Switching-Restricted Multi-Model Maneuvering Target Tracking Method and Device for Decision Aiding

Abstract

The embodiments of the present invention relate to the technical field of maneuvering target tracking, and particularly to a decision - assisted multi - model maneuvering target tracking method and device with limited switching. The method includes: after initializing the filters for each model sequence, filtering the measurement of the target at the current moment by using each filter respectively to obtain the state estimation result and likelihood function of each filter at the current moment; then, detecting whether the motion mode of the target has switched according to the likelihood function of each filter at the current moment. If it is detected that the mode has switched, the state estimation result of the corresponding filter at the current moment is output. If it is detected that the mode has not switched, the state estimation result after weighted fusion of the state estimation results of all filters at the current moment is output. This solution can ensure good estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of maneuvering target tracking, and in particular, to a decision-assisted multi-model maneuvering target tracking method and device with limited switching. Background Art

[0002] Target tracking refers to the technology of estimating the target motion state (such as position, velocity, etc.) from the noisy data observed by sensors. In actual tracking applications, the target usually changes its own motion law for some purpose, such as turning, accelerating, climbing, etc. of an aircraft. This unpredictable change in the target motion characteristics, that is, the maneuver of the target, will cause the tracking accuracy of the filter to decrease.

[0003] In the related art, there are mainly two tracking methods for maneuvering targets. One is the decision-based maneuvering target tracking method, which first detects whether the target maneuvers, and then selects the filter of the corresponding model for state estimation according to the detection result. Since the decision is irreversible and the possible decision errors are not considered in the estimation process, a wrong decision will affect the entire recursive estimation process, resulting in a lower estimation accuracy. The other is the multi-model filtering tracking method, which runs multiple filters of different models in parallel at each moment, and obtains the final estimation result by weighting the state estimations of all filters. This method does not detect whether the target maneuvers, and directly uses the method of setting transition probabilities, which is a prior modeling method, to describe the switching of the target motion mode, resulting in the multi-model filtering tracking method being unable to achieve satisfactory estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching. In summary, the existing tracking methods for maneuvering targets cannot ensure good estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching.

[0004] Therefore, there is an urgent need for a new method for maneuvering target tracking. Summary of the Invention

[0005] In order to solve the problem that the original maneuvering target tracking method cannot achieve good estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching, the embodiments of the present invention provide a decision-assisted multi-model maneuvering target tracking method and device with limited switching.

[0006] In a first aspect, the embodiments of the present invention provide a decision-assisted multi-model maneuvering target tracking method with limited switching, including:

[0007] Calculate the state estimation result at the second initial moment based on the measurements of the target at two initial moments, and use this state estimation result as the state estimation result at the previous moment of the filter corresponding to each model sequence at the current moment; the state estimation result includes state estimation and state estimation covariance; there is a one-to-one correspondence between the model sequence and the filter;

[0008] Calculate the mixing probability and the mixed state estimation result of each filter at the current moment according to the state estimation result of each filter at the previous moment and the corresponding model sequence probability; the mixing probability is used to characterize the weight of the state estimation result of each filter at the previous moment input to this filter at the current moment;

[0009] Respectively use the mixed state estimation result of each filter to filter the measurement of the target at the current moment to obtain the state estimation result and the likelihood function of each filter at the current moment;

[0010] Update the model sequence probability of each filter at the previous moment to obtain the model sequence probability of each filter at the current moment;

[0011] According to the likelihood function of each filter at the current moment, select the decision model sequence with the maximum joint probability among all the model sequences that have switched modes at the current moment;

[0012] Judge whether a mode switch occurs according to the maximum joint probability. If so, output the state estimation result of the filter corresponding to the decision model sequence at the current moment. If not, output the weighted fusion state estimation result obtained according to the state estimation results and model sequence probabilities of all filters at the current moment;

[0013] Take the current moment as the new previous moment, and jump to execute the calculation of the mixing probability and the mixed state estimation result of each filter at the current moment.

[0014] Preferably, before calculating the mixing probability and the mixed state estimation result of each filter at the current moment, it further includes:

[0015] Determine the number of pre-constructed motion models;

[0016] Determine the number of model sequences after permutation and combination according to the number of the motion models;

[0017] Determine the model sequence probability of the filter corresponding to each model sequence at the previous moment of the current moment according to the number of the model sequences;

[0018] Preferably, the calculating the mixing probability and the mixed state estimation result of each filter at the current moment according to the state estimation result of each filter at the previous moment and the corresponding model sequence probability includes:

[0019] According to the switching restriction rule that the target will not switch the motion model twice continuously within three consecutive moments, set the first transition probability that the model sequence with model switching at the previous moment maintains the model unchanged at the current moment and the second transition probability of the model sequence without model switching at the previous moment.

[0020] Calculate the mixing probability of each filter at the current moment according to the first transition probability and the second transition probability.

[0021] According to the mixing probability of each filter at the current moment and the state estimation result of each filter at the previous moment, calculate the mixing state estimation result of each filter at the current moment respectively.

[0022] Preferably, using the mixing state estimation results of each filter to filter the measurement of the target at the current moment respectively to obtain the state estimation result and the likelihood function of each filter at the current moment, including:

[0023] For each filter corresponding to a model sequence, perform:

[0024] Determine the modeling equation of the motion model with a later time in the model sequence at the previous moment.

[0025] Calculate the one-step state prediction and the one-step state prediction covariance according to the mixing state estimation result of the filter and the modeling equation.

[0026] Calculate the measurement prediction value according to the one-step state prediction.

[0027] Calculate the innovation according to the measurement of the target at the current moment and the measurement prediction value.

[0028] Calculate the innovation covariance according to the one-step state prediction covariance.

[0029] Calculate the likelihood function according to the innovation and the innovation covariance.

[0030] Calculate the filtering gain according to the one-step state prediction covariance and the innovation covariance.

