Maneuvering target tracking method and system based on interactive multi-model and intention reasoning

Abstract

The invention discloses a maneuvering target tracking method and system based on interactive multi-model and intention reasoning, and belongs to the technical field of target tracking. According to the maneuvering target tracking method based on interactive multi-model and intention reasoning, a random context-independent grammar is introduced to model a target behavior intention, the target behavior intention and an interactive multi-model filtering framework are subjected to closed-loop fusion, specifically, a dominant motion mode is quantized into a character sequence, grammar analysis is performed by using SCFG, and the target behavior intention is tracked. Therefore, accurate recognition of the structured behavior intention is achieved, the intention posterior probability obtained through analysis is fed back to the weight of the IMM model for fusion, closed-loop optimization is formed, the target tracking and positioning precision can be improved, and inherent defects caused by discrete processing of motion state estimation and behavior intention reasoning in the prior art are effectively overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of target tracking, in particular to a maneuvering target tracking method and system based on interactive multiple models and intention reasoning. BACKGROUND

[0002] In the technical field of target tracking, especially for the tracking of maneuvering targets, the traditional method usually treats the estimation of the motion state of the target (tracking) and the identification of the behavior intention (reasoning) as two separate processes.

[0003] The traditional interactive multiple model (IMM) tracking algorithm is based on a finite state Markov chain, and the state transition of which can only represent the maneuvering change on a time scale of seconds (such as acceleration and turning). However, the behavior pattern driven by the intention of the target (such as circling around a sensitive area or wandering along a preset path) often involves long-range spatiotemporal dependence on a time scale of minutes. The random walk assumption of the Markov model will lead to the over-flattening of the posterior distribution, and cannot effectively model such structured path patterns. Although the existing reasoning methods such as Bayesian networks can model complex behavior patterns, the computational complexity thereof increases exponentially with the level of the pattern, and cannot meet the real-time requirements of special area monitoring and other scenarios. This forces the system designer to make a compromise between the model expression ability and the computational efficiency.

[0004] The above defects make it difficult for the existing system to meet the actual requirements, especially in some key fields, where the delay or misjudgment of intention identification may cause serious consequences. SUMMARY

[0005] The purpose of the present application is to provide a maneuvering target tracking method and system based on interactive multiple models and intention reasoning, so as to overcome the problem of insufficient accuracy of target tracking and positioning results caused by the decoupling of the motion tracking and intention reasoning modules in the prior art.

[0006] The present application solves the above technical problems by the following technical solutions: The present application provides a maneuvering target tracking method based on interactive multiple models and intention reasoning, comprising the following steps: Initializing the parameters of the interactive multiple model, configuring the grammar rule library of the random context-free grammar; based on the sensor measurement data, performing interactive multiple model filtering tracking, and performing real-time filtering estimation on the motion state of the target to obtain the state estimation and model weight at the current time; According to the model weight, determining the dominant motion pattern, and quantizing the dominant motion pattern into a discrete character and adding it to the track segment string; Based on the grammar rule library of the random context-free grammar, performing syntax analysis on the track segment string to calculate the posterior probability of the track segment string belonging to each predefined intention; The posterior probability is fused with the model weight to generate a fused weight, and the fused weight is fed back to the interactive multi-mode for filtering tracking at the next moment.

[0007] The application further improves the real-time filtering estimation of the target motion state by reassigning the speed in different speed modes to the speed in the state variable and adjusting the model weight according to the residual error to obtain the state estimation at the current moment.

[0008] The application further improves the interactive multi-model parameters, which include a state transition matrix, a model weight, an initial state estimation and a covariance matrix; and the grammar rule library includes non-terminal symbols, terminal symbols and production rules corresponding to straight lines, curved lines and rectangular trajectories.

[0009] The application further improves the syntax analysis of the track segment string by using the Earley-Stolcke algorithm.

