Asynchronous track association method and device based on triple network, storage medium and electronic equipment

Abstract

The embodiment of the invention provides an asynchronous track association method and device based on a triple network, a storage medium and electronic equipment, and relates to the technical field of multi-source sensor track association fusion, and the method comprises the steps: carrying out the preprocessing of track data of two information sources; obtaining a depth track feature model based on the preprocessed track data and triple network training; performing calculation based on the preprocessed track data and the depth track feature model to obtain to-be-associated track features; performing correlation measurement calculation on the two tracks based on the track features to obtain a track feature correlation vector; obtaining a binary classifier based on the track feature correlation vector and binary classification network training; and calculating and obtaining a track correlation result based on the track feature correlation vector and the binary classifier. According to the technical scheme, the association problem in a dense scene and the error expansion problem caused by an asynchronous track are solved, and the association performance is improved.

Description

Technical Field

[0001] This application relates to the technical field of multi-source sensor track association and fusion. Specifically, it relates to an asynchronous track association method, device, storage medium, and electronic device based on a triple network. Background Technique

[0002] With the rapid development of modern information technology, it has promoted the wide use of multi-source sensors in modern warfare, making the data fusion processing of multi-source sensors increasingly complex. Multi-source sensor track association is the basis and prerequisite of multi-source fusion, which directly determines the accuracy and reliability of subsequent position fusion and target recognition, and is one of the important links in battlefield situation estimation.

[0003] Distributed track association is to obtain the local tracks of each information source and then send the data to the processing center for processing, that is, track-track association. Track-track association can obtain higher reliability by using a lower channel capacity. The purpose is to determine whether the measurements from different sensors belong to the same target, and then obtain the track sets of the same target on different sensors, providing a basis for subsequent track fusion. Therefore, a good track association algorithm is an important prerequisite for improving battlefield situation awareness. How to effectively mine the hidden features in the tracks is the key to track-track association research.

[0004] Traditional track association methods are divided into statistical-based association methods and fuzzy mathematics-based association methods. Statistical-based association methods include the nearest neighbor method, weighted method, sequential method, multi-dimensional assignment method, etc. Fuzzy mathematics-based methods include the fuzzy comprehensive function method, multi-factor fuzzy decision method, etc. The above traditional track association methods are highly sensitive to the distribution of data, and in the case of large measurement errors and dense targets, the correct association rate drops sharply. In addition, traditional methods need time registration before asynchronous track association, which makes the error expand step by step, making the association face severe challenges.

[0005] In recent years, with the continuous development of machine learning, some scholars have converted the track association problem into a binary classification problem of pattern recognition, and trained the model by constructing a track training set, such as using a convolutional neural network, a time series network, etc. to extract track features. However, the existing deep learning-based track association algorithms still face severe tests, specifically reflected in: First, the existing deep learning-based asynchronous track association methods need to first perform spatio-temporal registration on two tracks to be associated, then extract the statistical distance between the two tracks, then use the statistical distance as the network input to extract track features, and then perform association discrimination. However, performing spatio-temporal registration is likely to amplify errors and lose the weak differential features between the two tracks.

[0006] Second, multi-source tracks in complex and dense scenarios have great similarity, and the feature differences between positive and negative track samples are small. It is difficult to effectively perform correlation discrimination by only using a deep network to learn the weak feature differences between positive and negative samples, which increases the difficulty of deep track feature extraction and correlation. Summary of the Invention

[0007] Embodiments of the present application provide an asynchronous track association method, device, storage medium, and electronic device based on a triplet network to overcome the deficiencies of the above-mentioned traditional association methods and deep learning-based association methods.

[0008] Other features and advantages of the present application will become apparent through the following detailed description, or will be partially learned through the practice of the present application.

[0009] According to the first aspect of the embodiments of the present application, an asynchronous track association method based on a triplet network is provided, including: Preprocess the track data of two information sources; Based on the preprocessed track data and a triplet network, train a deep track feature model; Based on the preprocessed track data and the deep track feature model, calculate the track features to be associated; Based on the track features, perform a correlation metric calculation on two tracks to obtain a track feature correlation vector; Based on the track feature correlation vector and a binary classification network, train a binary classifier; Based on the track feature correlation vector and the binary classifier, calculate the track association result.

