Multi-model maneuvering target tracking method and device based on partial state interaction

Abstract

The invention discloses a multi-model maneuvering target tracking method and device based on partial state interaction, and belongs to the field of maneuvering target tracking. The method comprises the steps that firstly, maneuvering models of the same structure are used for jointly describing a maneuvering mode, maneuvering parameter jump modeling is conducted on the maneuvering models into switching between the homogeneous maneuvering models, and each maneuvering model initializes an exclusive maneuvering parameter subspace of the maneuvering model through a partitioned Gaussian probability density function; secondly, a partial state interaction mechanism is innovatively designed, model maneuvering parameter input is limited as a weighted combination of self historical estimation and parameter initial values, and cross interference of parameter estimation between models is eliminated; according to the technical scheme, matching coverage and consistent initialization of a parameter space can be realized when maneuvering parameters jump; and the optimal initialization filter is protected from being interfered by error parameter estimation in a maneuvering mode sojourn section. The problem of maneuvering parameter jump can be effectively solved, and the tracking precision of the maneuvering target is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of maneuvering target tracking, and in particular to a multi-model maneuvering target tracking method and device based on partial state interaction. Background Art

[0002] Maneuvering target tracking is a technique for estimating the state of a target with time-varying motion characteristics based on sensor observation data. Maneuvering parameters are physical quantities used to characterize the target's maneuvering motion, including kinematic parameters such as acceleration and turn rate. Tracking methods based on a single model often struggle to meet high-precision tracking requirements due to a mismatch between the pre-set model and the actual motion pattern. In contrast, multi-model approaches improve adaptability to target maneuvers by constructing a collection of models of the target's potential motion patterns.

[0003] In related technologies, existing multi-model methods often adopt parametric design methods, that is, discretizing the maneuvering parameter space into a finite number of quantized points, each quantized point corresponding to a fixed model parameter of a model; or structured design methods, that is, modeling the maneuvering parameters as random processes and augmenting them into state components for real-time estimation. However, these two design methods cannot effectively deal with maneuvering parameter jumps.

[0004] Based on this, there is an urgent need for a multi-model maneuvering target tracking method and device based on partial state interaction to solve the above technical problems. Summary of the Invention

[0005] The present invention provides a multi-model maneuvering target tracking method and device based on partial state interaction, which can solve the problem that related technologies cannot effectively deal with maneuvering parameter jumps. The technical solution is as follows:

[0006] In one aspect, a multi-model maneuvering target tracking method based on partial state interaction is provided, the method comprising:

[0007] Step 100: Establishing a motion model set based on the motion pattern of the maneuvering target to be tracked, partitioning the maneuvering parameter space of the maneuvering target into regions, and determining partition parameters of each partition and a state equation for each motion model; wherein the motion model includes a non-maneuvering model and a maneuvering model, and the same-structure maneuvering model is used to describe a maneuvering pattern of the maneuvering target;

[0008] Step 102, initializing the states and model probabilities of all the motion models, and setting the model transition probabilities;

[0009] Step 104, at each filtering moment, calculating the mixing probability of the motion model according to the model probability and the preset model transition probability;

[0010] Step 106: Perform partial state interaction on all the motion models according to the mixing probability and the partition parameter, input the interaction result into the filter corresponding to the motion model, and filter based on the observation data obtained at the current filtering time to obtain the conditional state estimate and likelihood function of each motion model;

[0011] Step 108 , based on the likelihood function, the model probability of each motion model is updated, and the model conditional state estimation is estimated and fused based on the updated model probability to obtain the final state estimation value of tracking the maneuvering target at the current moment.

