method of PolynomialCurveFitter in the Java package org.apache.commons.math3.fitting. Regarding the order of the polynomial function of the track curve in the track model, it is determined by configuring a functional preset maximum order max-degree. In this embodiment, the default value is max-degree = 6, or it can also be adjusted according to actual needs. If the data volume pointNum of the new target list where the target is located > max-degree + 1, the end of the polynomial function of the track curve is the maximum order M = max-degree. Otherwise, the maximum order M = pointNum - 1. The three two-dimensional models of longitude, latitude, altitude related to time obtained by fit calculation are assembled into a custom model entity, which is the track model. The obtained track model is used for point track position prediction calculation, and the target position aligned with the target track data time point of the new target list where the corresponding target is located under this historical track model is calculated to obtain the point track average error ε.

[0132] If the point track average error ε is less than the target threshold, it is considered that the corresponding target in the current time window is a historical target with lost track, and the corresponding target in the current time window and the historical target are subjected to batch number fusion processing. Otherwise, it is considered that the corresponding target in the current time window is a newly emerged target. When performing batch number fusion processing, considering the subsequent time, the observation data about this target is reported using a new purpose number for a period of time. If the purpose number is directly unified into the historical purpose number, it will cause a large number of repeated calls to the entire track lost target recognition logic. And a large number of repeated calculations will lead to a decline in the system processing ability. The strategy adopted here is to apply the target fusion logic to the current target and the corresponding historical target. For the convenience of description, the current new purpose number is denoted as tagNo1, and the corresponding recognized historical purpose number of the lost track is denoted as tagNo0. The core is the target batch number fusion logic. The specific operation is as follows: First, retrieve the fusion batch number fuseNo assigned to tagNo0 from the third data set tagFuseNoMap and the cache. If it exists, the mapping information (oldtagNo1 - fuseNo) between the current original target batch number oldtagNo1 of the corresponding target and the fusion batch number fuseNo assigned to the historical target is stored in the third data set tagFuseNoMap and the cache. As for the target track data corresponding to tagNo1, no other processing is required at this stage. In this way, it is possible to avoid frequent track lost target recognition calculations when this target arrives later. When this target arrives later, first, according to the purpose number, the data that has appeared in the current time window is used as the historical target track data. At the same time, since the fusion batch number already exists, it can be directly retrieved without repeated calculation, nor repeated target batch number association and conversion.

[0133] S34. Determine the identity of any two targets and perform fusion matching as follows:

[0134] In one embodiment, the following operations are also performed:

[0135] SA1. Determine whether two targets are from the same detection device. If so, it is considered that the targets do not meet the identity requirement. Otherwise, execute step SA2;

[0136] SA2. Determine whether fusion batch numbers have been assigned to the two targets respectively. If so, it is considered that the targets do not meet the identity requirement. Otherwise, execute step SA3;

[0137] SA3. Determine whether there is a non-fusion flag for the two targets. If so, it is considered that the targets do not meet the identity requirement. Otherwise, execute step S341.

[0138] Since the target identity determination algorithm requires relatively complex calculations, in order to improve data processing performance, it is necessary to intercept and screen the data entering the identity determination, filter unnecessary calculations. For the targets appearing in the current time window, when the target numbers are inconsistent, if they are from the same device, they cannot be data of the same target; if two targets with different target numbers can each retrieve a fusion batch number from the third data set tagFuseNoMap or the cache, they cannot be data of the same target; if a non-fusion flag is obtained through the retrieval of the cache data of the current two target numbers, it indicates that the current two targets have been calculated in the historical time window and are determined not to be the same target, and there is no need to perform the identity determination calculation again.

[0139] After the above screening process, most of the calculations will be reduced. Because in a general multi-target environment, the proportion of target data that needs to be fused is not very high, so there will be a large part of the identity calculation results between target data that do not have identity. Based on this premise, the above screening process has an obvious effect on improving the calculation performance.