[0010] In some embodiments of the present application, based on the foregoing solution, the preprocessing of the track data of two information sources includes: Normalize the track data of two information sources and supplement the track data of two information sources.

[0011] In some embodiments of the present application, based on the foregoing solution, the training of the deep track feature model based on the preprocessed track data and a triplet network includes: Based on the preprocessed track data, construct a training sample of the triplet network; Input the training sample into the long short-term memory network with weight sharing in the triplet network for processing to obtain an anchor track feature corresponding to the anchor track, a positive sample track feature corresponding to the positive sample track, and a negative sample track feature corresponding to the negative sample track; Calculate the triplet loss function corresponding to the anchor track feature, the positive sample track feature, and the negative sample track feature; Backpropagate to calculate gradients through the Adam optimizer and update the parameters of the triplet network; Perform iterative training until the triplet network converges or reaches a preset number of iterations to obtain the depth track feature model.

[0012] In some embodiments of the present application, based on the foregoing solution, the track features to be associated calculated based on the preprocessed track data and the depth track feature model include: Randomly select one track data from the preprocessed track data of two information sources and input it into the depth track feature extraction model to extract the track features to be associated.

[0013] In some embodiments of the present application, based on the foregoing solution, the calculation of the correlation metric for two tracks based on the track features to obtain a track feature correlation vector includes: Use the L1 norm distance algorithm to calculate the L1 norm of the track features to be associated to obtain the track feature distance, which is used as the track feature correlation vector.

[0014] In some embodiments of the present application, based on the foregoing solution, the training of a binary classifier based on the track feature correlation vector and a binary classification network includes: Use the track feature distance as the training sample of the binary classification network; Input the training sample into the binary classification network for processing to obtain an association result; Calculate the cross-entropy loss based on the association result; Backpropagate to calculate gradients through the Adam algorithm and update the network parameters of the binary classification network; Perform iterative training until the binary classification network converges or reaches a preset number of iterations to obtain the binary classifier.

[0015] In some embodiments of the present application, based on the foregoing solution, the calculation of the track association result based on the track feature correlation vector and the binary classifier includes: Input the track feature correlation vectors of two information sources into the binary classifier to calculate the track association result.

[0016] According to the second aspect of the embodiments of the present application, there is provided an asynchronous track association device based on a triplet network, including: A preprocessing unit for preprocessing the track data of two information sources; A first training unit for training a depth track feature model based on the preprocessed track data and a triplet network; A first calculation unit for calculating the track features to be associated based on the preprocessed track data and the depth track feature model; A second calculation unit, configured to calculate the correlation metric between two tracks based on the track features to obtain a track feature correlation vector; A second training unit, configured to train a binary classifier based on the track feature correlation vector and a binary classification network; A third calculation unit, configured to calculate a track association result based on the track feature correlation vector and the binary classifier.

[0017] According to a third aspect of the embodiments of the present application, there is provided a computer-readable storage medium storing computer instructions, which when running on a computer, cause the computer to execute the method as described in the first aspect.

[0018] According to a fourth aspect of the embodiments of the present application, there is provided an electronic device, including: a memory and a processor; The memory is configured to store computer instructions; The processor is configured to call the computer instructions stored in the memory, so that the electronic device executes the method as described in the first aspect.

[0019] The beneficial effects of the present application are as follows: First, the association problem is transformed into a metric learning problem. The feature extraction method based on the triplet network eliminates the need for time registration in track association, eliminating the error accumulation problem caused by time asynchrony; Second, the triplet loss function enables the feature extraction network to effectively widen the distance between positive and negative track samples, paying more attention to the learning process of similar track features in the deep track feature representation, increasing the feature difference of similar tracks. Especially in dense scenarios, the method provided by the present application has good association performance.