[0010] The application further improves the syntax analysis of the track segment string by using the Earley-Stolcke algorithm, which includes three state processes of Predict, Scan and Complete. The Predict state is to predict the non-terminal symbol that is not matched, generate a child node and update the forward probability and the inline probability; the Scan state is to move the point operator and update the probability if the terminal symbol matches the input character; and the Complete state is to backtrack and update the state and probability of the parent node when the non-terminal symbol is completely matched.

[0011] The application further improves the model set of the interactive multi-model filter, which is a plurality of discretized speed direction models; and the discretized speed direction model is a speed vector in eight directions.

[0012] The application further provides a maneuvering target tracking system based on the interactive multi-model and intention reasoning, which includes: The first module is used for initializing the parameters of the interactive multi-model, configuring the grammar rule library of the random context-free grammar, performing the interactive multi-model filtering based on the sensor measurement data to obtain the target state estimation and the model weight at the current moment; The second module is used for determining the dominant motion mode according to the model weight and quantizing the dominant motion mode into a discrete character to be added to the track segment string. The third module is used for performing the syntax analysis of the track segment string based on the grammar rule library of the random context-free grammar to calculate the posterior probability of the track segment string belonging to each predefined intention. A fourth module is configured to fuse the posterior probability with the model weight to generate a fused weight, and feed the fused weight back to the interactive multi-model for filtering tracking at the next moment.

[0013] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method for maneuvering target tracking based on an interactive multi-model and intention reasoning when executing the computer program.

[0014] The application further provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the method for maneuvering target tracking based on an interactive multi-model and intention reasoning when executed by a processor.

[0015] The application further provides a computer program product, which comprises a computer program, and the computer program implements the steps of the method for maneuvering target tracking based on an interactive multi-model and intention reasoning when executed by a processor.

[0016] Compared with the prior art, the application has the following positive progress effects: The method for maneuvering target tracking based on an interactive multi-model and intention reasoning provided by the application introduces a stochastic context-free grammar to model the target behavior intention, and fuses the stochastic context-free grammar with an interactive multi-model filtering framework in a closed loop, specifically, quantizes a dominant motion mode into a character sequence, performs syntax analysis by using the stochastic context-free grammar, so as to accurately identify a structured behavior intention, feeds the intention posterior probability obtained by analysis back to the interactive multi-model weight for fusion, forms a closed loop optimization, and can improve target tracking positioning accuracy, and effectively overcomes inherent defects caused by separate processing of motion state estimation and behavior intention reasoning in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the application. The schematic embodiments of the application and their descriptions serve to explain the application, and do not constitute an improper limitation on the application.

[0018] Figure 1 A flowchart of the method for maneuvering target tracking based on an interactive multi-model and intention reasoning is shown in FIG. 1.

[0019] Figure 2 The eight-directional quantized velocity vector is used as a model set of the IMM algorithm, and is also used for string modeling of the SCFG.

[0020] Figure 3 The syntax rule of the application (from left to right, straight line, arc line and rectangle).

[0021] Figure 4Schematic diagram of the intention reasoning-based maneuvering target tracking system of the present application.

[0022] Figure 5 Trajectory diagram of experiment one of the present application.

[0023] Figure 6 Result diagram of experiment one of the present application.

[0024] Figure 7 Target motion trajectory diagram of experiment two of the present application.

[0025] Figure 8 Corresponding motion pattern diagram of the method trajectory of the present application.

[0026] Figure 9 Motion pattern recognition accuracy of the method of the present application.

[0027] Figure 10 RMSE comparison diagram of the method of the present application and the traditional IMM.

[0028] Figure 11 Comparison diagram of the real trajectory and the tracking trajectory of different algorithms of the method of the present application. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0030] In the description of the present application, it should be understood that the terms “include” and “contain” indicate the existence of described features, whole, steps, operations, elements and / or components, but do not exclude the existence or addition of one or more other features, whole, steps, operations, elements, components and / or sets thereof.