[0031] Calculate the state estimation of the filter at the current moment according to the filtering gain, the one-step state prediction and the innovation.

[0032] Calculate the state estimation covariance of the filter at the current moment according to the one-step state prediction covariance, the filtering gain and the innovation covariance.

[0033] Preferably, according to the likelihood function of each filter at the current moment, select the decision model sequence with the maximum joint probability among all the model sequences with mode switching at the current moment, including:

[0034] According to the likelihood function of each filter at the current moment, calculate the effective probability of each model sequence at the current moment respectively;

[0035] According to the effective probability, calculate the joint probability that each model sequence is true while other model sequences are false respectively;

[0036] According to the joint probability of each model sequence, select the decision model sequence with the maximum joint probability among all model sequences with mode switching occurring at the current moment.

[0037] Preferably, the judging whether mode switching occurs according to the maximum joint probability includes:

[0038] Determine the preset mode switching threshold parameter;

[0039] Judge whether the maximum joint probability is greater than the mode switching threshold parameter.

[0040] Preferably, the calculation method of the state estimation result after weighted fusion is:

[0041] Calculate the state estimation after weighted fusion according to the state estimation and model sequence probability of each filter at the current moment;

[0042] Calculate the state estimation covariance after weighted fusion according to the state estimation, state estimation covariance and model sequence probability of each filter at the current moment.

[0043] In a second aspect, an embodiment of the present invention further provides a decision-assisted switching-restricted multi-model maneuvering target tracking device, including:

[0044] An initialization unit, configured to calculate the state estimation result at the second initial moment based on the measurements of the target at two initial moments, and use this state estimation result as the state estimation result at the previous moment of each filter corresponding to each model sequence; the state estimation result includes state estimation and state estimation covariance; there is a one-to-one correspondence between the model sequence and the filter;

[0045] A hybrid calculation unit, configured to calculate the hybrid probability and hybrid state estimation result of each filter at the current moment according to the state estimation result and corresponding model sequence probability of each filter at the previous moment; the hybrid probability is used to characterize the weight of the state estimation result of each filter at the previous moment input to this filter at the current moment;

[0046] A filtering unit, configured to respectively filter the measurement of the target at the current moment by using the hybrid state estimation result of each filter to obtain the state estimation result and likelihood function of each filter at the current moment;

[0047] An update unit for updating the model sequence probability of each filter at the previous moment to obtain the model sequence probability of each filter at the current moment;

[0048] A decision-making unit for selecting, according to the likelihood function of each filter at the current moment, a decision model sequence with the maximum joint probability among all the model sequences that have undergone mode switching at the current moment;

[0049] An output unit for judging whether a mode switch occurs according to the maximum joint probability. If so, outputting the state estimation result of the filter corresponding to the decision model sequence at the current moment. If not, outputting the weighted fusion state estimation result obtained according to the state estimation results and model sequence probabilities of all filters at the current moment;

[0050] A loop unit for taking the current moment as the new previous moment and jumping to execute the calculation of the mixed probability and mixed state estimation result of each filter at the current moment.

[0051] In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the method described in any embodiment of this specification is implemented.

[0052] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed on a computer, the computer is made to execute the method described in any embodiment of this specification.

[0053] An embodiment of the present invention provides a decision-assisted switching-limited multi-model maneuvering target tracking method and device. After initializing the filters of each model sequence, each filter is used to filter the measurement of the target at the current moment to obtain the state estimation result and likelihood function of each filter at the current moment. Then, according to the likelihood function of each filter at the current moment, it is detected whether the motion mode of the target has switched. If it is detected that the mode has switched, the state estimation result of the corresponding filter at the current moment is output. If it is detected that the mode has not switched, the weighted fusion state estimation result of the state estimation results of all filters at the current moment is output. Therefore, this solution can ensure better estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching. Description of the Drawings

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0055] Figure 1 It is a flowchart of a decision - assisted switching - restricted multi - model maneuvering target tracking method provided by an embodiment of the present invention;

[0056] Figure 2 It is a comparison chart of position errors of three methods provided by an embodiment of the present invention;

[0057] Figure 3 It is a comparison chart of speed errors of three methods provided by an embodiment of the present invention;

[0058] Figure 4 It is a hardware architecture diagram of an electronic device provided by an embodiment of the present invention;

[0059] Figure 5 It is a structural diagram of a decision - assisted switching - restricted multi - model maneuvering target tracking device provided by an embodiment of the present invention. Detailed implementation manners

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0061] As mentioned above, in the related art, one is a maneuvering target tracking method based on decision - making. First, it detects whether the target maneuvers, and then selects a filter of the corresponding model for state estimation according to the detection result. Since the decision is irreversible and possible decision errors are not considered in the estimation process, wrong decisions will affect the entire recursive estimation process, thereby resulting in low estimation accuracy. The other is a multi - model filtering tracking method. This method runs multiple filters of different models in parallel at each moment and obtains the final estimation result by weighting the state estimations of all filters. This method does not detect whether the target maneuvers and directly uses the method of setting transition probabilities, which is a method of prior - property modeling, to describe the switching of the target motion mode, resulting in the multi - model filtering tracking method being unable to achieve satisfactory estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching.

[0062] To solve the above technical problems, the inventor can consider selecting one of the state estimation results of the filters corresponding to the decision sequence model and the state estimation results after multi-model sequence fusion weighting as the output according to the detection result of whether the mode is switched, so as to effectively reduce the peak error at the moment of mode switching; in addition, it can be considered to detect whether the mode is switched according to the state estimation results of each filter after each filter performs state estimation. In this way, the detection result only affects the output at the current moment and does not affect the filtering and output at the next moment, so as to ensure the estimation accuracy of each filter and further ensure that better estimation accuracy can be obtained both during the period when the target motion mode remains unchanged and at the moment of mode switching.

