Multi-model motion state tracking method based on adaptive interaction

Abstract

The invention discloses a multi-model motion state tracking method based on adaptive interaction, and the method comprises the steps: introducing an adaptive Markov chain to improve the model switching precision of a target in different states in order to solve a problem that the model switching precision of an interactive multi-model during prediction is low and cannot meet engineering demands; a joint probability data association algorithm is introduced to realize high-precision tracking of multiple targets, and then a covariance matrix of the targets is adjusted according to the calculation probability of the joint data association algorithm, so that the association stability of the targets is improved; according to the uncertainty change of the motion state of the target, different states of the target are represented by adopting an interactive multi-model, and on the basis, a Kalman filtering algorithm is combined, so that switching of the target in different motion states and smooth filtering of the target after automatic stable tracking are realized. The method can realize more accurate and more stable tracking of multiple targets, has good adaptability and estimation precision, and has practical significance for tracking complex maneuvering targets.

Description

Technical Field

[0001] The present invention belongs to the field of target tracking, and in particular relates to a technology for multi-model motion state tracking based on adaptive interaction. Background Art

[0002] Target tracking is a tracking technology that uses the distance, pitch and azimuth of a moving target and combines algorithms to estimate the target's motion state. Target tracking is divided into two modes: single model tracking mode and multi-model tracking mode. The single model tracking mode mainly uses a motion model to describe the maneuvering state of the target, but a single model is often difficult to accurately describe the maneuvering state of the target, and may even cause the target to be lost due to large estimation errors in maneuvering; the multi-model just overcomes the defects of the single model. The multi-model mainly uses multiple motion models to match the changing maneuvering state of the target, so multi-model target tracking is widely used in actual engineering applications.

[0003] In complex environments, measurement information may come from the correct measurement information of the target, or from the erroneous measurement of clutter, interference targets or false targets. Therefore, it is very easy to cause erroneous association of measurement data during association. The solution to this phenomenon is mainly to eliminate the uncertainty of measurement through data association technology. Data association technology mainly includes nearest neighbor data association (NNDA), probabilistic data association (PDA), joint probabilistic data association (JPDA) and multiple hypothesis tracking (MHT) and other related algorithms. An important problem that needs to be solved in joint probabilistic data association is how to design an effective tracking gate to find the target with a lower computational cost. In a strong maneuvering environment, since the motion model cannot meet the diverse changes of the motion state, and in the interactive multimodality, the Markov transition probability matrix is ​​pre-set and fixed, so that the model cannot be switched according to the data characteristics, it is very easy to cause the filter algorithm to diverge when filtering the point data.

[0004] Therefore, a better method is urgently needed. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a multi-model motion state tracking method based on adaptive interaction, which is used to solve the erroneous association of multiple target tracking and the real-time correction of large fluctuations in prediction models, thereby improving the accuracy and stability of multi-target tracking.

[0006] In a first aspect, an embodiment of the present invention provides a multi-model motion state tracking method based on adaptive interaction, characterized by comprising:

[0007] S1: obtaining measurement data of multiple targets and establishing a motion model set of the targets;

[0008] S2: Obtain the motion state of the target, and according to the motion state of the target, obtain the state transfer matrix and measurement matrix corresponding to each motion model in the motion model set of the target, and then set the Markov transfer matrix probability according to the prior knowledge;

[0009] S3: Filter out the state transfer matrix and the measurement matrix corresponding to the model probability that meets the motion track of the current target in the motion model set of the target, predict the motion track of the target, and obtain the motion track value of the target in the next frame;

[0010] S4: input the motion track value of the target in the next frame, and perform difference calculation with the measurement information of the next frame to determine whether it enters the probability threshold, perform association probability calculation on the measurement information that enters the probability threshold, and correct the value of the covariance matrix according to the association probability;

[0011] S5: according to the initial model probability and the Markov state transfer matrix, inputting the measurement information after the associated probability calculation and the value of the corrected covariance matrix into the interactive multi-model for interaction;

[0012] S6: Performing parallel filtering through a Kalman filter, inputting the associated probability into the modified covariance matrix, and filtering and tracking different motion states of the target, calculating the likelihood function value of each model probability according to each model probability, and then calculating the corresponding model probability value according to the likelihood function value of each model;

[0013] S7: Calculate a probability adjustment coefficient containing probability information according to the corresponding model probability value, obtain an adjusted Markov state transfer matrix probability, and update the model probability value again;

[0014] S8: Obtaining a filtered value of the measurement information after the interaction according to the updated model probability value;

[0015] The measurement data includes the target's distance, azimuth, pitch and SNR characteristic information.