[0012] On the other hand, a multi-model maneuvering target tracking device based on partial state interaction is provided, the device comprising:

[0013] a partitioning module, configured to establish a motion model set based on the motion pattern of the maneuvering target to be tracked, partition the maneuvering parameter space of the maneuvering target into regions, determine partition parameters of each partition and a state equation of each motion model; wherein the motion model includes a non-maneuvering model and a maneuvering model, and the same-structure maneuvering model is used to describe a maneuvering pattern of the maneuvering target;

[0014] An initialization module, configured to initialize the states and model probabilities of all the motion models, and set the model transition probabilities;

[0015] A calculation module, configured to calculate the mixing probability of the motion model at each filtering moment according to the model probability and a preset model transition probability;

[0016] a filtering module, configured to perform partial state interaction on all the motion models according to the mixing probability and the partition parameter, input the interaction result into a filter corresponding to the motion model, and filter the conditional state estimate and likelihood function of each motion model based on the observation data obtained at the current filtering moment;

[0017] A fusion module is used to update the model probability of each motion model according to the likelihood function, and to estimate and fuse the model conditional state estimate according to the updated model probability to obtain the final state estimate of tracking the maneuvering target at the current moment.

[0018] On the other hand, a computer device is provided, which includes a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of the above-mentioned multi-model maneuvering target tracking method based on partial state interaction.

[0019] On the other hand, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of the multi-model maneuvering target tracking method based on partial state interaction are implemented.

[0020] On the other hand, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the above-mentioned multi-model maneuvering target tracking method based on partial state interaction.

[0021] The present invention provides a multi-model maneuvering target tracking method and device based on partial state interaction, belonging to the field of maneuvering target tracking. To address the model parameter deviation problem existing in existing multi-model methods in maneuvering parameter jump scenarios, an innovative solution is proposed: first, a maneuvering mode is jointly described by maneuvering models with the same structure, and the maneuvering parameter jump is modeled as a switch between isomorphic maneuvering models. Each maneuvering model initializes its exclusive maneuvering parameter subspace using a partitioned Gaussian probability density function. Secondly, an innovative partial state interaction mechanism is designed to limit the model maneuvering parameter input to a weighted combination of its own historical estimation and the initial value of the parameter, eliminating cross-interference in parameter estimation between models. This technical solution can achieve matching coverage and consistent initialization of the parameter space when the maneuvering parameters jump; and protect the optimal initialization filter from interference from erroneous parameter estimates during the maneuvering mode stay period. The present invention can effectively address the problem of maneuvering parameter jumps and improve the tracking accuracy of maneuvering targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 This is a flow chart of a multi-model maneuvering target tracking method based on partial state interaction provided by one embodiment of the present invention;

[0024] Figure 2 is a schematic diagram of position estimation error provided by an embodiment of the present invention;

[0025] Figure 3 is a schematic diagram of speed estimation error provided by an embodiment of the present invention;

[0026] Figure 4 This is a structural diagram of a multi-model maneuvering target tracking device based on partial state interaction provided by one embodiment of the present invention;

[0027] Figure 5This is a hardware architecture diagram of a computer device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] As mentioned above, existing multi-model tracking methods usually adopt parametric design or structured design, which cannot effectively deal with the maneuver parameter jumps of the tracked target.

[0030] Based on this, the concept of the present invention is to describe the maneuver parameter jump as a switch between the same structural models through partial state interaction, and to process the maneuver parameter components in the state separately to achieve robust tracking under slow and sudden changes of maneuver parameters.

[0031] The specific implementation of the above concept is described below.

[0032] Please refer to Figure 1 An embodiment of the present invention provides a multi-model maneuvering target tracking method based on partial state interaction, the method comprising:

[0033] Step 100: Establishing a motion model set based on the motion pattern of the maneuvering target to be tracked, partitioning the maneuvering parameter space of the maneuvering target into regions, and determining partition parameters of each partition and a state equation for each motion model; wherein the motion model includes a non-maneuvering model and a maneuvering model, and the same-structure maneuvering model is used to describe a maneuvering pattern of the maneuvering target;

[0034] Step 102, initializing the states and model probabilities of all the motion models, and setting the model transition probabilities;

[0035] Step 104, at each filtering moment, calculating the mixing probability of the motion model according to the model probability and the preset model transition probability;

[0036] Step 106: Perform partial state interaction on all the motion models according to the mixing probability and the partition parameter, input the interaction result into the filter corresponding to the motion model, and filter based on the observation data obtained at the current filtering time to obtain the conditional state estimate and likelihood function of each motion model;

[0037] Step 108 , based on the likelihood function, the model probability of each motion model is updated, and the model conditional state estimation is estimated and fused based on the updated model probability to obtain the final state estimation value of tracking the maneuvering target at the current moment.