[0140] The target identity determination algorithm mainly includes track model algorithm selection, track model reference target selection, track prediction, track evaluation, and fusion batch number assignment, as follows:

[0141] S341. Denote one of the targets as the model reference target and the other target as the alignment target, and select a model algorithm for the model reference target.

[0142] In one embodiment, in step S341, denote one of the targets as the model reference target and the other target as the alignment target, and select a model algorithm for the model reference target, as follows:

[0143] When the data volume of both targets in the new target list is equal to 2, the model algorithm selects a linear function model and uses the target with a larger time span as the model reference target;

[0144] When the data volumes of the two targets in the new target list are equal to 2 and 3 respectively, the model algorithm selects a parabola function model and uses the target with more data as the model reference target;

[0145] When the data volume of both targets in the new target list is equal to 3, the model algorithm selects a parabola function model and selects any target as the model reference target;

[0146] When the data volume of both targets in the new target list is greater than 3, obtain the data volume of the two targets that fall within each other's time range. If the data volume that falls within each other's time range is greater than the first preset threshold and the data volume of the other party is greater than the second preset threshold, the model algorithm selects a cubic spline interpolation model and uses the other party as the model reference target; otherwise, the model algorithm selects a least squares method model and uses the target with more data as the model reference target.

[0147] Among them, the selection of the model algorithm is mainly established to cope with the number of traces and the cross-distribution of trace times in the target track. For the data volume of the new target list corresponding to the two targets that is less than 2, it does not participate in the identity calculation and is directly reported or filtered.

[0148] When the data volume of both targets in the new target list is greater than 3, calculate the cross of the two target trace times. For example, represent the target track data sets corresponding to the two targets as follows:

[0149] target-a: [tag t1 , tag t2 , tag t3 ,..., tag tn ,

[0150] target-b: [tag k1 , tag k2 , tag k3 ,..., tag km

[0151] ​Among them, t1, t2…tm represent the timestamps of all points in target-a; k1, k2, …km represent the timestamps of all points in target-b, and they are in chronological order. Therefore, t1 is the earliest time point in the data of target-a, denoted as aMinTime, and tn is the latest time point in the data of target-a, denoted as aMaxTime; correspondingly, the earliest time point k1 in the data of target-b is denoted as bMinTime, and the latest time point km in the data of target-b is denoted as bMaxTime. Define aTimeInnerBSum to represent the number of points in target-a within the range of [bMinTime, bMaxTime], and define bTimeInnerASum to represent the number of points in target-b within the range of [aMinTime, aMaxTime]. If aTimeInnerBSum > 5 and the number of target points of target-b is greater than 10, then target-b is used as the model reference target, and the trajectory model selects the cubic spline interpolation model; if bTimeInnerASum > 5 and target-a > 10, then target-a is used as the model reference target, and the trajectory model selects the cubic spline interpolation model. Otherwise, the least squares method model is selected for the trajectory model, and the target with more target points is used as the model reference target. Denote the model reference target as baseTarget, and the remaining other target as the alignment target, denoted as alignTarget.

[0152] S342. Establish a prediction model for the model reference target based on the selected model algorithm, and train with the target trajectory data in the new target list where the model reference target is located to obtain the final prediction model;

[0153] S343. Predict the target position aligned with the time point of the target trajectory data of the alignment target on the final prediction model based on interpolation or extrapolation;

[0154] S344. Determine whether the average distance error between the predicted target position and the target position of the alignment target is less than the preset distance threshold. If so, the target meets the identity, and it is determined to be the same target. Check whether the retrieved target has been assigned a fusion batch number. If any target has a fusion batch number, apply the fusion batch number transfer logic to share the fusion batch number. If neither of the two targets has been assigned a fusion batch number, use the purpose number of the earlier-occurring target as the shared fusion batch number, and save the assigned fusion batch number to the third data set and the cache; otherwise, the target does not meet the identity, and add the non-fusion flag of the two targets to the cache.