[0016] Preferably, the motion model set includes a uniform linear motion model, a constant acceleration motion model and an adaptive current statistical motion model.

[0017] Preferably, S2 includes:

[0018] S21: setting the initial probability of each sub-model in the initial model probability according to prior knowledge;

[0019] S22: according to prior knowledge, setting a Markov transition probability matrix between the initial probabilities of each sub-model in the initial model probability;

[0020] The values ​​on the main diagonal of the Markov transition probability matrix should be greater than a preset value and should all be set to equal fixed data.

[0021] Preferably, S4 includes:

[0022] S41: Confirming the matrix, when the measurement data falls into the overlapping area of ​​different target-related gates, the corresponding relationship between the measurement data and the target is established through the Mahalanobis distance, so as to represent the relationship between the effective measurement and each target tracking gate;

[0023] S42: Joint events: splitting the confirmation matrix according to two assumptions and two principles to obtain an interconnection matrix of interconnected events, confirming the number of the joint events, and obtaining a measurement interconnection indication and a target detection indication;

[0024] S43: association probability and conditional probability, calculating the probability and normal distribution function of each measurement being associated with its possible source target according to the measurement interconnection indication and the target detection indication;

[0025] Preferably, S5 includes:

[0026] S51: Model interaction input, assuming that the state estimate corresponding to the target motion model at time k-1 is And the covariance matrix at this moment is P i (k-1|k-1), then the state vector of the target motion model at time k-1 is interactively calculated and its covariance matrix is ​​P oj (k-1|k-1);

[0027] S52: Filter calculation: The Kalman filter algorithm is used to calculate the probability of the current model. The interactive input value and the associated probability value of the joint data probability algorithm are used as input to perform parallel filtering on each sub-model in the model set to obtain the state estimation value and state covariance matrix of each sub-model at time k.

[0028] S53: Model probability update: Use the likelihood function to update the model probability m j (k);

[0029] S54: Comprehensive information output: Based on the model probability, the estimation results of each filter are weighted and combined to obtain the final state estimation and covariance estimation.

[0030] The method for tracking multi-model motion states based on adaptive interaction according to claim 5, wherein S53 comprises:

[0031] S531: Calculate the corresponding model probability change rate according to the model probability at different times;

[0032] S532: Setting the adjustment coefficient, and calculating and updating the Markov transition probability.

[0033] In a second aspect, an embodiment of the present invention further provides a device for a multi-model motion state tracking method based on adaptive interaction, characterized in that the device comprises:

[0034] A target information acquisition module, which acquires information of the target under different motion states;

[0035] Information processing module, which processes and filters the acquired and calculated information;

[0036] The data calculation and processing module calculates and processes various information and data.

[0037] In a third aspect, an embodiment of the present invention further provides a system for a multi-model motion state tracking method based on adaptive interaction, characterized in that the system is used to implement the method as described in any one of claims 1 to 7, and comprises:

[0038] Radar, used to track the motion of multiple targets

[0039] The sensor is used to collect the motion information of the target.

[0040] In a fourth aspect, an embodiment of the present invention further provides a storage medium having computer program instructions stored thereon, wherein when the computer program instructions are executed by a processor, a method as described in any one of claims 1 to 9 is implemented.