[0020] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the drawings: Figure 1 A schematic flowchart of a method for asynchronous track association based on a triplet network according to an embodiment of the present application is shown; Figure 2Shows a schematic diagram of the principle of an asynchronous track association method based on a triple network according to an embodiment of the present application; Figure 3 Shows a training schematic diagram of a track feature extraction model according to an embodiment of the present application; Figure 4 Shows a schematic diagram of the model structure of a binary classifier according to an embodiment of the present application; Figure 5 Shows a comparison schematic diagram of the correct association rates of the present method and a comparative method under different numbers of targets in a scenario according to an embodiment of the present application; Figure 6 Shows a block diagram of an asynchronous track association device based on a triple network according to an embodiment of the present application; Figure 7 Shows a block diagram of an electronic device according to an embodiment of the present application; Figure 8 Shows a schematic diagram of the structure of a computer system of an electronic device suitable for implementing the embodiments of the present application. Detailed implementation manners

[0022] Now, example embodiments will be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more complete and comprehensive, and will fully convey the concept of the example embodiments to those skilled in the art.

[0023] In addition, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present application. However, those skilled in the art will realize that the technical solutions of the present application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the present application.

[0024] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0025] The flowcharts shown in the accompanying drawings are merely illustrative and not necessarily include all content and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be decomposed, while some operations / steps can be combined or partially combined. Therefore, the actual execution order may change according to the actual situation.

[0026] It should be noted that: "a plurality of" mentioned in this text refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of this application and the above accompanying drawings are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the objects used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order other than those illustrated or described.

[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0029] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0030] See Figure 1 , which shows a schematic flowchart of an asynchronous track association method based on a triple network according to an embodiment of the present application.

[0031] As Figure 1 shown, a method for asynchronous track association based on a triple network is presented, specifically including steps S100 to S600.

[0032] Refer to Figure 1 , step S100, preprocess the track data of two information sources.

[0033] It can be understood that since the sampling periods, coordinate systems, dimensions, etc. of different information sources are different, it is necessary to first preprocess the track data of the two information sources for subsequent use.

[0034] In some feasible embodiments, based on the foregoing solution, the preprocessing of the track data of the two data sources includes: Normalize the track data of the two data sources and supplement the track data of the two data sources.

[0035] Exemplarily, referring to Figure 2 , a schematic diagram of the principle of an asynchronous track association method based on a triple network according to an embodiment of the present application is shown.

[0036] The specific process of preprocessing is as follows: In this example, it is assumed that both data sources are sampled in a rectangular coordinate system, and the measurement of a single moment of the data source is expressed as , where x represents the abscissa, y represents the measured ordinate, t represents the measurement time. It is assumed that data source 1 detects N 1 track, and the track set of data source 1 is expressed as Data source 2 detects N 2 tracks, and the track set of data source 2 is expressed as , where the k th i track of the data source is , where represents the total number of measurement points of this track.

[0037] The method of the present application transforms the association problem into a binary classification problem of pattern recognition. The track measurements contain characteristic data with different dimensions and units. Therefore, before inputting the track data into the network, it is necessary to normalize the track data of the data source: ; where, where represents the data to be normalized, represents the minimum value of the measurement data, represents the maximum value of the measurement data.

[0038] Since the sampling periods of the two data sources are different, the track lengths are different. It is necessary to supplement the track data. Assume that the data dimensions of the two tracks and are and respectively. The data dimensions of the unified and supplemented tracks and are , where ; represents the number of track points of the track , represents the number of track points of the track . Indicates the number of supplemented track points in the track.

[0039] Continue to refer to Figure 1 , step S200, a deep track feature model is trained based on the preprocessed track data and the triplet network.

[0040] In some feasible embodiments, based on the foregoing solution, the training of the deep track feature model based on the preprocessed track data and the triplet network includes: Construct training samples for the triplet network based on the preprocessed track data; Input the training samples into the long short-term memory network with weight sharing in the triplet network for processing to obtain the anchor track feature corresponding to the anchor track, the positive sample track feature corresponding to the positive sample track, and the negative sample track feature corresponding to the negative sample track; Calculate the triplet loss function corresponding to the anchor track feature, the positive sample track feature, and the negative sample track feature; Backpropagate to calculate the gradient through the Adam optimizer and update the triplet network parameters; Perform iterative training until the triplet network converges or reaches a preset number of iterations to obtain the deep track feature model.