[0031] It should also be understood that the terms used in the present application specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, unless otherwise clearly indicated by the context, the singular forms “a”, “an” and “the” are intended to include the plural forms.

[0032] It should be understood that, although the terms first, second, third, etc. can be employed in describing the preset ranges, etc. in the embodiments of the present application, the preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from each other. For example, the first preset range can also be referred to as the second preset range, and similarly, the second preset range can also be referred to as the first preset range, without departing from the scope of the embodiments of the present application.

[0033] Depending on the context, the word "if" as used herein can be interpreted to mean "when" or "while" or "in response to determining" or "in response to detecting." Similarly, the phrase "if it is determined" or "if [a stated condition or event] is detected" can be interpreted to mean "upon determining" or "in response to determining" or "upon detecting [the stated condition or event]" or "in response to detecting [the stated condition or event]."

[0034] The present application is further described in detail below with reference to the accompanying drawings and specific examples. The following description is explanatory only and not limiting of the present application.

[0035] The interactive multiple model (IMM) method is a multi-model filtering algorithm, especially suitable for target tracking scenarios with maneuvering behavior (such as sudden acceleration, turning), and can be applied to radar, autonomous driving, unmanned aerial vehicle navigation, and robot perception. Traditional single model filtering (such as Kalman filtering) produces large errors when the target maneuvers, because it assumes that the target motion pattern is fixed (such as uniform speed). In actual scenarios, the target may frequently switch motion patterns, and an adaptive multi-model fusion method is needed. The core idea of IMM is to filter multiple motion models in parallel, dynamically weight the results, and adapt to target maneuvers in real time. First, a model pool is set: a set of typical motion models (such as uniform speed, uniform acceleration, turning model) is preset. Then, the model probability is calculated: the confidence of each model matching the current motion is dynamically calculated according to the observation data. Finally, the interactive fusion is performed: the outputs of each model are weighted based on the probability to obtain a robust state estimation. For example, in a maneuvering motion, the target changes from approximately uniform motion (CV) to approximately uniform acceleration motion (CA), and the IMM algorithm will calculate the model matching degree through the residual between the prediction of each model and the measurement after receiving the measurement. The CA motion model probability will increase, and the CV motion model probability will decrease. After weighting the filtering results of each model, a more accurate estimation result than single model filtering (using CA or CV as the motion model) is obtained.

[0036] A stochastic context-free grammar (SCFG) is used to model the intent as a probabilistic context-free grammar rule, and the most likely intent structure is calculated by parsing the syntax tree. The general process is to first quantify and generate the velocity track segment, which is quantified as a string form, and the system input is the position, velocity and other kinematic state estimates output by the target tracker. The continuous velocity vector is mapped to the discrete track segment symbol, i.e. the string form (abcdefgh represents eight directions respectively) through the quantization operation. Through the quantized velocity track segment, the modeling and intent inference of SCFG can be realized. How SCFG describes the trajectory pattern of the velocity track segment through the generative rule will be described in detail below. The idea is to define the corresponding grammar for different intents, and the grammar probability implicitly indicates the priority of the target intent. For example, in the trajectory approaching the sensitive area, the production probability of the "detour" mode is significantly higher than that of the random maneuver. By extending the Earley-Stolcke parsing algorithm, the most likely generative rule for a given track segment sequence is calculated in real time, and the intent category with the highest posterior probability is output.

[0037] Referring to Figure 1 , the application provides a maneuvering target tracking method based on intent reasoning, which parses the target trajectory semantics through a grammar model to realize closed-loop tracking optimization, and specifically comprises the following steps: S1: initializing tracking parameters and a grammar model The state transition matrix, model weight, initial state estimate and covariance matrix of the interactive multiple model (IMM) are initialized, and the grammar rule library of the stochastic context-free grammar (SCFG) is configured. Specifically, it includes: The target motion model set is defined as eight discrete velocity directions, as shown in Figure 2 ; the initial weight of each model and the state transition probability matrix π are set; the initial state estimate and the state covariance matrix of each model are set. The SCFG grammar rule library is constructed, and non-terminal symbols, terminal symbols and production rules RulesLine, RulesArc and RulesRec are designed for typical intents: for a straight line, it belongs to a regular grammar, and all production rules are shown in Figure 3 The non-terminal symbol of an arc line is set as , the terminal symbol is set as , and the production rule setting is shown in Figure 3 . The non-terminal symbol of a rectangle is set as , the terminal symbol is set as , and the production rule setting is shown in Figure 3 .