[0155] Among them, the model is trained according to the selected model algorithm and the point data of the model benchmark target to obtain the final prediction model, and the target position aligned with the target track data time point of the alignment target is predicted on the final prediction model according to the interpolation method or extrapolation method to obtain the predicted point data. The final track data is aligned in time to the time point reported by the device with less target data, and then the multi-device reported data aligned in time is fused and calculated. If the average distance error between the predicted target position and the target position of the alignment target is less than the preset distance threshold minimum_distance_threshold, the two target tracks are determined to be the same target, that is, it means that these two targets are actually observations of the same target by different devices, and the fusion batch number needs to be assigned to them. If the average distance error is greater than minimum_distance_threshold, it indicates that the current two targets do not belong to the same track, and a non-fusion flag is added to the cache. The subsequent identity algorithm will use this data when combining targets and will not perform repeated calculations. The calculation method of the average distance error is the same as that of the track average error ε and will not be elaborated here.

[0156] Fusion batch number assignment: First, check whether each of the two destination numbers already has an assigned fusion batch number. If any one has a fusion batch number, the fusion batch number transfer logic is applied. The principle here is that if a and b are on the same target track, and b and c are also on the same target track after calculation, then a - b - c are all on the same target track. Therefore, there is no need to perform identity calculation between a and c, and the fusion batch number can be directly obtained through the fusion batch number transfer principle. If neither of them has obtained a fusion batch number, then select the destination number of the target that appears earlier among the two targets as their subsequent fusion batch number. Then save the assigned fusion batch number to the third data set tagFuseNoMap and the cache.

[0157] S35. Group the target track data of the same target under the current time window, and perform error correction on the target track data corresponding to each target after this grouping based on Kalman filter smoothing processing. The accuracy can be further improved.

[0158] S36. Evaluate the threat level of the corresponding target according to the target track data after error correction, and output the threat level score to the second message queue for subscription.

[0159] In one embodiment, in step S36, the threat level score final-threat-score is calculated as follows:

[0160]

[0161]

[0162] Among them, F y (FAC y ) is the threat assessment score of the threat factor FAC y , α y is the target threat weight coefficient, and N is the total number of threat factors.

[0163] In one embodiment, the threat factor FAC y includes the position of the target from the center point of the defense area, the target's real-time speed, the target type, the working state of the target, the target's heading angle, and the security level of the center point of the defense area.

[0164] Among them, the threat assessment of the target's current position is carried out using the target track data after error correction, and the center point of the defense area is defined as centerPoint. The quantity and type of threat factors can be adjusted according to actual requirements. As Figure 6 shown, in this embodiment, the threat factor FAC y includes the position of the target from the center point of the defense area FAC 1 (different distance ranges from the center point of the defense area have different levels of threat), the target's real-time speed FAC 2 (the greater the radial speed of the target relative to the direction of the center point of the defense area, the higher the threat), the target type FAC 3 (different types of targets have different inherent threats. For example, an unmanned aerial vehicle target has a greater threat than a bird target), the working state of the target FAC 4 (the advancing state of the unmanned aerial vehicle has a higher threat than the hovering state), the target's heading angle FAC 5 (corresponding to the angle between the target movement direction and the center point direction, the smaller the angle, the greater the threat), and the security level of the center point of the defense area FAC 6 (the security level of the current defense center point, the higher the importance, the greater the threat). The threat weights of the threat factors correspondingly include the position threat value TH1 of the target from the center point of the defense area, the real-time speed threat value TH2 of the target, the target type threat value TH3, the working state threat value TH4 of the target, the heading angle threat value TH5 of the target, and the security level threat value TH6 of the center point of the defense area. The evaluation can be carried out by providing a visual target threat factor evaluation configuration tool and a target threat weight configuration tool. Among them, the target threat factor evaluation configuration tool is used to implement threat factor evaluation configuration and factor weight (the threat weight of the threat factor) configuration. For example, input the corresponding factor configuration file on the interface of the target threat factor evaluation configuration tool. For example, for the position of the target from the center point of the defense area FAC 1 , the factor configuration file defines specific scoring rules such as the threat score rule for the high-risk range of 0 - 500m and the threat score rule for the medium-risk range of 500 - 1500m. At the same time, the factor configuration file will also attach the factor weight of this threat factor, and generate configuration information F based on this information.y (FAC y ) is saved in the cache for high-speed retrieval. As shown in Figure 7 and 8 , it is only for illustration and does not represent the complete configuration of this embodiment. The position of the center point of the target distance defense zone FAC is configured in the interface 1 (distance F1), and the heading angle of the target FAC 5 (heading angle F2). After the target threat factor evaluation and calculation, and then according to the threat weight allocation in the target threat weight configuration tool, the final comprehensive threat score final-threat-score is obtained. The threat level score is output to the second message queue for downstream systems to subscribe.