[0041] Exemplarily, based on the foregoing example, as Figure 3 shown, the training steps of the deep track feature model are specifically as follows: Step S201, construct triplet network training samples: The network input is the triplet track sample set { , , }, where represents the preprocessed anchor track (anchor), represents the preprocessed positive sample track (positive), represents the preprocessed negative sample track (negative). Randomly select a track from source 1 as the anchor track, then select a track generated by the same target as source 1 from source 2 as the positive sample track, and then randomly select a track generated by a different target from the positive sample from source 2 as the negative sample track.

[0042] Step S202, Forward Propagation: Based on the training samples of the triple network, use the Long Short-Term Memory network (LSTM) as the feature extraction module to extract its deep features. LSTM is a special recurrent neural network designed specifically to address the vanishing gradient and exploding gradient problems in traditional RNNs when dealing with long sequence data. The key of LSTM lies in the introduction of memory cells and gating mechanisms, which can selectively remember or forget information, so it can effectively capture long-term dependencies. Each time step of LSTM contains the following calculation steps: a) Forget gate Calculate the output of the forget gate , , where is the Sigmoid function, is the weight matrix, is the hidden state at the previous time step, is the current input, is the bias term.

[0043] b) Input gate Calculate the output of the input gate and the candidate memory cell value : ; ; c) Update memory cell Update the memory cell , ; d) Output gate Calculate the output of the output gate and the current hidden state : ; ; After the preprocessed track data passes through the LSTM-based feature extraction module, track features , and are obtained, where represents the anchor feature corresponding to the anchor track , represents the positive sample track feature corresponding to the positive sample track , represents the negative sample track feature corresponding to the negative sample track .

[0044] Step S203, Calculate Feature Distance: After extracting the track features using the weight-sharing feature extraction module, the corresponding feature expressions of each triple track are obtained , and , based on this, calculate the distance between the anchor feature and the positive sample feature , the distance between the anchor feature and the negative sample feature , where , .

[0045] Step S204, calculate the triplet loss: The triplet loss requires that the interval between the positive and negative sample pairs satisfies a set threshold , where is a positive constant, which is the minimum distance interval between the positive and negative sample pairs. The purpose of the triplet network is to make and the distances after feature expression as small as possible, and the distances after feature expression as large as possible, and and the minimum interval between them is the distance , and the triplet loss function is expressed as: ; where , it can be seen from the triplet loss function that the process of minimizing the loss function is the feature learning process of increasing the positive and negative sample intervals. In feature expression learning, more attention is paid to the feature learning of similar tracks, increasing the difference of similar track features, which can play an important role in dense scenarios.

[0046] Step S205, backpropagation and parameter update: Calculate the gradient through the Adam optimizer for backpropagation and update the network parameters.

[0047] Step S206, repeat Steps S201 to S205 for iterative training until the triplet network model converges or reaches the preset number of iterations. Continue to refer to Figure 1 , Step S300, calculate the track features to be associated based on the preprocessed track data and the deep track feature model.

[0048] In some feasible embodiments, based on the foregoing solution, calculating the track features to be associated based on the preprocessed track data and the deep track feature model includes: Randomly select one track data from the preprocessed track data of two data sources and input it into the deep track feature extraction model to extract the track features to be associated.

[0049] Exemplarily, based on the foregoing example, a track data after preprocessing of the source 1 is randomly selected and denoted as , and a track data after preprocessing of the source 2 is denoted as . Then, and are respectively input into the trained feature extraction network, and the output is the track feature, denoted as and .

[0050] Continue to refer to Figure 1 , step S400. Based on the track features, the correlation metric calculation is performed on the two tracks to obtain a track feature correlation vector.