[0063] The following describes the specific implementation manners of the above concepts.

[0064] Please refer to Figure 1 , an embodiment of the present invention provides a decision-assisted switching-constrained multi-model maneuvering target tracking method, and the method includes:

[0065] Step 100: Calculate the state estimation result at the second initial moment based on the measurements of the target at two initial moments, and use this state estimation result as the state estimation result at the previous moment of the current moment for the filters corresponding to each model sequence; the state estimation result includes state estimation and state estimation covariance; there is a one-to-one correspondence between the model sequence and the filter;

[0066] Step 102: Calculate the mixed probability and the mixed state estimation result of each filter at the current moment according to the state estimation result of each filter at the previous moment and the corresponding model sequence probability; the mixed probability is used to characterize the weight of the state estimation result of each filter at the previous moment input to this filter at the current moment;

[0067] Step 104: Respectively use the mixed state estimation result of each filter to filter the measurement of the target at the current moment to obtain the state estimation result and the likelihood function of each filter at the current moment;

[0068] Step 106: Update the model sequence probability of each filter at the previous moment to obtain the model sequence probability of each filter at the current moment;

[0069] Step 108: According to the likelihood function of each filter at the current moment, select the decision model sequence with the maximum joint probability among all the model sequences that have mode switching at the current moment;

[0070] Step 110: Determine whether a mode switch occurs according to the maximum joint probability. If so, output the state estimation result of the filter corresponding to the decision model sequence at the current moment. If not, output the weighted fusion state estimation result obtained based on the state estimation results of all filters at the current moment and the model sequence probability.

[0071] Step 112: Take the current moment as the new previous moment, and jump to execute the calculation of the mixing probability and the mixed state estimation result of each filter at the current moment.

[0072] In the embodiment of the present invention, after initializing the filters of each model sequence, each filter is respectively used to filter the measurement of the target at the current moment to obtain the state estimation result and the likelihood function of each filter at the current moment; then, according to the likelihood function of each filter at the current moment, it is detected whether the motion mode of the target has switched. If it is detected that a mode switch has occurred, the state estimation result of the corresponding filter at the current moment is output. If it is detected that no mode switch has occurred, the state estimation result after weighted fusion of the state estimation results of all filters at the current moment is output. Therefore, this solution can ensure good estimation accuracy both during the period when the target motion mode remains unchanged and at the moment of mode switching.

[0073] The following describes Figure 1 the execution manner of each step shown.

[0074] Regarding step 100:

[0075] In this step, the measurements z 1 , z 2 of the radar at two initial moments are used to calculate the state estimation and the state estimation covariance at the second initial moment, i.e., when k = 2:

[0076]

[0077]

[0078] In the formula, is the state estimation at the second moment, z k = [x k y k T represents the measurement received by the radar at the kth moment, z k (1) indicates that the position of the target in the x direction in the measurement at the kth moment is x k , z k (2) indicates that the position of the target in the y direction in the measurement at the kth moment is y k , so represents the velocity of the target in the x direction at the second moment, ​Denote the velocity of the target in the y direction at the second moment as \(v_y\), and \(T\) represents the radar data sampling interval. \(P\) 2|2 is the state estimation covariance at the second moment, \(r\) ij represents the element in the \(i\)-th row and \(j\)-th column of the observation noise covariance. The observation noise covariance is a known quantity, and its expression formula is:

[0079]

[0080] In some embodiments, before step 102, it may further include:

[0081] Determine the number of pre-constructed motion models;

[0082] Determine the number of model sequences after permutation and combination according to the number of motion models;

[0083] Determine the model sequence probability of each filter corresponding to the model sequence at the previous moment of the current moment according to the number of model sequences.

[0084] In the embodiments of the present invention, the required motion models can be pre-constructed. The construction steps of the motion models are described below.

[0085] First, the motion equation is modeled as:

[0086] \(X\) k \(=\) \(F\) k-1 \((m\) j )\(X\) k-1 \(+\) \(G\) k-1 \((m\) j )\(u\) k-1 \((m\) j )\(+\) \(\Gamma\) k-1 \((m\) j )\(v\) k-1 \((m\) j )

[0087] In the formula, \(X\) k represents the state vector composed of the position \(x\) k in the \(x\) direction, the position \(y\) k in the \(y\) direction, the velocity \(\dot{x}\) in the \(x\) direction, and the velocity \(\dot{y}\) in the \(y\) direction at the \(k\)-th moment, that is, \(F\) k-1 \((m\) j ) is the state transition matrix of the model \(m\) j at the \((k - 1)\)-th moment, \(G\) k-1 \((m\) j ) is the input control matrix of the model \(m\) j at the \((k - 1)\)-th moment, \(u\) k-1 \((m\) j ) is the input of the model \(m\) j at the \((k - 1)\)-th moment, \(\Gamma\) k-1(m j ) is the noise gain matrix of the model m at time k-1, v j . k-1 (m j ) is the zero-mean white Gaussian process noise of the model m at time k-1, v j . The process noise covariance matrix of v k-1 (m j ) is Q k-1 (m j ).

[0088] Among them, in some embodiments, the input control matrix G k-1 (m j ) and the noise gain matrix Γ k-1 (m j ) can be expressed as:

[0089]

[0090] And the state transition matrix F k-1 (m j ) and the input G k-1 (m j ) can be determined as follows:

[0091] Case 1, when the motion model is a uniform motion model,

[0092]

[0093] Case 2, when the motion model is a coordinated turn model,

[0094]

[0095] In the formula, ω is the turning angular velocity.

[0096] Case 3, when the motion model is a uniformly accelerated motion model,

[0097]

[0098] Among them, a x is the acceleration in the x direction, and a y is the acceleration in the y direction.

[0099] It should be noted that the above three cases listed are the three most commonly used models in maneuvering target tracking, but only as examples, and there can be other model cases in actual scenarios.