[0038] In an embodiment of the present invention, a structured maneuver model is first used to describe the target's maneuvers based on its possible motion patterns. Multiple maneuver models with the same structure are used to describe a specific target maneuver. The maneuver parameter space is then partitioned based on a set number of maneuver models with the same structure, with each partition corresponding to a maneuver model, to construct a filter model set. The filtering algorithm is then initialized, model transition probabilities are set, and the filter states and model probabilities are initialized. Recursive filtering then begins. The recursive filtering process sequentially involves model mixing probability calculation, partial state interaction, model matching filtering, model probability update, model condition estimation fusion, and recursive filtering until tracking is complete. This method achieves significantly better performance when the target's acceleration changes, improving the accuracy of maneuvering target tracking.

[0039] Described below Figure 1 How to perform the steps shown.

[0040] First, for step 100, a motion model set is established according to the motion pattern of the maneuvering target to be tracked, and the maneuvering parameter space of the maneuvering target is divided into regions, and the partition parameters of each partition and the state equation of each motion model are determined.

[0041] In this embodiment of the present invention, considering that the maneuvering parameters of a maneuvering target are not fixed, a structured model is used to describe the target's maneuvers to account for subtle changes in the maneuvering parameters. A set of motion models is then established based on the motion pattern. The motion model includes non-maneuvering models and maneuvering models. A structured maneuvering model is used to describe a maneuvering pattern of the maneuvering target. For example, multiple collaborative turning models (NCT) can be used to jointly describe the target's near-constant turning maneuvers.

[0042] Specifically, we assume that the target's possible motion modes include near-uniform velocity and near-uniform acceleration / deceleration within a plane. In this case, the motion model set consists of a two-dimensional near-uniform velocity (NCV) model and a two-dimensional near-uniform acceleration (NCA) model. The NCV model is a non-maneuvering model, and there is only one such model in the model set. The NCA model is a maneuvering model, and there can be many NCAs. All NCAs have the same model structure, i.e., identically structured maneuvering models.

[0043] It is worth noting that if the maneuvering target has other motion modes such as turning, the maneuvering model also includes a turning NCT model. There can be many turning models. Similarly, the structure of each turning model is the same and is specifically used to describe the turning motion of the target.

[0044] In an embodiment of the present invention, the maneuvering parameter space of the maneuvering target is divided into regions, and the partition parameters of each partition and the state equation of each motion model are determined, including the following steps: determining the maneuvering parameter space according to the maneuvering range of the maneuvering target to be tracked; dividing the maneuvering parameter space corresponding to the maneuvering model according to the number of the maneuvering models with the same structure, and calculating the partition parameters of each partition; establishing the state equation of the non-maneuvering model according to the position and velocity of the maneuvering target; and establishing the state equation of the maneuvering model with the same structure according to the position, velocity and maneuvering parameters of the maneuvering target.

[0045] Specifically, according to the number of maneuvering models with the same structure, the maneuvering parameter space of this type of maneuvering model is divided. For NCA, the possible acceleration range of the maneuvering target is the maneuvering parameter space of the maneuvering model, which is recorded as

[0046]

[0047] where l x and h x They represent the acceleration component a in the x direction respectively x The minimum and maximum values of l y and h y They represent the acceleration component a in the y direction y The minimum and maximum values of .

[0048] Assuming that the number of NCA models with the same structure is n, the maneuver parameter space is divided into n regions. For example, when n = 9, the acceleration is divided into three equal parts in the x and y directions, and n is used to divide the space into n regions. x and n y Respectively represents the number of segments of maneuvering parameters in the x and y directions, that is, n x =n y = 3. For another example, when n = 8, x and y can be divided into 2 × 4. The specific division method can be adjusted according to actual needs and will not be described in detail here.