[0051] In some feasible embodiments, based on the foregoing solution, the correlation metric calculation is performed on the two tracks based on the track features to obtain a track feature correlation vector, including: Using the L1 norm distance algorithm to calculate the L1 norm of the track features to be associated, and obtaining a track feature distance as the track feature correlation vector.

[0052] It can be understood that the correlation metric is used to measure the correlation degree between two track features. As a correlation metric method, the L1 norm has significant advantages in terms of sparsity, robustness, feature selection, and preventing overfitting, and is especially suitable for high-dimensional data and scenarios with noise. Therefore, in this embodiment, the L1 norm is selected as the correlation metric method for features.

[0053] Exemplarily, based on the foregoing example, the process of the correlation metric calculation is as follows: Assume that the feature representation of track 1 after the feature extraction module is , and the feature representation of track 2 after the feature extraction module is . Then, the similarity metric based on the L1 norm is , where i represents the i th feature.

[0054] Continue to refer to Figure 1 , step S500. Based on the track feature correlation vector and the binary classification network, a binary classifier is trained.

[0055] In some feasible embodiments, based on the foregoing solution, the training of the binary classifier based on the track feature correlation vector and the binary classification network includes: Taking the track feature distance as the training sample of the binary classification network; Inputting the training sample into the binary classification network for processing to obtain an association result; Calculating the cross-entropy loss based on the association result; Calculate the gradient through backpropagation using the Adam algorithm and update the network parameters of the binary classification network; Perform iterative training until the binary classification network converges or reaches a preset number of iterations to obtain the binary classifier.

[0056] The model structure of the binary classifier is as Figure 4 shown. The binary classifier consists of two fully connected layers and three activation functions. The activation functions of connection layer 1 and fully connected layer 2 are ReLU functions. The output of fully connected layer 2 is the input of the activation function Sigmoid, and the output is the associated label, where 1 represents association and 0 represents non - association.

[0057] Exemplarily, based on the foregoing example, the training steps of the binary classifier are as follows: Step S501, construct binary classifier training samples: Combine track features pairwise. 50% are two - source data generated by the same target, and 50% are two - source data generated by different targets. Use their distances as the training samples for binary classification and input them into the binary classifier module.

[0058] Step S502, forward propagation: Construct training samples of track feature distances and input them into the binary classification network. The binary classification network includes two fully connected layers, and the input layer, hidden layer, and output layer are 128, 512, and 2 respectively. The activation function is the Sigmoid function. After passing through the binary classification network, obtain the association result.

[0059] Step S503, calculate cross - entropy loss: , where y i represents the label of the i th track sample pair, where association is 1 and otherwise 0, p i represents the association probability of the i th track sample pair.

[0060] Step S504, backpropagation and parameter update: Calculate the gradient through backpropagation using the Adam algorithm and update the network parameters of the binary classifier.

[0061] Step S505, repeat steps S501 to S504 for iterative training until the binary classification network model converges or reaches a preset number of iterations.

[0062] Continue to refer to Figure 1 , step S600, calculate the track association result based on the track feature correlation vector and the binary classifier.

[0063] In some feasible embodiments, based on the foregoing solution, calculating the track association result based on the track feature correlation vector and the binary classifier includes: Input the track feature correlation vectors of two information sources into the binary classifier, and calculate the track association result.

[0064] Exemplarily, take the track feature correlation vectors of two information sources as the input of the binary classifier, and output the association probability of two tracks.

[0065] Compare the correct association rates of this method and four comparison methods (the four comparison methods are the nearest neighbor method, the weighted method, the long short-term memory network method, and the siamese neural network method). The correct association rates of this method and the comparison methods under different numbers of targets in the scenario are compared as Figure 5 shown. Compared with other methods, the asynchronous track association method based on the triplet network proposed in this application has the highest association accuracy rate in both dense and sparse environments, which shows the superiority of this method.

[0066] In summary, based on the extraction of track time series features by the long short-term memory network, this method uses deep metric learning based on triplet loss to map tracks to a metric space, effectively increasing the feature distance between positive and negative samples, so as to effectively distinguish the weak track features of similar targets and achieve effective association of asynchronous tracks in a dense environment.