[0038] Table 1 Production rules of three track shapes

[0039] Initialize the velocity track segment string str to be empty.

[0040] S2: Perform IMM filtering tracking and state update Real-time filtering estimation of the target motion state is performed by the IMM algorithm. In this step, the present invention innovatively. From the perspective of the state transition equation formula, in the common IMM algorithm, different model sets have different motion models F, while in the present invention, different model sets are set to have different velocities, and the velocities are assigned to the target state when each model is used for one-step prediction.

[0041] Input sensor measurement data (position, velocity), perform multi-model interactive filtering, update model weight u and state estimation X; Determine the dominant motion mode according to the current model weight, quantize the corresponding velocity direction to a discrete character (abcdefgh eight-direction encoding), and add it to the track segment string str.

[0042] The specific process is as follows: 1. Read the last time weight and model transition matrix The model weight of the last time:

[0043] Markov transition probability matrix: , where represents the probability of "model i at time t-1 transitions to model j at time t".

[0044] 2. Calculate the mixing probability

[0045] For each current model i, the "prediction" weight transferred from all last time models j is summarized:

[0046] 3. Calculate the conditional mixing probability matrix U(j,i) U(j,i) represents "if the current selected model i, then the proportion of model j at the last time":

[0047] 4. Mix the state and covariance Weight the last time state and covariance of each model by U(j,i):

[0048] 5. Conditional filtering, calculate the likelihood A(i) For each model i, use the square root volume Kalman filter to make one-step prediction and update, and output the updated state , covariance , likelihood A(i) (Gaussian probability density value of the current observation under model i).

[0049] 6. Posterior weight update Finally, the prior mixture probability c(i) and the likelihood A(i) are fused by Bayes formula to get the posterior weight at time t:

[0050] The new weight is used to weight the updated state of each model to get the final estimate

[0051]

[0052] And the is taken as the input of the next iteration, and the above steps are repeated.

[0053] According to the IMM algorithm, the is selected as the current model with the maximum probability, and it is added to the end of the string str for subsequent SCFG use.

[0054] S3: SCFG parsing and intent probability calculation Earley algorithm is an excellent top-down interval dynamic programming algorithm based on point operator, which is used for parsing SCFG / CFG (stochastic context-free grammar). Due to the existence of transition probability in SCFG, it makes each syntactic tree correspond to a certain probability. See Table 2 for the advantages and disadvantages of Earley algorithm as follows: Table 2 Advantages and disadvantages of Earley algorithm

[0055] Earley algorithm is divided into three states, and Earley algorithm defines two probabilities: forward probability: is the sum of all path probabilities ending with the current state. Inner probability: is the sum of all path probabilities starting with the current state and matching the length of characters backward. For different states, there are different ways of calculating probability, as follows: Predict:

[0056] The above formula indicates that the characters from position j to i have been matched, and the A non-terminal needs to be predicted. The prediction process is a process of generating a child node.​

[0057] Scan:

[0058] The above formula indicates that the characters from j to i positions have been matched, corresponding to the c terminator, if the c terminator matches the real input at this time, the right shift of the dot operator can be realized. A new state is obtained. The state is marked as Scan.

[0059] Complete:

[0060] The above formula indicates that all elements corresponding to A have been matched, which also means that the A non-terminator has also been matched, so the following state can be obtained:

[0061] Through the above SCFG grammar, the parse tree corresponding to the string can be generated, and the parse tree is a tree structure used to parse the multiple possible structures of the string generated by the stochastic context-free grammar and mark the probability and other related information. The following directly displays the SCFG parser through a simple example.