[0049] Furthermore, assuming that the target prior acceleration obeys a uniform distribution, for the partitions obtained by division, the mean of the β-th (β∈{1,2,...,n}) acceleration partition is for:

[0050]

[0051] in, and is the component of the mean of the βth partition in the x and y directions; i represents the index of the segment in the x direction, and the number of segments in the x direction is nx , i is from 1 to n x .

[0052] The variance of the βth partition is:

[0053]

[0054] The acceleration mean and corresponding variance of the above partitions are the partition parameters.

[0055] Furthermore, the multi-motion model set constructed in this embodiment includes 1 NCV model and 9 NCA models. Each NCA model is assigned different initial maneuvering parameters, and these NCA models are effective in different acceleration partitions. The NCV model is denoted as m 1 , the 9 NCA models are marked as {m 2 ,…,m 10 The total number of models in the model set is denoted as r, and there are a total of r = n + 1 models.

[0056] The state vector of the NCV model is The state vector of the NCA model is where x k ,y k are the positions of the target in the x and y directions respectively, are the velocities in the x and y directions respectively, are the accelerations in the x and y directions respectively.

[0057] The NCV model m is established 1 The state equation is:

[0058]

[0059] in is the state transfer matrix of the NCV model, is the process noise gain matrix of the NCV model, To obey zero mean covariance is Gaussian white noise.

[0060] State transition matrix and the process noise gain matrix As shown below:

[0061]

[0062] Where T is the data sampling interval.

[0063] NCA model m j The state equation of (j∈{2,...,r}) is:

[0064]

[0065] in is the state transition matrix of the NCA model, is the process noise gain matrix of the NCA model, To obey zero mean covariance is Gaussian white noise.

[0066] State transition matrix and the process noise gain matrix As shown below:

[0067]

[0068] With respect to step 102 , the states and model probabilities of all the motion models are initialized, and the model transition probabilities are set.

[0069] In an embodiment of the present invention, algorithm initialization includes the following steps: calculating an estimated value of the initial state of the motion model at the second moment based on observation values at the first moment and the second moment; initializing the model probability of the motion model, where different model structures have the same probability and the probabilities of models with the same structure are equal; determining the model transition probability of the motion model at two adjacent moments based on the switching relationship between the motion models and the switching relationship within the maneuvering model with the same structure;

[0070] Specifically, this embodiment uses a two-point initialization method to initialize the state, and other initialization methods may also be used. The state estimate at the second moment is initialized using the observations z1 and z2 at the first two moments.

[0071] NCV Model m 1 The state estimate and the corresponding covariance at time k = 2 are:

[0072]

[0073]

[0074] Where z k (1) represents the position observation in the x direction at time k, z k (2) represents the position observation in the y direction at time k, and T represents the data sampling interval. is the state estimation covariance of the NCV model at time k = 2, and R is the observation noise covariance.

[0075] NCA model m j The state estimation and corresponding covariance of (j∈{2,…,r}) when k=2 are:

[0076]

[0077] Where β∈{1,2,…,n}, 0 m×n Represents an m-row, n-column matrix consisting of 0s.

[0078] Furthermore, the model probability at the second moment is initialized. The model probability is the probability that the model is the true motion model at the current moment. Under the initial conditions, different model structures have the same probability, and the probabilities of models with the same structure are equal. In this embodiment, the initial model probability of the NCV model is:

[0079]

[0080] The initial model probability of NCA is:

[0081]

[0082] Furthermore, the maneuver parameter jump is modeled as a switch between models with the same structure. It can be seen that when setting the model transition probability of the motion model, not only the switch between different structural models but also the switch between models with the same structure should be considered. That is, given the k-1 time model m i Effective, model m at time k j The probability of effectiveness is expressed as:

[0083]

[0084] in, Represents model m i Effective at time k-1, Represents model m j Takes effect at time k.

[0085] Model transition probability π j|i The settings are as follows:

[0086]

[0087] In the first formula, j=i means that the model remains unchanged, that is, P max represents the probability that the model remains unchanged. The second formula represents the switch from non-maneuvering mode to maneuvering mode. The third formula represents the switch from maneuvering mode to non-maneuvering mode. The last formula represents the switch between isomorphic maneuvering models and is used to describe the jump of maneuvering parameters.