[0165] This method has better comprehensibility and system maintainability compared with the method that strongly relies on device identification information by uniformly batch-numbering the target batch numbers reported by detection devices and allocating fusion batch numbers; pre-grouping the reported target track data based on a dynamic time window can obtain more target track data of the same target within a processing time window, which helps improve the accuracy of the fusion calculation results. At the same time, it uses the method of associating based on historical target track data, which can well solve the problem of fluctuations in the accuracy of real-time results caused by insufficient target data volume within the current time window, and can avoid accidental errors caused by calculating each target track one by one and reduce the performance overhead of the information processing system, ensuring the real-time and high availability of data; saving and retrieving historical target track data of target tracks through caching, and dynamically allocating historical target track data to participate in track fusion calculation. On the one hand, it supplements the data volume in the current time window. On the other hand, for the case of a large amount of historical data, the historical data is eliminated and discarded based on quantity and time to avoid the overfitting effect of long-term historical track data on the results, support the identification and association of lost target tracks, and can well make up for the target statistics and analysis capabilities in the case of hardware deployment and abnormal device operation at the software level; excluding abnormal data points through a dynamically expandable abnormal data point verification process to reduce interference, which helps improve the operation speed and accuracy; compared with the existing fusion methods based on device linkage and the single-type device data fusion method, this method has stronger versatility, is not restricted by the device type and the number of devices. As long as the device can stably report target track data, it can be uniformly processed, and there are no special requirements for the frequency of device-reported data; compared with the identification of target track identity in traditional fusion methods, this method proposes the concept of target fusion batch number transfer through data cache matching, greatly reducing the amount of identity calculation and improving the system data processing ability; compared with the existing fusion methods based on target track spatio-temporal alignment, this method can also realize the identification of the same target track based on non-spatio-temporally aligned data. At this stage, all original target data is not selected or discarded, and the respective real reported data is used during the same fusion, which can make more full use of the original target track data of each detection device and improve the authenticity of the track; a perfect fusion data retrieval system, during the process of identifying lost target tracks and fusion matching, there is no need to frequently repeat the calculation of data matching and selection, reducing unnecessary complex core calculation logic; the dynamic allocation logic of the track model during the target identity determination process, and at the same time, there is a dynamic model reference target selection strategy. Compared with a single matching logic, it can better adapt to the quantity uncertainty and uneven time distribution of target track data, improve the accuracy, and make the same determination result more reliable;Based on the Kalman filter smoothing processing method, trajectory smoothing processing and position error correction are performed on the full-scale target track after identity determination. The targets with the same identity are fused and matched to achieve the calculation of multi-track association recognition of the same target, which can accurately identify the true number of targets in the defense area and restore the true movement trajectory of the targets, and is beneficial to improving the evaluation accuracy when realizing the all-round dynamic and scalable target threat level evaluation.

[0166] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered that the scope described in this specification is covered.