[0041] In summary, the beneficial effects of the present invention are as follows:

[0042] 1. The present invention avoids the phenomenon of erroneous association when multiple targets meet, uses the joint probability data association algorithm to effectively associate the measurement data, and adjusts the covariance matrix of the target, thereby improving the accuracy of the track during the tracking process; stably obtains the trajectory of the target, and judges the movement trend of the target;

[0043] 2. The present invention establishes multiple motion state models to meet the diverse changes of the target as much as possible, obtains the trajectory of the target more stably, and judges the motion trend of the target to make the track close to the motion trajectory of the real target;

[0044] 3. The present invention dynamically and adaptively adjusts the Markov transition probability matrix between different models according to the different target motion characteristics, thereby improving the tracking performance of the radar. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are used to provide further understanding of the present invention and constitute a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0046] In the attached picture:

[0047] Figure 1 A schematic diagram of the process provided by the present invention;

[0048] Figure 2 The adaptive optimization flow chart provided by the present invention;

[0049] Figure 3 It is the target trajectory diagram of the present invention;

[0050] Figure 4 This is the target tracking diagram without Kalman filter combined with joint probabilistic data association algorithm;

[0051] Figure 5 This is a target tracking diagram using Kalman filtering combined with a joint probabilistic data association algorithm;

[0052] Figure 6 For Figure 5 Based on this, a target tracking diagram of the interactive multi-model with adaptively adjusted Markov transfer matrix is ​​added. DETAILED DESCRIPTION

[0053] 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 only part of the embodiments of the present invention, rather than all of 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.

[0054] Example 1

[0055] See also Figure 1 and Figure 2 The embodiment of the present invention provides a multi-model motion state tracking method based on adaptive interaction, characterized by comprising:

[0056] S1: obtaining measurement data of multiple targets and establishing a motion model set of the targets;

[0057] Acquire measurement data of multiple vehicles, the measurement data includes various parameters related to vehicle performance, status and environment, which can be measured and recorded by the vehicle's own sensors, external devices or vehicle systems; such as speed: measuring the speed of the vehicle; acceleration: measuring the acceleration of the vehicle; speed: measuring the speed of the vehicle engine; engine parameters: measuring and monitoring various parameters of the engine, such as oil temperature, coolant temperature, oil pressure, exhaust temperature, etc.; braking force and braking distance: measuring the force generated by the vehicle braking system and the distance required for the vehicle to stop; vehicle position: measuring the geographic location and navigation data of the vehicle, usually obtained through a positioning system; vehicle attitude: measuring the attitude parameters of the vehicle, such as pitch angle, roll angle and yaw angle; preferably, the measurement data mainly includes characteristic information of the target's distance, azimuth, pitch and SNR;

[0058] And according to the motion characteristic information of the vehicle, a motion model set of the vehicle is established;

[0059] Establish a motion model set of multiple vehicles m = [m1, m2, ..., m r ](m i represents the probability of a sub-model), and determines the number of models.

[0060] S2: Acquire the motion state of the target, and based on the motion state of the target, obtain the state transfer matrix and measurement matrix corresponding to each motion model in the motion model set of the target, and then set the initial model probability and the Markov transfer matrix probability according to prior knowledge; prior knowledge refers to the common sense and conventional technical means already possessed in this field, and since they are conventional technical means, they will not be elaborated on.

[0061] S3: Filter out the state transfer matrix and measurement matrix corresponding to the model probability that meets the motion track of the current target in the motion model set of the target, predict the motion track of the target, and obtain the motion track value of the target in the next frame; specifically as follows:

[0062] X j (k+1)=F j (k)X j (k)+Γ j (k)υ j (k)

[0063] Z(k)=H(k)X j (k)+w j (k)

[0064] Among them, X j(k+1) is the state vector at time k+1, Z(k) is the observation vector at time k, and F j (k) is the state transfer matrix corresponding to the model probability, H represents the observation matrix, and the rest are the measurement noise and process noise;

[0065] S4: Input the motion track value of the target in the next frame and perform difference calculation with the measurement information of the next frame to determine whether it enters the probability threshold, perform association probability calculation on the measurement information that enters the probability threshold, and correct the value of the covariance matrix according to the association probability; the measurement information is the meaningful information obtained after processing and interpreting the measurement data.