[0062] Suppose the grammar rule is

[0063] In order to parse the string aab, the state set evolution is as follows according to the above rule process:

[0064] The parse tree can be visualized as follows:

[0065] Through the above SCFG grammar, the trajectory intention of the speed trajectory segment can be realized.

[0066] In a specific embodiment of the present application, the SCFG parser obtains the mixed probability through the following steps : 1. Weighted fusion: using the posterior probability of the current grammar structure obtained by the SCFG parser as the weight, the prediction probabilities of the three grammar structures are weighted and averaged: (12) Where prob ln, prob arc and prob rec are the likelihoods of each mode (a-h) obtained by the SCFG parser (for example, prob ln is the probability array of the next character being a-h under the straight line grammar); 2. Normalization: if ​If non-zero, normalization is performed; if If all values ​​are zero (no effective prediction), then the value regresses to a uniform distribution [0.125, ..., 0.125]. S4: Closed-loop feedback and weight fusion After fusing the intent inference result with the IMM filtering result, a single-frame tracking loop is completed. The result is then used as the input to the IMM parser for the next frame, enabling dynamic optimization. Update model probabilities: Model probability compared to the previous time step Perform exponential smoothing: (13) Combining IMM filtering results ( ) and the current syntax fusion result ( This enables dynamic adjustment; it adjusts the state transition matrix. This enhances the model's probability of conforming to the semantic intent. Model weights and state transition matrix The data is then passed to the next IMM, starting a new loop. The fusion weights can be adaptively adjusted based on the predictability of the actual scenario or target, and are not limited to a fixed 0.5.

[0067] Based on the same inventive concept, this invention also provides a maneuvering target tracking system based on interactive multi-model and intent reasoning, which can perform the above-described method. See also Figure 4 The system may specifically include: The initialization module is used to initialize the state transition matrix, model weights, initial state estimates, and covariance matrix of the interactive multi-model (IMM), and to configure the syntax rule base of the stochastic context-free grammar (SCFG).

[0068] The filtering and tracking module is used to perform multi-model interactive filtering, monitor the target's motion state in real time, and quantize the dominant motion pattern into discrete characters based on the model weights, thereby updating the track segment string.

[0069] The parsing and calculation module is used to call the SCFG parser to perform syntax parsing on the track segment string and calculate the meaning of different trajectories such as straight lines, arcs, and rectangles. Figure 4 The posterior probability.

[0070] The closed-loop feedback module is used to perform weighted fusion of the intent probability output by the parsing and calculation module and the model weight output by the filtering and tracking module, and feed the fused result back to the filtering and tracking module for dynamic optimization in the next time step.

[0071] The performance of the method is verified by simulation experiment. In the experimental design part, two main experiments are implemented. The first experiment is mainly used for verifying the effectiveness of the Earley Stolcke algorithm, and the second experiment is used for verifying whether the SCFG grammar has performance improvement on the traditional target tracking positioning problem.

[0072] Experiment one: For example: for the speed track segment: aahabbbbccccdddedd, it can be considered as a rectangular track segment, and the specific trajectory is shown in Figure 5 ; the Earley Stolcke algorithm is applied to parse the above string track segment according to different track modes, and the result is shown in Figure 6 .

[0073] In order to verify the effectiveness, specific probability values need to be set for the SCFG production rules of the three track modes (straight line, arc line, and rectangle) defined in Figure 3 . In this experiment, higher probabilities are given to the rules that meet the typical characteristics of each mode, and lower probabilities are given to the rules that are not typical or represent the end, so as to simulate the real intention recognition scene. For example, for the rectangular grammar, the production rule probability of describing continuous straight line is higher, and the rule probability of describing turning is relatively lower.