[0088] With respect to step 104 , at each filtering moment, the mixed probability of the motion model is calculated according to the model probability and the preset model transition probability.

[0089] In the embodiment of the present invention, the prediction probability and the mixed probability of the model can be calculated based on the model probability and the preset model transition probability.

[0090] The calculation of the mixing probability is:

[0091]

[0092] in, For the model The predicted probability of:

[0093]

[0094] For step 106, partial state interaction is performed on all the motion models according to the mixing probability and the partition parameters, and the interaction results are input into the filter corresponding to the motion model. Based on the observation data obtained at the current filtering moment, the conditional state estimation and likelihood function of each motion model are obtained by filtering.

[0095] State interaction refers to the state estimates from each model at time k-1 being weighted and combined as the input to the model matching filter at time k. Different from the traditional full-state interaction strategy, this method proposes a partial state interaction strategy, which only interacts with the common state components in the model set. In this embodiment, the NCV model state includes the target position and velocity, and the NCA model state includes the target position, velocity and acceleration. Only the target position and velocity state components in the model set interact. The acceleration input to the corresponding filter of each NCA model is a weighted combination of the initial acceleration and the model's own estimate at the previous moment.

[0096] Interacting the position and velocity state components results in:

[0097]

[0098] where j∈{1,2,…,r}, The k-1 moment comes from the model State Estimation The position and velocity components in The k-1 moment comes from the model State Estimation Covariance The position and velocity covariance matrices in .

[0099] For the NCV model m 1 In terms of state interaction results and Directly as the input of its corresponding filter, recorded as:

[0100]

[0101] For the NCA model m j (j∈{2,...,r}), we also need to calculate the acceleration component in the input:

[0102]

[0103] where j∈{2,...,r}, The k-1 moment comes from the model State Estimation The acceleration component in The k-1 moment comes from the model State Estimation Covariance The acceleration covariance matrix in η j-1 is the mean value of the acceleration partition corresponding to the β=j-1th NCA model, that is,

[0104] Σ j-1 is the covariance matrix of the acceleration partition corresponding to the β=j-1th NCA model:

[0105]

[0106] The input of the NCA model matched filter is obtained as follows:

[0107]

[0108] In an embodiment of the present invention, the interaction result is input into the filter corresponding to the motion model, and based on the observation data obtained at the current filtering moment, the conditional state estimation and likelihood function of each motion model are obtained by filtering, including the following steps: establishing an observation model for filtering processing; inputting the first interaction result into the filter corresponding to the non-maneuvering model, and inputting the second interaction result into the filter corresponding to each maneuvering model respectively; updating the filtering result according to the observation value obtained at the current filtering moment and the observation model to obtain the state estimation value, estimated covariance and likelihood function of the target at the current filtering moment.

[0109] Specifically, an observation model is first established. This implementation scheme considers a linear observation model:

[0110]

[0111] Where z k represents the observation received at time k, For model m j The observation matrix, w k is the observation noise with zero mean and covariance R.

[0112] NCV Model m 1 The observation matrix is:

[0113]

[0114] NCA model m j The observation matrix of (j∈{2,...,r}) is:

[0115]

[0116] Then, based on the observation value obtained at the current filtering moment, Kalman filtering is performed to update the state estimate and the corresponding covariance

[0117]

[0118]

[0119]

[0120] Where, is the innovation covariance; is the Kalman gain.

[0121] And calculate the likelihood function

[0122]

[0123] With respect to step 108, the model probability of each motion model is updated according to the likelihood function, and the model conditional state estimation is estimated and fused according to the updated model probability to obtain the final state estimation value of tracking the maneuvering target at the current moment.