[0100] Next, the number of pre-constructed motion models can be determined. The motion model is m j , where j∈{1,2,…,r} represents the serial number of the motion model, and r is the number of motion models.

[0101] In this embodiment, the time length of the model sequence is 2, that is, the motion models are arranged and combined in pairs.

[0102] Specifically, the model sequence with a length of 2 at time k-1 is (j≠i):

[0103]

[0104] in, It means that the motion model l′∈{1,2,…,r} is effective at time k′, represents all model sequences that have mode switching from time k-2 to time k-1, represents all model sequences without mode switching from time k-2 to time k-1. All model sequences with mode switching and all model sequences without mode switching can be uniformly recorded as When the model sequence time is not specified, the model sequence can be simply represented by i n j 2-n .

[0105] Since the number of motion models is r, the number of model sequences is r 2 , so when k=2 is initialized, the model sequence probability of the filter corresponding to each model sequence is The k=2 moment is taken as the previous moment of the current moment, that is, the model sequence probability of each filter is

[0106] It should be noted that, in order to facilitate subsequent cycles, the subscript k is used to represent the current moment, and k-1 represents the moment before the current moment.

[0107] In addition, the calculated state estimate and state estimate covariance at the second initial moment are used as the state estimate result of the filter corresponding to each model sequence at the previous moment of the current moment, that is,

[0108] In this embodiment, the initialization of each filter can be completed at this point.

[0109] Regarding step 102:

[0110] In some implementations, step 102 may include the following steps S1-S3:

[0111] Step S1, according to the switching restriction rule that the target will not switch the motion model twice in three consecutive moments, set a first transition probability that the model sequence that had model switching at the previous moment maintains the model unchanged at the current moment and a second transition probability that the model sequence that did not have model switching at the previous moment;

[0112] Step S2, calculate the mixing probability of each filter at the current moment according to the first transition probability and the second transition probability;

[0113] Step S3, calculate the mixed state estimation results of each filter at the current moment respectively according to the mixing probability of each filter at the current moment and the state estimation results of each filter at the previous moment.

[0114] In this embodiment, first calculate the mixing probability of each filter according to Steps S1 - S2, and the following is an explanation of the specific calculation process.

[0115] In Step S1, according to the switching - limited rule that the target will not switch the motion model twice continuously within three consecutive moments, so for the model sequence i that switched the model at k - 1 1 j 1 (j ≠ i), it will maintain the motion model j unchanged at k. Therefore, the first transition probability And for the model sequence i that did not switch the model at k - 1 2 j 0 has the possibility of switching at k, and set the second transition probability

[0116]

[0117] where, p max is a pre - set parameter, and its range is [0, 1]. In the embodiment of the present invention, its value range is [0.99, 1).

[0118] In Step S2, when n = 1, the mixing probability of each model sequence that did not switch the model at k can be calculated. Then, the expression of the mixing probability of the filter corresponding to each model sequence where the mode did not switch is:

[0119]

[0120] where, is the first normalization constant, and its expression is:

[0121]

[0122] In the formula, is the model sequence probability of the sequence i 1 j 1 at the previous moment.

[0123] When n = 2 (j ≠ i), the mixing probability of each model sequence that switched the model at k can be calculated. Then, the expression of the mixing probability of the filter corresponding to each model sequence where the mode switched is:

[0124]

[0125] Among them, is the first normalization constant, and the expression is:

[0126]

[0127] In the formula, is the model sequence probability of the corresponding filter at the previous moment.

[0128] Therefore, according to steps S1 - S2, the mixing probability of each filter can be calculated.

[0129] It should be noted that in the existing multi - model filtering and tracking method, the value range of p max is [0.95, 0.99). In practical applications, the larger the value of p max , the higher the estimation accuracy of using the state estimation after multi - model sequence weighted fusion as the output during the period when the motion mode remains unchanged.

[0130] Since in the existing technology, the switching of the target motion mode is directly described by setting the transition probability, in order to balance the estimation accuracy during the period when the motion mode remains unchanged and at the moment of mode switching, the value of p max cannot be too large.

[0131] In the embodiment of the present invention, in the subsequent step 110, it will be determined whether mode switching occurs according to the maximum joint probability. Therefore, in this step, the value of p max will not affect the estimation accuracy at the moment of mode switching. Thus, in order to improve the estimation accuracy during the period when the motion mode remains unchanged, the value range of p max in this embodiment can be [0.99, 1).

[0132] In step S3, according to the mixing probability of each filter at the current moment and the state estimation result of each filter at the previous moment, the mixing state estimation result of each filter at the current moment can be calculated respectively.

[0133] Specifically, when n = 1, the calculation formulas for the mixing state estimation and the mixing state estimation covariance of the filter corresponding to each model sequence where mode switching does not occur are respectively:

[0134]

[0135]

[0136] In the formula, represents the state estimation of the model sequence i 1 j 1 at the previous moment, Represents the model sequence i at the previous moment 1 j 1 The estimated input is fed into the model sequence i at the current moment 0 j 2 The weights in the corresponding filter Represents the model sequence i at the previous moment 1 j 1 The state estimation covariance

[0137] When n = 2 (j ≠ i), the calculation formulas for the mixed state estimation and the mixed state estimation covariance of the filter corresponding to each model sequence with mode switching are respectively as follows:

[0138]

[0139]

[0140] In the formula Represents the state estimation of the model sequence i at the previous moment 2 j 0 The state estimation Represents the state estimation covariance of the model sequence i at the previous moment 2 j 0 The state estimation covariance

[0141] Regarding step 104:

[0142] In some embodiments, step 104 may include the following steps B1 - B9:

[0143] Step B1, for the filter corresponding to each model sequence, perform: determine the modeling equation of the motion model with a later time in the model sequence at the previous moment.

[0144] In this step, for the filter corresponding to each model sequence, perform: determine the modeling equation of the motion model with a later time in the model sequence.