[0066] S5: according to the value of the initial model probability and the Markov state transfer matrix, inputting the measurement information after the associated probability calculation and the value of the corrected covariance matrix into the interactive multi-model for interaction;

[0067] S6: Performing parallel filtering through a Kalman filter, inputting the associated probability into the modified covariance matrix, and filtering and tracking different motion states of the target, calculating the likelihood function value of each model probability according to each model probability, and then calculating the corresponding model probability value according to the likelihood function value of each model;

[0068] S7: Calculate a probability adjustment coefficient containing probability information according to the corresponding model probability value, obtain an adjusted Markov state transfer matrix probability, and update the model probability value again;

[0069] S8: Obtaining a filtered value of the measurement information after the interaction according to the updated model probability value;

[0070] The method for tracking the motion states of multiple vehicles based on the adaptive interactive multi-model Kalman filter combined with the joint probabilistic data association algorithm as described in claim 1 is characterized in that the motion model set includes a uniform linear motion model, a constant acceleration motion model and an adaptive current statistical motion model.

[0071] The method for tracking the motion states of multiple vehicles based on the adaptive interactive multi-model Kalman filter combined with the joint probabilistic data association algorithm as claimed in claim 1, comprising:

[0072] S21: According to prior knowledge, the initial probability of each sub-model in the initial model probability is set:

[0073] m(i), i=1,2,...,n;

[0074] in,

[0075] S22: According to prior knowledge, a Markov transition probability matrix c between the initial probabilities of each sub-model in the initial model probability is set: in,

[0076] The values ​​on the main diagonal of the Markov transition probability matrix should be greater than a preset value and should all be set to equal fixed data.

[0077] And in general, the value of the main diagonal of the transfer matrix should be set to be greater than 0.8.

[0078] Preferably, S4 includes:

[0079] S41: Confirming the matrix, when the measurement data falls into the overlapping area of ​​different target-related gates, the corresponding relationship between the measurement data and the target is established through the Mahalanobis distance, so as to represent the relationship between the effective measurement and each target tracking gate;

[0080] The specific expressions are as follows:

[0081]

[0082] Among them, ω jt is a binary variable, ω jt =1 indicates that measurement j (j=1,2,...,m) enters the confirmation gate of target t (t=1,2,...,T), otherwise, no measurement enters the confirmation gate of the target; when all the column elements corresponding to Ω are 1, it is considered that each measurement may come from clutter or false alarm;

[0083] S42: Joint Event θ i (k) The confirmation matrix is ​​split according to two assumptions and two principles to obtain the interconnection matrix of the interconnection events, confirm the number of the joint events, and obtain the measurement interconnection indication τ i (θ i (k)) and target detection indication; δ t (θ i (k)), which is specifically expressed as follows:

[0084]

[0085] in, is the event in which measurement j in the ith joint event at time k originates from target t;

[0086]

[0087] Among them, τ i (θ i (k)) represents the measurement j at the joint event θ i(k) is associated with a real target, if τ i (θ i (k)) = 1 and t> 0, indicating that measurement j is in the joint event θ i (k) is associated with a real target t, otherwise, the measurement j does not originate from any real target but from clutter or false alarm;

[0088]

[0089] In the formula, δ t (θ i (k)) represents the joint event θ i (k) Whether there is a connection between the measurement and the target t (whether the target t is detected), if δ t (θ i (k)) = 1, indicating that there exists a measurement j such that There exists a measurement j interconnected with the target t, otherwise, there is no measurement interconnected with the target t;

[0090] S43: association probability β jt And the conditional probability P r {θ i (k)|Z k}, according to the measurement interconnection indication and the target detection indication, calculate the probability of each measurement being interconnected with its possible source target and the normal distribution function; specifically as follows:

[0091]

[0092] Among them, θ i (k) is the i-th joint event, L is the number of joint events;

[0093]

[0094]

[0095]

[0096] Among them, Φ(θ i (k)) is the joint event θ i The number of false measurements in (k), N tj [Z j (k)] is the measurement interconnected with target t that follows Gaussian distribution;

[0097] Preferably, S5 includes:

[0098] S51: Model interaction input, assuming that the state estimate corresponding to the target motion model at time k-1 is And the covariance matrix at this moment is Pi (k-1|k-1), then the state vector of the target motion model at time k-1 is interactively calculated and its covariance matrix is ​​P oj (k-1|k-1); the specific formula is as follows:

[0099]

[0100]

[0101] in: yes The interaction result of P oj (k-1|k-1) is P i (k-1|k-1) interaction result; m ij (k-1|k-1) is the mixing probability from model i to model j, c ij is the probability of conversion according to the motion model, that is, the transfer probability transfer matrix from model i to model j. The mixing probability can be obtained by the predicted probability c of the model. ij and model probability m i (k-1|k-1) means:

[0102]

[0103] S52: Filter calculation: The Kalman filter algorithm is used to calculate the probability of the current model. Each sub-model in the model set is filtered in parallel according to the interactive input value and the associated probability value of the joint data probability algorithm as input, and the state estimation value and state covariance matrix of each sub-model at time k are obtained respectively. The process based on the Kalman filter method is as follows:

[0104] One-step state prediction:

[0105]

[0106]

[0107] Kalman gain calculation:

[0108] K j (k) = P j (k|k-1)H T [HP j (k|k-1)H T +R] -1

[0109] The filter is updated as:

[0110]

[0111]

[0112] Among them: A j (k) is the innovation covariance;

[0113] A j (k)=H(k)P j (k|k-1)H(k) T +R(k)

[0114] S53: Model probability update: Use the likelihood function to update the model probability m j (k), the likelihood function of model j and the model probability formula are as follows:

[0115]

[0116] Where: a j For new breath;

[0117] a j =Z(k)-H(k)X j (k|k-1)

[0118] Model Probability:

[0119]

[0120] S54: Comprehensive information output: Based on the model probability, the estimation results of each filter are weighted and combined to obtain the final state estimation and covariance estimation, as follows:

[0121]

[0122]

[0123] Preferably, the S53 includes:

[0124] S531: Calculate the corresponding model probability change rate according to the model probability at different times, as follows:

[0125] Δm j (k) = m j (k)-m j (k-1)

[0126] S532: Setting the adjustment coefficient and calculating the updated Markov transition probability, as follows:

[0127]

[0128] Among them, ε v is the speed adjustment parameter, when 0<ε v When ε is less than 1, the rate of adjusting the coefficient will decrease.v When it is greater than 1, the adjustment rate will be faster, so the appropriate value can be selected according to actual needs;

[0129] c ij (k)' = ζ ij (k) c ij (k-1)

[0130] In summary, when Δm j When (k)>0, ζ ij (k)>1 and c ij When (k)' rises, it means that model j is closer to the real model and should be in the maintenance stage. By increasing the proportion of the model, the accuracy of track tracking is improved; otherwise, it is in the state of model switching, which improves the speed of rapid model switching.

[0131] Example 2

[0132] See also Figure 1 and Figure 2 The embodiment of the present invention provides a multi-model motion state tracking method based on adaptive interaction, characterized by comprising:

[0133] Step 1: Obtain measurement data of multiple vehicles, including characteristic information such as target distance, azimuth, pitch, and SNR, and establish a vehicle motion model set based on the vehicle motion characteristic information;

[0134] Step 2: According to the motion state of the vehicle, the state transfer matrix and measurement matrix corresponding to each motion model in the model set are obtained, and the probability of the initial model and the probability of the Markov transfer matrix are set according to prior knowledge;

[0135] Step 3: Select the state transfer matrix and observation matrix corresponding to the model probability value of the current track in the model set to predict the track, so as to obtain the track prediction value at the next moment;

[0136] Step 4: Use the predicted data information as input to perform difference calculation with the measurement information of the next frame to determine whether it enters the probability threshold, calculate the association probability of the measurement information that enters the probability threshold, and update the value of the modified covariance matrix according to the association probability;

[0137] Step 5: Based on the model probability value at the previous moment and the Markov state transfer matrix, the associated measurement information and the corrected covariance matrix are used as input to interact in the interactive multi-model;

[0138] Step 6: The Kalman filter performs parallel filtering to filter and track different motion states of the target, and calculates the corresponding model probability value according to the likelihood function value of each model;

[0139] Step 7: Calculate a probability adjustment coefficient containing probability information according to the model probability value, obtain the adjusted Markov state transition probability, and update the model probability value again;

[0140] Step 8: Obtain the filtered value of the measurement information after the interaction according to the updated model probability value.