[0074] In addition, the simulated string aahabbbbccccdddedd has intentionally added interference characters (such as h and e), which do not belong to the core components of any preset track mode. By setting very low generation probability or not assigning rules to these interference characters, the robustness of the parsing algorithm used in the application when facing non-ideal input is tested.

[0075] Simulation result interpretation: in the initial stage, the system considers that the current track is only a straight line, because there is no sign of turning. After the first turning occurs, the system considers that the current track segment is an arc track segment with the highest probability. But after the third turning occurs, the system considers that the rectangular track segment is a more suitable choice, which is consistent with the subjective idea. In addition, in the current experiment, interference characters are added in the simulated string, but this does not affect the result of the overall track recognition of the system, which also shows that the current algorithm has a certain robustness.

[0076] Experiment two: Experiment two aims to compare the tracking performance difference between the traditional IMM algorithm and the IMM algorithm embedded with SCFG of the application. The design scene is as shown in Figure 7 , the target makes an approximate uniform motion of rectangular trajectory, and the corresponding quantized speed direction vector category changes over time can be described in the form of Figure 8 .

[0077] The evaluation index is the root mean square error RMSE, defined as: (14) wherein, and respectively represent the true value and the i-th Monte Carlo simulation estimated value of the to-be-determined target position coordinate, and L is the total number of Monte Carlo simulations.

[0078] In order to make a fair and comprehensive comparison between the algorithm (SCFG-IMM) proposed in the application and the traditional IMM algorithm, the second experiment designs specific simulation scenarios and parameters, which are set as follows: 1. Simulation scenario and target trajectory: In this simulation, the true trajectory of the target is a rectangular trajectory generated by combining four orthogonal uniform straight line motion models, with a total duration of 50 seconds. The specific segments are as follows: 1-12 seconds: the target travels to the right at an approximate speed of [10, 0] m / s.

[0079] 13-25 seconds: the target switches to an approximate speed of [0, 10] m / s and travels upwards.

[0080] 26-37 seconds: the target switches to an approximate speed of [-10, 0] m / s and travels to the left.

[0081] 38-50 seconds: the target switches to an approximate speed of [0, -10] m / s and travels downwards.

[0082] 2. Observation model parameters Sampling period: set to T = 1 second.

[0083] Measurement noise: simulate the observation data of GMTI (Ground Moving Target Indication Radar), with zero-mean Gaussian white noise added. The covariance matrix R is set to diag([0.025, 0.0000019, 0.00001]), corresponding to the measurement variances of range, azimuth and Doppler velocity respectively. This setting represents a high-precision observation scenario, aiming to test the algorithm's ability to finely track the trajectory details in a low-noise environment.

[0084] 3. Algorithm initialization and comparison settings: To ensure the effectiveness of the comparison, the two algorithms share some identical initialization parameters: Process noise: the process noise power spectral density of the filter model is set to 0.1.

[0085] Initial state covariance: the initial state covariance matrix of the two algorithms All are set to diag([0.01, 0.001, 0.01, 0.001]).

[0086] The core difference between the two algorithms lies in the nature and updating mechanism of their Markov transition probability matrix Π: The comparative algorithm (traditional IMM algorithm): this algorithm adopts the same 8 discrete direction uniform motion model set as the present application. However, the key lies in that its Markov transition probability matrix Π is set as a static, uniformly distributed matrix, that is, the probability of transition between models in any direction is equal. This represents no prior knowledge of the target's next time maneuver, making it a "reactive" tracker that relies entirely on the current observation.

[0087] The algorithm of the present application (SCFG-IMM): the core advantage of the present application lies in the introduction of high-level semantic cognition. Through the SCFG grammar analyzer, the algorithm can learn and infer the long-term motion intention of the target (such as "executing a straight-line maneuver") from the historical trajectory of the target. Subsequently, the intention judgment is used to dynamically modify the Markov transition probability matrix Π. For example, when the "straight-line" intention is identified, the system will greatly increase the transition probability of "maintaining straight-line", while suppressing the probability of "sudden turn". Therefore, the present application is a "cognitive-reactive" tracker that revalues the state transition by fusing long-term intention and short-term observation, so as to make more stable and accurate tracking.