[0124] In the embodiment of the present invention, after obtaining the likelihood function of the model, the model probability of all models in the model set can be updated:

[0125]

[0126] where c k is the normalization constant

[0127] Finally, the final state estimate and the corresponding covariance at time k are calculated. Since the state estimates from the NCV model and the NCA model have unequal dimensions, the estimate from the NCV model is padded with zero mean and zero variance. State Estimation and covariance After dimension alignment, they are recorded as and The model conditional estimate is weighted by the model probability, and the final estimate is:

[0128]

[0129] In summary, the above steps can achieve tracking of a maneuvering target at a filtering moment; after initialization of step 100 and step 102, repeating the above steps 104-108 can achieve continuous tracking of a maneuvering target.

[0130] The feasibility of the above method is demonstrated below using a simulation example.

[0131] A maneuvering target with initial state x0 = [8000m, 8000m, 25m / s, 200m / s] T Start moving. True acceleration a=[a x ,a y ] T The sequences are shown in Table 1, where a x and a y are the acceleration components in the x and y directions. The sensor located at the Cartesian origin reports the Cartesian position of the target, and the observations are subject to covariance R k =1250I2m 2 The sampling period is T = 1s, and the lower bounds of the acceleration components in the x-direction and y-direction are assumed to be -40m / s. 2 , the upper limit is 40m / s 2 1000 Monte Carlo simulations are performed, and the probability that the model remains unchanged is set to P max =0.98.

[0132] Table 1

[0133]

[0134] The average Euclidean error of the position of PSIMM and traditional IMM algorithms under 1000 Monte Carlo simulations is as follows: Figure 2 As shown in the figure, the average Euclidean error of the PSIMM and traditional IMM algorithms under 1000 Monte Carlo simulations is as follows: Figure 3 As shown in the figure, it can be seen that the PSIMM algorithm proposed in the present invention achieves significantly better performance when the target acceleration jumps, and improves the tracking accuracy of the maneuvering target.

[0135] Please refer to Figure 4 The embodiment of the present invention provides a multi-model maneuvering target tracking device based on partial state interaction, the device comprising:

[0136] A partitioning module 400 is configured to establish a motion model set based on the motion pattern of the maneuvering target to be tracked, partition the maneuvering parameter space of the maneuvering target into regions, determine partition parameters for each partition and a state equation for each motion model; wherein the motion model includes a non-maneuvering model and a maneuvering model, and a maneuvering model with the same structure is used to describe a maneuvering pattern of the maneuvering target;

[0137] Initialization module 402, used to initialize the states and model probabilities of all the motion models, and set the model transition probabilities;

[0138] A calculation module 404 is configured to calculate the mixing probability of the motion model at each filtering moment based on the model probability and a preset model transition probability;

[0139] A filtering module 406 is configured to perform partial state interaction on all the motion models according to the mixing probability and the partition parameter, input the interaction result into the filter corresponding to the motion model, and filter the result based on the observation data obtained at the current filtering time to obtain a conditional state estimate and likelihood function for each motion model;

[0140] Fusion module 408 is configured to update the model probability of each motion model based on the likelihood function, and to perform an estimated fusion of the model conditional state estimates based on the updated model probabilities to obtain a final state estimate for tracking the maneuvering target at the current moment. In this embodiment of the present invention, the motion model includes a non-maneuvering model and a maneuvering model, and the same-structure maneuvering model is used to describe a maneuvering pattern of the maneuvering target.

[0141] In an embodiment of the present invention, the partitioning module 400 is specifically configured to perform the following operations when partitioning the maneuvering parameter space of the maneuvering target and determining the partition parameters of each partition and the state equation of each motion model: determining the maneuvering parameter space according to the maneuvering range of the maneuvering target to be tracked; partitioning the maneuvering parameter space corresponding to the maneuvering model according to the number of maneuvering models with the same structure, and calculating the partition parameters of each partition; establishing the state equation of the non-maneuvering model according to the position and velocity of the maneuvering target; and establishing the state equation of the maneuvering model with the same structure according to the position, velocity, and maneuvering parameters of the maneuvering target.

[0142] In an embodiment of the present invention, the initialization module 402 is specifically used to perform the following operations when performing state initialization and model probability initialization for all the motion models and setting the model transition probability: calculating the initial state estimation value of the motion model at the second moment based on the observation values at the first moment and the second moment; initializing the model probability of the motion model, different model structures have the same probability, and the probabilities of models with the same structure are equal; determining the model transition probability of the motion model at two adjacent moments based on the switching relationship between the motion models and the switching relationship within the maneuvering model with the same structure.