[0067] The core of this method lies in using a two-stage network for association judgment, which can make the judgment network have good interpretability. The first stage is feature extraction. The target feature embedding space learned by the triplet network makes the same-class samples close to each other in the space, and different-class samples far from each other in the space, increasing the positive and negative sample distances in the feature space and enabling the network to learn more discriminative features. The second stage is association judgment. The binary classification network discriminates whether the tracks belong to the same target. The benefits of this method are mainly reflected in: First, it transforms the association problem into a metric learning problem. The feature extraction method based on the triplet network eliminates the need for time registration in track association and solves the problem of error accumulation caused by time asynchrony. Second, using the triplet loss function enables the feature extraction network to effectively increase the distance between positive and negative track samples, pays more attention to the learning process of similar track features in the deep track feature expression, and increases the feature difference of similar tracks. Especially in a dense scenario, this method has good association performance.

[0068] Compared with traditional association algorithms and deep learning algorithms, this method overcomes the defects existing in existing methods (the association problem in dense scenarios and the problem of error expansion caused by asynchronous tracks), and improves the association performance.

[0069] The following introduces the device embodiments of this application, which can be used to execute an asynchronous track association method based on the triplet network in the above embodiments of this application. For the details not disclosed in the device embodiments of this application, please refer to the method embodiments of this application above.

[0070] Reference Figure 6 As shown, an asynchronous track association device 600 based on a triple network according to an embodiment of the present application includes: A preprocessing unit 601 for preprocessing the track data of two information sources; A first training unit 602 for training a deep track feature model based on the preprocessed track data and a triple network; A first calculation unit 603 for calculating the track features to be associated based on the preprocessed track data and the deep track feature model; A second calculation unit 604 for calculating the correlation measure of two tracks based on the track features to obtain a track feature correlation vector; A second training unit 605 for training a binary classifier based on the track feature correlation vector and a binary classification network; A third calculation unit 606 for calculating a track association result based on the track feature correlation vector and the binary classifier.

[0071] As Figure 7 shown, an embodiment of the present application also provides an electronic device 700, including a memory 710, a processor 720, and a computer program 711 stored on the memory 710 and executable on the processor. When the processor 720 executes the computer program 711, the steps of the above-mentioned asynchronous track association method based on a triple network are implemented.

[0072] Since the electronic device introduced in this embodiment is the device adopted by an asynchronous track association device based on a triple network in an embodiment of the present application, based on the method introduced in an embodiment of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in an embodiment of the present application will not be described in detail here. As long as the device adopted by those skilled in the art to implement the method in an embodiment of the present application belongs to the scope to be protected by the present application.

[0073] In the specific implementation process, when the computer program 711 is executed by the processor, it can implement any implementation manner in the corresponding embodiment of the first aspect.

[0074] Figure 8 Shows a schematic structural diagram of a computer system of an electronic device suitable for implementing an embodiment of the present application.

[0075] It should be noted that Figure 8 the computer system 800 of the electronic device shown is only an example and should not bring any limitation to the functions and usage scope of the embodiments of the present application.

[0076] As shown Figure 8 in FIG. 800, a computer system 800 includes a central processing unit (CPU) 801 that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage section 808 into a random access memory (RAM) 803, such as executing the method described in the above embodiments. In the RAM 803, various programs and data required for system operations are also stored. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0077] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read from it can be installed into the storage section 808 as needed.

[0078] Specifically, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program codes for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 809, and / or installed from the removable medium 811. When the computer program is executed by a central processing unit (CPU) 801, various functions defined in the system of the present application are executed.

[0079] It should be noted that the computer-readable medium shown in the embodiments of the present application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. In the present application, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries the computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0080] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. Among them, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the above module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0081] The units involved in the embodiments of the present application can be implemented in software or in hardware, and the described units can also be provided in a processor. In some cases, the names of these units do not constitute a limitation on the units themselves.

[0082] As another aspect, the present application also provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes a method for asynchronous track association based on a triple network described in the above embodiments.