[0088] The target mode recognition accuracy results of Monte Carlo simulation 100 times are shown in Figure 9 It can be seen that the present application has always maintained a higher recognition accuracy compared with the traditional IMM throughout the whole process. Especially when the target makes a maneuver based on abnormal trajectory, the present application can still maintain a very high accuracy, while the traditional IMM cannot accurately identify the target motion mode when the maneuver occurs. The tracking accuracy comparison of the two methods is shown in Figure 10 To more intuitively show the tracking effect, Figure 11 the path comparison of the two algorithms with the real trajectory is given. From the figure, it can be clearly seen that the method proposed in the present application can more closely fit the real trajectory when the target makes a turn, and the tracking accuracy is significantly better than the traditional IMM algorithm.

[0089] Table 3 Performance comparison of the present application and the IMM algorithm

[0090] As can be seen from Table 3, whether it is the position RMSE or the velocity RMSE, the present application has a significant performance improvement. This shows that the SCFG-IMM makes full use of the target trajectory results obtained by the Earley algorithm, so that the target tracking and positioning results are greatly improved.

[0091] In addition, the SCFG grammar rule library of the present application can also be extended as needed to identify more complex intent patterns such as "circling / wandering", "S-shaped maneuver", etc.

[0092] The results of Experiment Two show that the present application has a significant improvement in target trajectory prediction and tracking positioning compared to traditional methods, especially in handling target behavior in complex environments, showing higher accuracy and stability. The experimental results show that the root mean square error (RMSE) of the target trajectory prediction of the present application is reduced by about 50% compared to traditional methods, indicating its potential in high-precision target tracking applications.

[0093] The method of the present application provides a new approach in the field of maneuvering target tracking, i.e., combining intent reasoning technology in the tracking process to improve the intelligent level of the tracking system. Future research can further optimize the algorithm structure and explore more practical application scenarios to achieve more efficient target tracking.

[0094] Based on the same inventive concept, the embodiments of the present application provide a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the maneuvering target tracking method based on interactive multiple models and intent reasoning when executing the computer program. The memory can include an internal memory, such as a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk memory. The processor, network interface, and memory are connected to each other through an internal bus, which can be an industry standard architecture bus, a peripheral component interconnect standard bus, an extended industry standard architecture bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs, specifically, the programs can include program codes, and the program codes include computer operation instructions. The memory can include an internal memory and a non-volatile memory, and provide instructions and data to the processor.

[0095] Based on the same inventive concept, the embodiments of the present application provide a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the maneuvering target tracking method based on interactive multiple models and intent reasoning. Specifically, the computer readable storage medium includes but is not limited to, for example, a volatile memory and / or a non-volatile memory. The volatile memory can include a RAM (Random Access Memory) and / or a cache memory, etc. The non-volatile memory can include a ROM (Read-Only Memory), a hard disk, a flash memory, an optical disk, a magnetic disk, etc.

[0096] Based on the same inventive concept, the embodiment of the present application provides a computer program product, which comprises a computer program stored on a computer readable storage medium, and the computer program comprises program instructions, which, when executed by a computer device, cause the computer device to perform the steps of the above-mentioned method for tracking a maneuvering target based on interactive multi-model and intention reasoning.

[0097] Those skilled in the art should understand that the embodiments of the present application can be provided as a method or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM (Compact Disc Read-Only Memory), optical storage, etc.) containing computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the method, device (system), computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer device or other programmable data processing device produce the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flow(s) or block(s).

[0099] These computer program instructions can also be stored in a computer readable storage medium that can direct the computer device or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable storage medium produce the manufactured product comprising the instruction means, which implement the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flow(s) or block(s).