[0141] In the problem of multi-target tracking, incorrect association and slow model switching often occur, resulting in poor tracking performance accuracy. The joint probability data association algorithm determines the track trajectory to which the measurement belongs according to the weight of each measurement. The IMM algorithm establishes multiple model sets to meet the diversity of targets and determine the target's motion model. The traditional Markov transition probability is a fixed value and cannot switch the model according to the data characteristics, which can easily lead to the failure to reflect the target's true motion trajectory. Therefore, the adaptive Markov transition probability can quickly switch the model according to the target's motion change characteristics, improve the target's tracking accuracy, and can more stably reflect the target's true track.

[0142] Example 3

[0143] See also Figure 1 and Figure 2 The embodiment of the present invention provides a multi-model motion state tracking method based on adaptive interaction, characterized by comprising:

[0144] Step 1: To solve the problem that the model switching accuracy of interactive multi-model is low during prediction and cannot meet engineering requirements, an adaptive Markov chain is introduced to improve the accuracy of model switching in different states of the target;

[0145] Step 2: Use the joint probability data association algorithm to achieve high-precision tracking of multiple targets, and then adjust the covariance matrix of the target according to the calculated probability of the joint data association algorithm to improve the association stability of the target;

[0146] Step 3: According to the uncertainty changes of the target's motion state, the interactive multi-model is used to represent the different states of the target. On this basis, the Kalman filter algorithm is combined to achieve the switching of the target in different motion states and the smoothing filtering after automatic and stable tracking of the target.

[0147] Example 4

[0148] See also Figures 3 to 6 An embodiment of the present invention provides a multi-model motion state tracking method based on adaptive interaction, and the effect of the present invention is further illustrated by the following simulation experiments.

[0149] In order to verify the accuracy of the method of the present invention, a simulation experiment is performed to prove it.

[0150] The experimental data are as follows:

[0151]

[0152]

[0153] The simulation results are analyzed as follows: Figures 3 to 6 ;

[0154] Figure 3 It is the target trajectory diagram of the present invention;

[0155] Figure 4 This is the target tracking diagram without Kalman filter combined with joint probabilistic data association algorithm;

[0156] Figure 5 This is a target tracking diagram using Kalman filtering combined with a joint probabilistic data association algorithm;

[0157] Figure 6 For Figure 5 Based on this, a target tracking diagram of an interactive multi-model with adaptively adjusted Markov transfer matrix is ​​added;

[0158] from Figure 4 It can be seen that due to the encounter of the targets, it is difficult to separate the target tracks by conventional association algorithms alone. Figure 5 It can be seen that the target can be accurately separated by using the method of the present invention, thus ensuring a higher tracking accuracy.

[0159] from Figure 4 It can be seen that when the UAV turns, the track is interrupted and the wrong association prediction occurs. Figure 6 It can be seen that the use of all the methods of the present invention can not only accurately separate the targets, but also ensure that the targets have high switching accuracy under different models, thereby reducing the erroneous association of tracks and improving the tracking accuracy of tracks, which is beneficial to engineering implementation. Figure 5 and Figure 6 The effectiveness of this method is also verified;

[0160] In summary, the simulation experiments have verified the correctness, effectiveness and reliability of the present invention.

[0161] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article or device.