[0145] Step B2, calculate the one - step state prediction and the one - step state prediction covariance according to the mixed state estimation result of the filter and the modeling equation.

[0146] Calculate the one - step state prediction of the state estimation according to the following formula

[0147]

[0148] In the formula, F k-1 (m j ) is the state transition matrix of the motion model m at time k - 1 j The state transition matrix, G k-1 (m j) is the motion model m at time k-1 j of the input control matrix, u k-1 (m j ) is the input of the motion model m at time k-1 j , is the hybrid state estimate of this filter at the current time.

[0149] Calculate the one-step state prediction covariance according to the following formula

[0150]

[0151] where Γ k-1 (m j ) is the noise gain matrix of model m at time k-1 j , Q k-1 (m j ) is the process noise covariance matrix of model m at time k-1 j , is the hybrid state estimate covariance of this filter at the current time.

[0152] Step B3, calculate the measurement prediction value according to the one-step state prediction.

[0153] Calculate the measurement prediction value according to the following formula

[0154]

[0155] where H k is the observation matrix.

[0156] Step B4, calculate the innovation according to the measurement of the target at the current time and the measurement prediction value.

[0157] Calculate the innovation according to the following formula

[0158]

[0159] where z k is the measurement at the current time.

[0160] Step B5, calculate the innovation covariance according to the one-step state prediction covariance.

[0161] Calculate the innovation covariance according to the following formula

[0162]

[0163] where H k is the observation matrix and R is the observation noise covariance.

[0164] Step B6, calculate the likelihood function according to the innovation and the innovation covariance.

[0165] Calculate the likelihood function according to the following formula

[0166]

[0167] In the formula, denotes z k obeys a Gaussian distribution with a mean of and a covariance of

[0168] Step B7, calculate the filtering gain according to the one-step state prediction covariance and the innovation covariance.

[0169] Calculate the filtering gain according to the following formula

[0170]

[0171] Step B8, calculate the state estimate of the filter at the current moment according to the filtering gain, the one-step state prediction, and the innovation.

[0172] Calculate the state estimate of the filter at the current moment according to the following formula

[0173]

[0174] Step B9, calculate the state estimate covariance of the filter at the current moment according to the one-step state prediction covariance, the filtering gain, and the innovation covariance.

[0175] Calculate the state estimate covariance of the filter at the current moment according to the following formula

[0176]

[0177] In the embodiments of the present invention, the state estimate, the state estimate covariance, and the likelihood function of each filter at the current moment can be calculated.

[0178] For step 106:

[0179] In the embodiments of the present invention, calculate the model sequence probability of each filter at the current moment according to the following formula

[0180]

[0181] Wherein, is the model sequence i n-1 j 3-n ​The corresponding likelihood function is the first normalization constant corresponding to each filter in step 102, c is the second normalization constant, and the calculation formula is:

[0182]

[0183] In the formula, represents the probability of the model sequence at the previous moment, represents the transition probability.

[0184] Regarding step 108:

[0185] In some embodiments, step 108 may include the following steps H1 - H3:

[0186] Step H1, calculate the effective probability of each model sequence at the current moment according to the likelihood function of each filter at the current moment.

[0187] In this step, the model sequence i n-1 j 3-n is represented by the symbol s l Each model sequence has a corresponding subscript l ∈ {1, 2,..., r 2}, and calculate the effective probability of each model sequence respectively:

[0188]

[0189] Step H2, calculate the joint probability that each model sequence is true while the other model sequences are false according to the effective probability.

[0190] Calculate the joint probability that each model sequence is true while the remaining sequences are false according to the following formula

[0191]

[0192] where C is the third normalization constant, and the calculation formula is:

[0193]

[0194] Step H3, select the decision model sequence with the maximum joint probability among all the model sequences with mode switching at the current moment according to the joint probability of each model sequence.

[0195] In this step, according to the joint probability of each model sequence calculated in step H2, select the model sequence with the maximum joint probability among all the model sequences with mode switching at the current moment, and determine this model sequence as the decision model sequence at the current moment.

[0196] Specifically, the sequence of decision models is selected according to the following formula:

[0197]

[0198] For step 110:

[0199] In some embodiments, the step of "judging whether a mode switch occurs according to the maximum joint probability" may include:

[0200] Determine the pre-set mode switch threshold parameter;

[0201] Judge whether the maximum joint probability is greater than the mode switch threshold parameter.

[0202] In the embodiments of the present invention, it is necessary to pre-determine the set mode switch threshold parameter γ, and the value range of the mode switch threshold parameter γ can be [0.8, 1).

[0203] Then compare the maximum joint probability determined in step 108 with the mode switch threshold parameter γ to judge whether the following conditions are met:

[0204]

[0205] If it is satisfied, it is judged that a mode switch occurs at the current moment of the target, and then output the state estimation and state estimation covariance of the filter corresponding to the decision model sequence at the current moment;

[0206] If it is not satisfied, it is judged that no mode switch occurs at the current moment of the target, and then output the weighted fusion state estimation result obtained according to the state estimation results and model sequence probabilities of all filters at the current moment.

[0207] In some embodiments, the calculation method of the weighted fusion state estimation result is:

[0208] Calculate the weighted fusion state estimation according to the state estimation and model sequence probability of each filter at the current moment;

[0209] Calculate the weighted fusion state estimation covariance according to the state estimation, state estimation covariance and model sequence probability of each filter at the current moment.

[0210] Specifically, the weighted fusion state estimation and state estimation covariance can be calculated according to the following formula:

[0211]

[0212]

[0213] It should be noted that the state estimation result output in this step is the algorithm output result at the current moment. The state estimation and state estimation covariance of each filter at the current moment are both the results calculated in step 104. Therefore, in the embodiments of the present invention, the decision made based on the detection result will not affect the recursive filtering process of each filter.