[0143] In the embodiment of the present invention, when performing partial state interaction on all the motion models according to the mixing probability and the partition parameter, the filtering module 406 is specifically configured to perform the following operations:

[0144] Interaction processing is performed on the non-motorized model according to the common state component estimation of the non-motorized model and the motorized model, to obtain a first interaction result of the non-motorized model. and

[0145]

[0146] in, The k-1 moment comes from the model State Estimation Position and velocity components in ; The k-1 moment comes from the model State Estimation Covariance The position and velocity covariance matrices;

[0147] Interactively processing the maneuvering model according to the common state component estimate and the unique state component estimate of the maneuvering model to obtain a second interactive result for each maneuvering model and

[0148]

[0149] in, The k-1 moment comes from the model State Estimation The acceleration component in The k-1 moment comes from the model State Estimation Covariance The acceleration covariance matrix in .

[0150] In an embodiment of the present invention, the filtering module 406 inputs the interaction result into the filter corresponding to the motion model, and filters to obtain the conditional state estimation and likelihood function of each motion model based on the observation data obtained at the current filtering moment, including: establishing an observation model for filtering processing; inputting the first interaction result into the filter corresponding to the non-maneuvering model, and inputting the second interaction result into the filter corresponding to each maneuvering model respectively; updating the filtering result according to the observation value obtained at the current filtering moment and the observation model to obtain the state estimation value, estimated covariance and likelihood function of the target at the current filtering moment.

[0151] In an embodiment of the present invention, when the fusion module 408 performs the estimation fusion of the model conditional state estimation according to the updated model probability to obtain the final state estimation value of tracking the maneuvering target at the current moment, it is specifically used to perform the following operations: fill and align the model conditional state estimation value and the estimated covariance at the current filtering moment according to zero mean and zero variance respectively; perform weighted estimation on the processed model conditional state estimation value and the estimated covariance according to the updated model probability, and calculate the final state estimation value.

[0152] It should be noted that the multi-model maneuvering target tracking device based on partial state interaction provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the multi-model maneuvering target tracking device based on partial state interaction provided in the above embodiment and the multi-model maneuvering target tracking method based on partial state interaction are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0153] The embodiment of the present application also provides a computer device, please refer to Figure 5 The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the multi-model maneuvering target tracking method based on partial state interaction provided by the above-mentioned method embodiments.

[0154] An embodiment of the present application also provides a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the multi-model maneuvering target tracking method based on partial state interaction provided by the above-mentioned method embodiments.

[0155] An embodiment of the present application also provides a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium, and the processor executes the computer program, so that the computer device executes the multi-model maneuvering target tracking method based on partial state interaction described in any of the above embodiments.

[0156] For the convenience of description, the above systems or devices are described as being divided into various modules or units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0157] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0158] Finally, it should be noted that, in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0159] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A multi-model maneuvering target tracking method based on partial state interaction, characterized in that: The method comprises: Step 100: Establishing a motion model set based on the motion pattern of the maneuvering target to be tracked, partitioning the maneuvering parameter space of the maneuvering target into regions, and determining partition parameters of each partition and a state equation for each motion model; wherein the motion model includes a non-maneuvering model and a maneuvering model, and the same-structure maneuvering model is used to describe a maneuvering pattern of the maneuvering target; Step 102, initializing the states and model probabilities of all the motion models, and setting the model transition probabilities; Step 104, at each filtering moment, calculating the mixing probability of the motion model according to the model probability and the preset model transition probability; Step 106: Perform partial state interaction on all the motion models according to the mixing probability and the partition parameter, input the interaction result into the filter corresponding to the motion model, and filter based on the observation data obtained at the current filtering time to obtain the conditional state estimate and likelihood function of each motion model; Step 108 , based on the likelihood function, the model probability of each motion model is updated, and the model conditional state estimation is estimated and fused based on the updated model probability to obtain the final state estimation value of tracking the maneuvering target at the current moment.