[0162] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0163] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A multi-model motion state tracking method based on adaptive interaction, characterized in that: include: S1: obtaining measurement data of multiple targets and establishing a motion model set of the targets; S2: Obtain the motion state of the target, and according to the motion state of the target, obtain the state transfer matrix and measurement matrix corresponding to each motion model in the motion model set of the target, and then set the Markov transfer matrix probability according to the prior knowledge; S3: Filter out the state transfer matrix and the measurement matrix corresponding to the model probability that meets the motion track of the current target in the motion model set of the target, predict the motion track of the target, and obtain the motion track value of the target in the next frame; S4: input the motion track value of the target in the next frame, and perform difference calculation with the measurement information of the next frame to determine whether it enters the probability threshold, perform association probability calculation on the measurement information that enters the probability threshold, and correct the value of the covariance matrix according to the association probability; S5: according to the initial model probability and the Markov state transfer matrix, inputting the measurement information after the associated probability calculation and the value of the corrected covariance matrix into the interactive multi-model for interaction; S6: Performing parallel filtering through a Kalman filter, inputting the associated probability into the modified covariance matrix, and filtering and tracking different motion states of the target, calculating the likelihood function value of each model probability according to each model probability, and then calculating the corresponding model probability value according to the likelihood function value of each model; S7: Calculate a probability adjustment coefficient containing probability information according to the corresponding model probability value, obtain an adjusted Markov state transfer matrix probability, and update the model probability value again; S8: Obtaining a filtered value of the measurement information after the interaction according to the updated model probability value; The measurement data includes the target's distance, azimuth, pitch and SNR characteristic information.

2. The method for tracking multi-model motion states based on adaptive interaction according to claim 1, characterized in that: The motion model set includes a uniform linear motion model, a constant acceleration motion model and an adaptive current statistical motion model.

3. The method for tracking multi-model motion states based on adaptive interaction according to claim 1, wherein S2 comprises: S21: setting the initial probability of each sub-model in the initial model probability according to prior knowledge; S22: according to prior knowledge, setting a Markov transition probability matrix between the initial probabilities of each sub-model in the initial model probability; The values ​​on the main diagonal of the Markov transition probability matrix should be greater than a preset value and should all be set to equal fixed data.

4. The method for tracking multi-model motion states based on adaptive interaction according to claim 1, characterized in that: The S4 includes: S41: Confirming the matrix, when the measurement data falls into the overlapping area of ​​different target-related gates, the corresponding relationship between the measurement data and the target is established through the Mahalanobis distance, so as to represent the relationship between the effective measurement and each target tracking gate; S42: Joint events: splitting the confirmation matrix according to two assumptions and two principles to obtain an interconnection matrix of interconnected events, confirming the number of the joint events, and obtaining a measurement interconnection indication and a target detection indication; S43: Association probability and conditional probability: according to the measurement interconnection indication and the target detection indication, calculate the probability of each measurement being associated with its possible source target and the normal distribution function.

5. The method for tracking motion states of multiple models based on adaptive interaction according to claims 1 to 4, characterized in that: The S5 includes: S51: Model interaction input, assuming that the state estimation corresponding to the target motion model at time k-1 is And the covariance matrix at this moment is P i (k-1|k-1), then the state vector of the target motion model at time k-1 is interactively calculated and its covariance matrix is ​​P oj (k-1|k-1); S52: Filter calculation: The Kalman filter algorithm is used to calculate the probability of the current model. The interactive input value and the associated probability value of the joint data probability algorithm are used as input to perform parallel filtering on each sub-model in the model set to obtain the state estimation value and state covariance matrix of each sub-model at time k. S53: Model probability update: Use the likelihood function to update the model probability m j (k); S54: Comprehensive information output: Based on the model probability, the estimation results of each filter are weighted and combined to obtain the final state estimation and covariance estimation.

6. The method for tracking multi-model motion states based on adaptive interaction as claimed in claim 5, wherein S53 comprises: S531: Calculate the corresponding model probability change rate according to the model probability at different times; S532: Setting the adjustment coefficient, and calculating and updating the Markov transition probability.

7. A device for a multi-model motion state tracking method based on adaptive interaction, characterized in that: The device comprises: A target information acquisition module, which acquires information of the target under different motion states; Information processing module, which processes and filters the acquired and calculated information; The data calculation and processing module calculates and processes various information and data.

8. A system for tracking multi-model motion states based on adaptive interaction, characterized in that , used to implement the method as described in any one of claims 1-7, the system comprising: Radar, used to track the motion of multiple targets The sensor is used to collect the motion information of the target.

9. A storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images