[0167] The above-described embodiments only express the embodiments of the present application that are relatively specific and detailed, but should not be construed as limiting the scope of the patent application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A real-time and efficient method for fusing heterogeneous target track data, characterized in that: The real-time and efficient method for fusing heterogeneous target track data includes the following steps: S1. Obtain the target track data of all detection devices and put them into the first message queue. The target track data includes the detection device number, the original target batch number, the target position, and the timestamp; S2. Subscribe to the target track data in the first message queue, and calculate the MD5 value based on the detection device number and the original target batch number as the destination number corresponding to the target track data; S3. Perform batch processing on the target track data based on a preset time window. The processing process of each batch of target track data is as follows: S31. Group according to the destination number, and each group of target track data forms a target list; S32. Match the historical target track data of the targets corresponding to each target list, and perform the following operations: S321. Obtain all cached historical destination numbers and put them into the first data set; S322. Judge whether there is a destination number corresponding to the target in the current time window in the cache. If so, it is considered that there is a historical track. Remove the destination number corresponding to the target from the first data set, cache the target track data corresponding to the target in the current time window, and insert the historical target track data corresponding to the target into the target list corresponding to the target in the current time window to form a new target list. Otherwise, it is considered that there is no historical track, and the destination number and target track data corresponding to the target in the current time window are respectively put into the second data set and the cache; S323. Retrieve the fusion batch number of the targets existing in the current time window in the cache. If the target has obtained a fusion batch number, use this fusion batch number as the latest fusion batch number corresponding to the target and put it into the third data set; S33. Identify the targets with lost tracks corresponding to the remaining destination numbers in the first data set, and perform the following operations: S331. Traverse the first data set to judge whether there is a track model for each remaining target. If so, load the corresponding track model. Otherwise, create and train a track model based on the historical target track data of the corresponding target, and store the trained track model in the model set and the cache; S332. Traverse the second data set to obtain the target track data of each target in the new target list; S333. Use the loaded track model or the trained track model to predict the target position aligned with the time point of the target track data corresponding to the target, and calculate the point average error ε; S334. Judge whether the point average error ε is less than the target threshold. If so, consider the corresponding target in the current time window as a historical target with lost track, and perform batch number fusion processing on the corresponding target in the current time window and the historical target. Otherwise, consider the corresponding target in the current time window as a newly emerged target; S34. Determine the identity of any two targets and perform fusion matching, specifically as follows: S341. Denote one of the targets as the model reference target and the other target as the alignment target, and select a model algorithm for the model reference target; S342. Establish a prediction model for the model benchmark target based on the selected model algorithm, and train it with the target track data in the new target list where the model benchmark target is located to obtain the final prediction model; S343. Predict the target position aligned with the time point of the target track data of the alignment target on the final prediction model based on interpolation or extrapolation; S344. Determine whether the average distance error between the predicted target position and the target position of the alignment target is less than the preset distance threshold. If so, the target meets the identity, and it is determined as the same target. Check whether the retrieved target has been assigned a fusion batch number. If any target has a fusion batch number, apply the fusion batch number transfer logic to share this fusion batch number. If neither of the two targets has been assigned a fusion batch number, use the purpose number of the earlier-occurring target as the shared fusion batch number, and save the assigned fusion batch number to the third data set and the cache; Otherwise, the target does not meet the identity, and add non-fusion identifiers for the two targets to the cache; S35. Group the target track data of the same target under the current time window, and perform error correction on the target track data corresponding to each target after this grouping based on Kalman filter smoothing processing; S36. Evaluate the threat level of the corresponding target according to the target track data after error correction, and output the threat level score to the second message queue for subscription.

2. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S2, the subscription of the target track data in the first message queue is implemented based on the Flink computing framework.

3. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S31, an abnormal point picking and verification process is also performed on each target list, and the abnormal target track data is deleted, specifically as follows: S311. Determine whether the target position of the target track data exceeds the preset range. If so, delete the target track data. Otherwise, execute step S312; S312. Obtain the current time of the system after time synchronization and read the time stamp of the target track data. Determine whether the difference between the time of the target track data and the current time of the system exceeds the preset time delay threshold or whether the time of the target track data exceeds the average time of the target track data in the target list. If so, delete the target track data. Otherwise, execute step S313; S313. Calculate the average speed of the target track data in the target list, and calculate the instantaneous speed of each target track data according to the position difference and time difference between adjacent two points. Determine whether the instantaneous speed is higher than the average speed. If so, delete the target track data. Otherwise, retain it.

4. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S322, the following operations are also performed on the new target list: When the data volume in the new target list is greater than the preset retention value, delete the earlier historical target track data until the data volume reaches half of the preset retention value.

5. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S331, creating and training a track model based on the historical target track data corresponding to the target is specifically as follows: S3311. Respectively establish models for the longitude, latitude, and altitude dimensions of the target based on the least squares method, and the formulas are as follows: where t is time, f(t) is the objective function, and a x is the constant term coefficient of the x-th term of the polynomial, and M is the maximum order; S3312. Determine whether pointNum > max-degree + 1 is satisfied, where pointNum is the data volume of the target in the new target list, and max-degree is the preset maximum order. If so, the maximum order M = max-degree; otherwise, the maximum order M = pointNum - 1; S3313. Train the models for the three dimensions of the target based on the corresponding historical target track data using the fit method, and regard the trained three models as the track model.

6. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S334, the batch number fusion processing is specifically as follows: Retrieve and judge from the third data set and the cache whether the corresponding historical target has been assigned a fusion batch number. If so, store the mapping information between the original target batch number of the corresponding target and the fusion batch number assigned to the historical target in the third data set and the cache; otherwise, reassign the fusion batch number.

7. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S34, when determining the identity of any two targets and performing fusion matching, the following operations are also performed: SA1. Judge whether the two targets are from the same detection device. If so, it is considered that the targets do not meet the identity; otherwise, execute step SA2; SA2. Judge whether the two targets have been respectively assigned fusion batch numbers. If so, it is considered that the targets do not meet the identity; otherwise, execute step SA3; SA3. Judge whether there is a non-fusion flag for the two targets. If so, it is considered that the targets do not meet the identity; otherwise, execute step S341.

8. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S341, regarding one of the targets as the model reference target and the other target as the alignment target, and selecting a model algorithm for the model reference target is specifically as follows: When the data volumes in the new target lists of both targets are equal to 2, the model algorithm selects a linear function model, and the target with a larger time span is used as the model reference target; When the data volumes in the new target lists of the two targets are equal to 2 and 3 respectively, the model algorithm selects a parabola function model, and the target with a larger data volume is used as the model reference target; When the data volumes in the new target lists of both targets are equal to 3, the model algorithm selects a parabola function model, and any target is selected as the model reference target; When the data volume of both targets in the new target list is greater than 3, obtain the data volume of the two targets that fall within each other's time range. If the data volume that falls within each other's time range is greater than the first preset threshold and the data volume of the other party is greater than the second preset threshold, the model algorithm selects the cubic spline interpolation model with the other party as the model reference target. Otherwise, the model algorithm selects the least squares method model with the target with the larger data volume as the model reference target.

9. The real-time and efficient fusion method for heterogeneous target track data according to claim 1, characterized in that: In step S36, the threat level score final-threat-score is calculated as follows: Among them, F y (FAC y ) is the threat assessment score of the threat factor FAC y , α y is the target threat weight coefficient, and N is the total number of threat factors.

10. The real-time and efficient fusion method for heterogeneous target track data according to claim 9, characterized in that: The threat factor FAC y includes the position of the target from the center point of the defense area, the real-time speed of the target, the target type, the working state of the target, the heading angle of the target, and the security level of the center point of the defense area.

Images