[0214] Regarding step 112:

[0215] For example, after the initialization in step 100, the state estimation and state estimation covariance with k = 2 are obtained. After steps 102 - 110, the state estimation and state estimation covariance of each filter when k = 3 are calculated, and the final estimation result when k = 3 is output by determining whether a mode switch occurs.

[0216] In this step, taking the moment of k = 3 as the new previous moment and returning to step 102, the state estimation and state estimation covariance of each filter when k = 4 can be calculated through steps 102 - 110, and the final estimation result when k = 4 is output by determining whether a mode switch occurs. By repeating this process, the maneuvering tracking of the target can be achieved.

[0217] To verify the effectiveness of this solution, a simulation experiment was carried out according to the embodiments of the present invention. At the same time, a comparative experiment was conducted between the proposed decision - assisted switching - constrained multiple - model maneuvering target tracking method (DA - SC - IMM) in this solution and the multiple - model maneuvering target tracking filtering method with limited model switching times (SC - IMM) and the interactive multiple - model filtering method (IMM).

[0218] The experimental simulation is based on the following linear observation equation:

[0219] z k =H k X k +w k

[0220] where w k is the observation noise, and its covariance is R.

[0221]

[0222] The simulation scenario is as follows: The maneuvering target first moves at a constant speed for 15 s with an initial state of X 1 =[73km, 30km, - 270m / s, - 130m / s] T , then turns at a turning rate of 13° / s for 5 s, moves on the new flight path for 10 s after turning, then turns at a turning rate of 9° / s for 10 s for a 90° turn, and finally the target moves in a straight line at a constant speed. The total simulation duration is 50 s, and the sampling interval T = 1 s. The process noise covariance is Qk (m j ) = 0.01Im 2 / s 4 , and the measurement noise covariance is R = 64Im 2 . The models used are the constant velocity motion model and the coordinated turn model. Assuming that the turn rate of the motion is known, the models are numbered. The constant velocity motion model is denoted as Model 1, the coordinated turn model with a turn rate of 13° / s is Model 2, and the coordinated turn model with a turn rate of 9° / s is Model 3. 2000 Monte Carlo simulations are carried out.

[0223] The position average Euclidean errors of the DA-SC-IMM, SC-IMM, and IMM algorithms under 2000 Monte Carlo simulations are as Figure 2 shown; the velocity average Euclidean errors of the DA-SC-IMM, SC-IMM, and IMM algorithms under 2000 Monte Carlo simulations are as Figure 3 shown. It can be seen from the figure that DA-SC-IMM achieves similar performance to the SC-IMM algorithm during the period when the mode remains unchanged, is superior to the IMM algorithm, and the peak error at the mode switching moment is smaller than that of the other two algorithms, realizing an overall improvement in the tracking performance of maneuvering targets.

[0224] Therefore, in the embodiments of the present invention, according to the detection result of whether the mode switches, one of the state estimation results of the filters corresponding to the decision sequence model and the state estimation results after multi-model sequence fusion and weighting is selected as the output, thereby effectively reducing the peak error at the mode switching moment. In addition, the detection in the present invention is performed after the filtering estimation, and the detection result only affects the output of the algorithm at the current moment, avoiding introducing the decision error into the filtering estimation recursive process, thereby ensuring the estimation accuracy of each filter. Therefore, the present invention can not only obtain ideal tracking performance during the period when the motion mode remains unchanged, but also achieve a lower peak error at the mode switching moment.

[0225] As Figure 4 , Figure 5 shown, the embodiments of the present invention provide a decision-assisted switched constrained multi-model maneuvering target tracking device. The device embodiments can be implemented by software, or by hardware or a combination of software and hardware. From the hardware level, as Figure 4 shown, it is a hardware architecture diagram of an electronic device where a decision-assisted switched constrained multi-model maneuvering target tracking device provided by the embodiments of the present invention is located. In addition to Figure 4 the processor, memory, network interface, and non-volatile memory shown, the electronic device where the device is located in the embodiments usually may also include other hardware, such as a forwarding chip responsible for processing packets, etc. Taking software implementation as an example, as Figure 5As shown, as a device in a logical sense, it is formed by reading the corresponding computer program in the non-volatile memory into the memory by the CPU of the electronic device where it is located and running it.

[0226] As Figure 5 shown, a switching-restricted multi-model maneuvering target tracking device for decision assistance provided in this embodiment includes:

[0227] An initialization unit 501, configured to calculate a state estimation result at a second initial moment based on measurements of the target at two initial moments, and use this state estimation result as the state estimation result at the previous moment of each filter corresponding to each model sequence; the state estimation result includes a state estimation and a state estimation covariance; there is a one-to-one correspondence between the model sequence and the filter;

[0228] A hybrid calculation unit 502, configured to calculate the hybrid probability and the hybrid state estimation result of each filter at the current moment according to the state estimation result of each filter at the previous moment and the corresponding model sequence probability; the hybrid probability is used to represent the weight of the state estimation results of each filter at the previous moment input to this filter at the current moment;

[0229] A filtering unit 503, configured to respectively use the hybrid state estimation result of each filter to filter the measurement of the target at the current moment to obtain the state estimation result and the likelihood function of each filter at the current moment;

[0230] An update unit 504, configured to update the model sequence probability of each filter at the previous moment to obtain the model sequence probability of each filter at the current moment;

[0231] A decision unit 505, configured to select, according to the likelihood function of each filter at the current moment, a decision model sequence with the maximum joint probability among all the model sequences that have mode switching at the current moment;

[0232] An output unit 506, configured to judge whether mode switching occurs according to the maximum joint probability. If so, output the state estimation result of the filter corresponding to the decision model sequence at the current moment. If not, output the weighted fusion state estimation result obtained according to the state estimation results and the model sequence probabilities of all filters at the current moment;

[0233] A loop unit 507, configured to use the current moment as the new previous moment, and jump to execute the calculation of the hybrid probability and the hybrid state estimation result of each filter at the current moment.