[0100] These computer program instructions can also be loaded into the computer device or other programmable data processing device, so that a series of operation steps are performed on the computer device or other programmable device to produce a processing implemented by the computer device, so that the instructions executed on the computer device or other programmable device provide the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1steps of the functions specified in the block or blocks.

[0101] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the preferred embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to encompass within their scope all such variations and modifications as are included within the scope of the application.

[0102] It is clear that the application can be subject to many modifications and variants without thereby departing from the scope of the application. Thus, the application is intended to encompass all such modifications and variants which fall within the scope of the application, as defined by the following claims and their equivalents.

Claims

1. A maneuvering target tracking method based on interactive multi-model and intent reasoning, characterized in that, Includes the following steps: Initialize the parameters of the interactive multi-model and configure the syntax rule base of the random context-free grammar; based on sensor measurement data, perform interactive multi-model filtering and tracking to perform real-time filtering estimation of the target motion state and obtain the current state estimate and model weights. The dominant motion pattern is determined based on the model weights, and then quantized into discrete characters and added to the track segment string. Based on a grammar rule base of random context-free grammar, the track segment string is parsed to calculate the posterior probability of the track segment string belonging to each predefined intent. The posterior probability is fused with the model weights to generate fused weights, which are then fed back to the interactive multi-mode for filtering and tracking in the next time step.

2. The maneuvering target tracking method based on interactive multi-model and intent reasoning according to claim 1, characterized in that, When performing real-time filtering estimation of the target's motion state, the velocity in different velocity modes is reassigned to the velocity in the state variables, and the model weights are adjusted according to the residuals. The state estimate at the current moment is obtained through state fusion.

3. The maneuvering target tracking method based on interactive multi-model and intent reasoning according to claim 1, characterized in that, The parameters of the interactive multi-model include the state transition matrix, model weights, initial state estimates, and covariance matrix; the syntax rule base contains non-terminal symbols, terminal symbols, and production rules corresponding to straight lines, arcs, and rectangular trajectories.

4. The maneuvering target tracking method based on interactive multi-model and intent reasoning according to claim 3, characterized in that, The Earley-Stolcke algorithm is used to perform syntax parsing on the track segment strings.

5. The maneuvering target tracking method based on interactive multi-model and intent reasoning according to claim 4, characterized in that, The Earley-Stolcke algorithm is used to perform syntax parsing on track segment strings, including three states: Predict, Scan, and Complete. The Predict state is used to predict unmatched non-terminal symbols, generate child nodes, and update the forward and inline probabilities; the Scan state is used to move the dot operator and update the probabilities if the terminator matches the input character; and the Complete state is used to backtrack and update the state and probability of the parent node when the non-terminal symbol is completely matched.

6. The maneuvering target tracking method based on interactive multi-model and intent reasoning according to claim 1, characterized in that, The model set for interactive multi-model filtering consists of multiple discretized velocity direction models; the discretized velocity direction models are velocity vectors in eight directions.

7. A maneuvering target tracking system based on interactive multi-model and intent reasoning, characterized in that, include: The first module is used to initialize the parameters of the interactive multi-model and configure the syntax rule base of the random context-free grammar; based on sensor measurement data, it performs interactive multi-model filtering to obtain the target state estimate and model weights at the current time. The second module is used to determine the dominant motion mode based on the model weights, and quantize the dominant motion mode into discrete characters, which are then added to the track segment string. The third module is used to perform grammatical parsing on track segment strings based on a random context-free grammar rule base, and to calculate the posterior probability that the track segment string belongs to each predefined intent. The fourth module is used to fuse the posterior probability with the model weights to generate fused weights, and then feeds the fused weights back to the interactive multi-mode for filtering and tracking in the next time step.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the maneuvering target tracking method based on interactive multi-model and intent reasoning as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the maneuvering target tracking method based on interactive multi-model and intent reasoning as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the maneuvering target tracking method based on interactive multi-model and intent reasoning as described in any one of claims 1 to 6.

Images