2. The method according to claim 1, wherein The step of dividing the maneuvering parameter space of the maneuvering target into regions and determining the partition parameters of each partition and the state equation of each motion model includes: Determine the maneuvering parameter space according to the maneuvering range of the maneuvering target to be tracked; Dividing the corresponding maneuver parameter space according to the number of maneuver models with the same structure, and calculating the partition parameters of each partition; Establishing a state equation of the non-maneuvering model according to the position and velocity of the maneuvering target; The state equation of the same-structure maneuvering model is established according to the position, speed and maneuvering parameters of the maneuvering target.

3. The method according to claim 1, wherein The step of initializing the states and model probabilities of all the motion models and setting the model transition probabilities includes: Calculating an estimated value of the initial state of the motion model at the second moment based on the observation values at the first moment and the second moment; Initializing the model probability of the motion model, wherein different model structures have the same probability, and the probabilities of models with the same structure are equal; The model transition probabilities of the motion models at two adjacent moments are determined according to the switching relationship between the motion models and the switching relationship within the maneuvering model with the same structure.

4. The method according to claim 1, wherein The performing partial state interaction on all the motion models according to the mixing probability and the partition parameter includes: Interaction processing is performed on the non-motorized model according to the common state component estimation of the non-motorized model and the motorized model, to obtain a first interaction result of the non-motorized model. and in, The k-1 moment comes from the model State Estimation Position and velocity components in ; The k-1 moment comes from the model State Estimation Covariance The position and velocity covariance matrices; Interactively processing the maneuvering model according to the common state component estimate and the unique state component estimate of the maneuvering model to obtain a second interactive result of the maneuvering model and in, The k-1 moment comes from the model State Estimation The acceleration component in The k-1 moment comes from the model State Estimation Covariance The acceleration covariance matrix in .

5. The method according to claim 4, wherein The interaction result is input into the filter corresponding to the motion model, and based on the observation data obtained at the current filtering moment, the conditional state estimation and likelihood function of each motion model are obtained by filtering, including: Establishing an observation model for filtering processing; Inputting the first interaction result into the filter corresponding to the non-maneuvering model, and inputting the second interaction result into the filter corresponding to each maneuvering model respectively; The filtering result is updated according to the obtained observation value at the current filtering moment and the observation model to obtain the state estimation value, estimated covariance and likelihood function of the target at the current filtering moment.

6. The method according to claim 1, wherein The step of estimating and fusing the model conditional state estimate according to the updated model probability to obtain a final state estimate for tracking the maneuvering target at the current moment includes: Performing padding and alignment processing on the model conditional state estimation value and the estimated covariance at the current filtering moment according to zero mean and zero variance respectively; The processed model conditional state estimate and the estimated covariance are weightedly estimated according to the updated model probability to calculate the final state estimate.

7. A multi-model maneuvering target tracking device based on partial state interaction, characterized in that: The device comprises: a partitioning module, configured to establish a motion model set based on the motion pattern of the maneuvering target to be tracked, partition the maneuvering parameter space of the maneuvering target into regions, determine partition parameters of each partition and a state equation of each motion model; wherein the motion model includes a non-maneuvering model and a maneuvering model, and the same-structure maneuvering model is used to describe a maneuvering pattern of the maneuvering target; An initialization module, configured to initialize the states and model probabilities of all the motion models, and set the model transition probabilities; A calculation module, configured to calculate the mixing probability of the motion model at each filtering moment according to the model probability and a preset model transition probability; a filtering module, configured to perform partial state interaction on all the motion models according to the mixing probability and the partition parameter, input the interaction result into a filter corresponding to the motion model, and filter the conditional state estimate and likelihood function of each motion model based on the observation data obtained at the current filtering moment; A fusion module is used to update the model probability of each motion model according to the likelihood function, and to estimate and fuse the model conditional state estimate according to the updated model probability to obtain the final state estimate of tracking the maneuvering target at the current moment.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the steps of any one of the methods described in claims 1-6.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images