[0234] In an embodiment of the present invention, in the initialization unit 501, the following operations may also be performed:

[0235] Determine the number of pre-constructed motion models;

[0236] Determine the number of model sequences after permutation and combination according to the number of motion models;

[0237] Determine the model sequence probability of the previous moment of each filter corresponding to each model sequence according to the number of model sequences.

[0238] In an embodiment of the present invention, the hybrid computing unit 502 is configured to perform the following operations:

[0239] According to the switching restriction rule that the target will not switch the motion model twice continuously within three consecutive moments, set the first transition probability that the model sequence that switched the model at the previous moment maintains the model unchanged at the current moment and the second transition probability of the model sequence that did not switch the model at the previous moment.

[0240] Calculate the hybrid probability of each filter at the current moment according to the first transition probability and the second transition probability.

[0241] Calculate the hybrid state estimation results of each filter at the current moment respectively according to the hybrid probability of each filter at the current moment and the state estimation results of each filter at the previous moment.

[0242] In an embodiment of the present invention, the filtering unit 503 is configured to perform the following operations

[0243] For each filter corresponding to each model sequence, perform:

[0244] Determine the modeling equation of the motion model with a later time in the model sequence at the previous moment;

[0245] Calculate the one-step state prediction and the one-step state prediction covariance according to the hybrid state estimation result of the filter and the modeling equation;

[0246] Calculate the measurement prediction value according to the one-step state prediction;

[0247] Calculate the innovation according to the true measurement of the target at the current moment and the measurement prediction value;

[0248] Calculate the innovation covariance according to the one-step state prediction covariance;

[0249] Calculate the likelihood function according to the innovation and the innovation covariance;

[0250] Calculate the filtering gain according to the one-step state prediction covariance and the innovation covariance;

[0251] Calculate the state estimation of the filter at the current moment according to the filtering gain, the one-step state prediction and the innovation;

[0252] Calculate the state estimation covariance of the filter at the current moment according to the one-step state prediction covariance, the filtering gain, and the innovation covariance.

[0253] In an embodiment of the present invention, the decision-making unit 505 is configured to perform the following operations:

[0254] Calculate the effective probability of each model sequence at the current moment according to the likelihood function of each filter at the current moment;

[0255] Calculate the joint probability that each model sequence is true while the other model sequences are false according to the effective probability;

[0256] Select the decision model sequence with the maximum joint probability from all the model sequences with mode switching at the current moment according to the joint probability of each model sequence.

[0257] In an embodiment of the present invention, when the output unit 506 executes the determination of whether mode switching occurs according to the maximum joint probability, it is configured to perform the following operations:

[0258] Determine the preset mode switching threshold parameter;

[0259] Judge whether the maximum joint probability is greater than the mode switching threshold parameter.

[0260] In an embodiment of the present invention, in the output unit 506, the calculation method of the weighted fusion state estimation result is as follows:

[0261] Calculate the weighted fusion state estimation according to the state estimation and the model sequence probability of each filter at the current moment;

[0262] Calculate the weighted fusion state estimation covariance according to the state estimation, the state estimation covariance, and the model sequence probability of each filter at the current moment.

[0263] It can be understood that the structure schematically shown in the embodiments of the present invention does not constitute a specific limitation on a decision-assisted switching-restricted multi-model maneuvering target tracking device. In other embodiments of the present invention, a target tracking device in the distance square domain may include more or fewer components than those shown in the figure, or combine certain components, or split certain components, or have different component arrangements. The components shown in the figure can be implemented in hardware, software, or a combination of software and hardware.

[0264] Regarding the information interaction, execution process, etc. between the modules in the above device, since they are based on the same concept as the method embodiments of the present invention, the specific content can be referred to the description in the method embodiments of the present invention and will not be elaborated here.

[0265] An embodiment of the present invention further provides an electronic device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, a switching-limited multi-model maneuvering target tracking method for decision assistance in any embodiment of the present invention is implemented.

[0266] An embodiment of the present invention further provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the processor is enabled to execute a switching-limited multi-model maneuvering target tracking method for decision assistance in any embodiment of the present invention.

[0267] Specifically, a system or device equipped with a storage medium can be provided. A software program code for implementing the functions in any one of the above embodiments is stored on the storage medium, and the computer (or CPU or MPU) of the system or device reads and executes the program code stored in the storage medium.

[0268] In this case, the program code read from the storage medium itself can implement the functions in any one of the above embodiments. Therefore, the program code and the storage medium storing the program code constitute a part of the present invention.

[0269] Embodiments of the storage medium for providing the program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Optionally, the program code can be downloaded from a server computer via a communication network.

[0270] In addition, it should be clear that not only can the functions in any one of the above embodiments be implemented by executing the program code read by the computer, but also by causing an operating system or the like operating on the computer based on the instructions of the program code to complete part or all of the actual operations.

[0271] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in the expansion board inserted into the computer or into the memory provided in the expansion module connected to the computer. Subsequently, based on the instructions of the program code, the CPU or the like installed on the expansion board or the expansion module executes part and all of the actual operations, thereby implementing the functions in any one of the above embodiments.

[0272] It should be noted that in this article, relational terms such as first and second are only used 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 term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the said element.