[0083] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist alone without being assembled into the electronic device. The above computer-readable medium carries one or more programs. When the one or more programs are executed by an electronic device, the electronic device implements a method for asynchronous track association based on a triple network described in the above embodiments.

[0084] It should be noted that although several modules or units of a device for action execution are mentioned in the above detailed description, such a division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more of the above-mentioned modules or units can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0085] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software or by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (such as a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0086] Other embodiments of the present application will be readily contemplated by those skilled in the art after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include the common general knowledge or conventional technical means in the technical field not disclosed in the present application. It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. An asynchronous track association method based on a triplet network, characterized in that: include: Preprocess the track data of two sources; A deep track feature model is obtained based on the preprocessed track data and triplet network training; Calculating the track features to be associated based on the preprocessed track data and the deep track feature model; Based on the track characteristics, correlation measurement calculation is performed on the two tracks to obtain a track characteristic correlation vector; A binary classifier is obtained based on the track feature correlation vector and binary classification network training; The track correlation result is obtained based on the track feature correlation vector and the binary classifier calculation.

2. The method according to claim 1, characterized in that The preprocessing of the track data of the two information sources includes: The track data of the two sources are normalized and supplemented.

3. The method according to claim 1, characterized in that The deep track feature model is obtained based on the pre-processed track data and triplet network training, including: Construct training samples of triplet network based on preprocessed track data; Input the training sample into the weight-shared long short-term memory network in the triple network for processing to obtain anchor point track features corresponding to the anchor point track, positive sample track features corresponding to the positive sample track, and negative sample track features corresponding to the negative sample track; Calculating a triplet loss function corresponding to the anchor point track feature, the positive sample track feature and the negative sample track feature; The gradient is calculated through back propagation of the Adam optimizer to update the triplet network parameters; Iterative training is performed until the triplet network converges or reaches a preset number of iterations to obtain the deep track feature model.

4. The method according to any one of claims 1 to 3, characterized in that The calculating the track features to be associated based on the preprocessed track data and the deep track feature model includes: A piece of track data is randomly selected from the preprocessed track data of the two information sources and input into the deep track feature extraction model to extract the track features to be associated.

5. The method according to claim 4, characterized in that The step of performing correlation measurement calculation on the two tracks based on the track characteristics to obtain a track characteristic correlation vector includes: The L1 norm distance algorithm is used to calculate the L1 norm of the track feature to be associated, and the track feature distance is obtained as the track feature correlation vector.

6. The method according to claim 5, characterized in that The binary classifier is obtained based on the track feature correlation vector and the binary classification network training, including: The track feature distance is used as the training sample of the binary classification network; Input the training sample into a binary classification network to obtain an association result; Calculate the cross entropy loss based on the association results; The gradient is calculated through back propagation of the Adam algorithm to update the network parameters of the binary classification network; Iterative training is performed until the binary classification network converges or reaches a preset number of iterations to obtain the binary classifier.

7. The method according to claim 6, characterized in that The track correlation result is obtained based on the track feature correlation vector and the binary classifier, including: The track feature correlation vectors of the two information sources are input into the binary classifier, and the track correlation result is calculated.

8. An asynchronous track association device based on a triplet network, characterized in that: include: A preprocessing unit, used for preprocessing the track data of two information sources; A first training unit is used to obtain a deep track feature model based on the preprocessed track data and triplet network training; A first calculation unit, configured to calculate the track features to be associated based on the preprocessed track data and the deep track feature model; A second calculation unit, configured to perform correlation measurement calculation on the two tracks based on the track characteristics to obtain a track characteristic correlation vector; A second training unit is used to obtain a binary classifier based on the track feature correlation vector and the binary classification network training; The third calculation unit is used to calculate and obtain the track correlation result based on the track feature correlation vector and the binary classifier.

9. A computer-readable storage medium, characterized in that: The storage medium stores computer instructions, and when the computer instructions are executed on a computer, the computer is enabled to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer instructions; The processor is used to call the computer instructions stored in the memory so that the electronic device executes the method as described in any one of claims 1-7.

Images