[0273] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for tracking a maneuvering target with switching constraints for decision assistance, characterized in that, it includes: Calculating the state estimation result at the second initial moment based on the measurements of the target at two initial moments, and using this state estimation result as the state estimation result at the previous moment of each filter corresponding to each model sequence at the current moment; The state estimation result includes state estimation and state estimation covariance; There is a one-to-one correspondence between the model sequence and the filter; Calculating the mixing probability and the mixed state estimation result of each filter at the current moment according to the state estimation result and the corresponding model sequence probability of each filter at the previous moment; the mixing probability is used to characterize the weight of the state estimation results of each filter at the previous moment input into this filter at the current moment; Respectively using the mixed state estimation result of each filter to filter the measurement of the target at the current moment to obtain the state estimation result and the likelihood function of each filter at the current moment; Updating the model sequence probability of each filter at the previous moment to obtain the model sequence probability of each filter at the current moment; According to the likelihood function of each filter at the current moment, selecting the decision model sequence with the maximum joint probability among all the model sequences that have mode switching at the current moment; Judging whether there is a mode switching according to the maximum joint probability. If so, outputting the state estimation result of the filter corresponding to the decision model sequence at the current moment. If not, outputting the weighted fusion state estimation result obtained according to the state estimation results and the model sequence probabilities of all filters at the current moment; Taking the current moment as the new previous moment, and jumping to execute the calculation of the mixing probability and the mixed state estimation result of each filter at the current moment; Respectively using the mixed state estimation result of each filter to filter the measurement of the target at the current moment to obtain the state estimation result and the likelihood function of each filter at the current moment, including: For each filter corresponding to each model sequence, execute: Determining the modeling equation of the motion model that is later in time in this model sequence at the previous moment; Calculating the one-step state prediction and the one-step state prediction covariance according to the mixed state estimation result of this filter and the modeling equation; Calculating the measurement prediction value according to the one-step state prediction; Calculating the innovation according to the measurement of the target at the current moment and the measurement prediction value; Calculating the innovation covariance according to the one-step state prediction covariance; Calculating the likelihood function according to the innovation and the innovation covariance; Calculating the filtering gain according to the one-step state prediction covariance and the innovation covariance; Calculating the state estimation of this filter at the current moment according to the filtering gain, the one-step state prediction and the innovation; Calculating the state estimation covariance of this filter at the current moment according to the one-step state prediction covariance, the filtering gain and the innovation covariance.

2. The method according to claim 1, characterized in that, Before calculating the mixing probability and the mixed state estimation result of each filter at the current moment, it further includes: Determining the number of pre-constructed motion models; Determine the number of model sequences after permutation and combination according to the number of the motion models; Determine the model sequence probability of the previous moment of each filter corresponding to each model sequence at the current moment according to the number of the model sequences.

3. The method according to claim 1, characterized in that the calculating of the mixed probability and the mixed state estimation result of each filter at the current moment according to the state estimation result of each filter at the previous moment and the corresponding model sequence probability includes: According to the switching restricted rule that the target will not switch the motion model twice continuously within three consecutive moments, set the first transition probability that the model sequence with model switching at the previous moment keeps the model unchanged at the current moment and the second transition probability of the model sequence without model switching at the previous moment; Calculate the mixed probability of each filter at the current moment according to the first transition probability and the second transition probability; Calculate the mixed state estimation result of each filter at the current moment respectively according to the mixed probability of each filter at the current moment and the state estimation result of each filter at the previous moment.

4. The method according to claim 1, characterized in that the selecting of the decision model sequence with the maximum joint probability among all the model sequences with mode switching at the current moment according to the likelihood function of each filter at the current moment includes: Calculate the effective probability of each model sequence at the current moment respectively according to the likelihood function of each filter at the current moment; Calculate the joint probability that each model sequence is true while other model sequences are false according to the effective probability; Select the decision model sequence with the maximum joint probability among all the model sequences with mode switching at the current moment according to the joint probability of each model sequence.

5. The method according to claim 1, characterized in that the judging whether mode switching occurs according to the maximum joint probability includes: Determine the preset mode switching threshold parameter; Judge whether the maximum joint probability is greater than the mode switching threshold parameter.

6. The method according to claim 1, characterized in that the calculation method of the state estimation result after weighted fusion is: Calculate the state estimation after weighted fusion according to the state estimation and the model sequence probability of each filter at the current moment; Calculate the state estimation covariance after weighted fusion according to the state estimation, the state estimation covariance and the model sequence probability of each filter at the current moment.

7. A decision-assisted switching restricted multi-model maneuvering target tracking device for implementing the method according to any one of claims 1-6, characterized in that it includes: An initialization unit, configured to calculate the state estimation result of the second initial moment based on the measurements of the target at two initial moments, and use this state estimation result as the state estimation result of the previous moment of each filter corresponding to each model sequence at the current moment; The state estimation result includes state estimation and state estimation covariance; There is a one-to-one correspondence between the model sequences and the filters; A hybrid computing unit, configured to calculate the hybrid probability and the hybrid state estimation result of each filter at the current moment according to the state estimation result of each filter at the previous moment and the corresponding model sequence probability; the hybrid probability is used to represent the weight of the state estimation results of each filter at the previous moment input to this filter at the current moment; A filtering unit, configured to respectively use the hybrid state estimation result of each filter to filter the measurement of the target at the current moment, so as to obtain the state estimation result and the likelihood function of each filter at the current moment; An updating unit, configured to update the model sequence probability of each filter at the previous moment to obtain the model sequence probability of each filter at the current moment; A decision-making unit, configured to select, according to the likelihood function of each filter at the current moment, the decision model sequence with the maximum joint probability among all the model sequences that have undergone mode switching at the current moment; An output unit, configured to judge whether a mode switch occurs according to the maximum joint probability. If so, output the state estimation result of the filter corresponding to the decision model sequence at the current moment. If not, output the weighted fusion state estimation result obtained according to the state estimation results and the model sequence probabilities of all filters at the current moment; A loop unit, configured to use the current moment as the new previous moment, and jump to execute the calculation of the hybrid probability and the hybrid state estimation result of each filter at the current moment.

8. An electronic device, comprising a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the method according to any one of claims 1-6 is implemented.

9. A computer-readable storage medium, on which a computer program is stored. When the computer program is executed in a computer, the computer is made to execute the method according to any one of claims 1-6.

Images