A soft-constraint target tracking method and system based on genetic resampling

Abstract

This invention discloses a soft-constraint target tracking method and system based on genetic resampling, belonging to the field of target tracking technology. The method includes: constructing a target tracking model based on the target being tracked; generating multiple particles required for particle filtering based on the initial state value of the target being tracked; updating each particle over time based on the target tracking model to predict the state and weight of each particle; resampling each particle using a genetic algorithm; calculating the corrected likelihood function of the resampled particles based on the measurement equation; correcting the predicted weights based on the corrected likelihood function; and updating the predicted state based on the corrected weights to obtain the final estimated state of the target being tracked. This invention can integrate known nonlinear inequality soft constraints in the tracking of ground moving targets, improving the tracking accuracy of ground targets under such soft constraints.

Description

Technical Field

[0001] This invention relates to the field of target tracking technology, and in particular to a soft-constraint target tracking method and system based on genetic resampling. Background Technology

[0002] In the process of tracking ground targets, Kalman filters are often used to process the measurement results of sensors and the statistical characteristics of target motion in order to effectively track the target's position, velocity and other state information. They have wide applications in military, autonomous driving and traffic control.

[0003] However, in the application of Kalman filters, some prior known information is often ignored. For example, the motion of ground targets is often limited by road information, and their speed is often restricted. Aircraft at airports face airspace restrictions during takeoff and landing. The motion of almost all targets is constrained by factors such as dynamics or the physical environment. Incorporating this constraint information into the state estimation during target tracking can improve the accuracy of the target tracking system. Currently, a series of filtering methods that consider constraints usually treat these constraints as hard constraints, meaning that the target's state is strictly limited by the constraint conditions. However, in reality, although the constraint conditions provide additional prior information about the target, the target may not strictly conform to the constraint. For example, a car traveling in a speed-limited area may still exceed the speed limit. This kind of constraint that does not strictly limit the target's state is called a soft constraint. How to integrate prior known soft constraint information when tracking such ground moving targets to improve the accuracy of ground target tracking is a problem that urgently needs to be solved. Summary of the Invention

[0004] Therefore, the purpose of this invention is to provide a soft-constraint target tracking method and system based on genetic resampling, so as to integrate known nonlinear inequality soft constraints in the tracking of ground moving targets and improve the tracking accuracy of ground targets under such soft constraints.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A soft-constraint target tracking method based on genetic resampling includes:

[0007] A target tracking model is constructed based on the target being tracked; the target tracking model includes a state equation, a measurement equation, and a soft constraint model;

[0008] Generate multiple particles required for particle filtering based on the initial state value of the tracked target;

[0009] Based on the target tracking model, each particle is updated over time to predict its state and weight.

[0010] Resampling of each particle is performed using a genetic algorithm;

[0011] The corrected likelihood function of the resampled particles is calculated based on the measurement equation.

[0012] The predicted weights are adjusted based on the modified likelihood function;

[0013] The final estimated state of the tracked target is obtained by updating the predicted state based on the corrected weights.

[0014] Optionally, the expression for the state equation is as follows:

[0015] X k =F k X k-1 +w k-1

[0016] Among them, X k Let X be the state vector of the tracked target at time k. k-1 Let F be the state vector of the tracked target at time k-1. k Let w be the state transition matrix at time k. k-1 The noise in the 4D process at time k-1;

[0017] The expression for the measurement equation is as follows:

[0018] z k =H k X k +v k

[0019] Among them, z k Let H be the 2D measurement vector at time k. k Let v be the measurement matrix at time k. k The 2D Gaussian measurement noise at time k;

[0020] The expression for the soft constraint model is as follows:

[0021]

[0022]

[0023] Where Γ is a random variable, g1(X) k ) and g2(X k ) is an intermediate variable, R min R max These represent the upper and lower boundaries of the road, respectively.

[0024] Optionally, based on the target tracking model, each particle is updated over time to predict its state and weight, specifically including:

[0025] The state equation is used to update the state of each particle over time, and the state of each particle is predicted.

[0026] Based on the soft constraint model, each particle is divided into particles that satisfy the soft constraints and particles that do not satisfy the soft constraints.

[0027] The quasi-Newton method is used to optimize the particles that do not meet the soft constraint conditions, and the optimization results are obtained in the region that meets the soft constraint conditions.

[0028] The optimization results are used to replace the particles that do not meet the soft constraints, and the weights of each particle are predicted based on the likelihood function, which is calculated based on the measurement equation.

[0029] Optionally, a genetic algorithm is used to resample each particle, specifically including:

[0030] Select, crossover, and mutation operations are performed on the particles to obtain resampled particles.

[0031] Optionally, the corrected likelihood function of the resampled particles is calculated based on the measurement equation, specifically including:

[0032] The measurement vector of the resampled particle is calculated based on the measurement equation.

[0033] The resampled information is calculated based on the measurement vector of the resampled particles;

[0034] The corrected likelihood function is calculated based on the resampled information.

[0035] This invention also provides a soft-constraint target tracking system based on genetic resampling, comprising:

[0036] The target tracking model construction module is used to construct a target tracking model based on the tracked target; the target tracking model includes a state equation, a measurement equation, and a soft constraint model.

[0037] The particle generation module is used to generate multiple particles required for particle filtering based on the initial state value of the tracked target;

[0038] The prediction module is used to update each particle over time based on the target tracking model and predict the state and weight of each particle.

[0039] The resampling module is used to resample each particle using a genetic algorithm;

[0040] The modified likelihood function calculation module is used to calculate the modified likelihood function of the resampled particles based on the measurement equation.

[0041] The weight correction module is used to correct the predicted weights based on the corrected likelihood function.

[0042] The final estimated state determination module updates the predicted state according to the corrected weights to obtain the final estimated state of the tracked target.

[0043] Optionally, the prediction module specifically includes:

[0044] The state prediction unit is used to update the state of each particle over time using the state equation and predict the state of each particle.

[0045] A partitioning unit is used to divide each particle into particles that satisfy the soft constraints and particles that do not satisfy the soft constraints based on the soft constraint model.

[0046] An optimization unit is used to optimize the particles that do not meet the soft constraint conditions using a quasi-Newton method, and to obtain optimization results in the region that meets the soft constraint conditions.

[0047] The weight prediction unit is used to replace the particles that do not meet the soft constraints with the optimization results, and predict the weight of each particle based on the likelihood function; the likelihood function is calculated based on the measurement equation.

[0048] Optionally, the modified likelihood function calculation module specifically includes:

[0049] The measurement information calculation unit is used to calculate the measurement vector of the resampled particle based on the measurement equation.

[0050] The innovation calculation unit is used to calculate the resampled innovation based on the measurement vector of the resampled particle;

[0051] The modified likelihood function calculation unit is used to calculate the modified likelihood function based on the resampled information.

[0052] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0053] This invention models soft constraints by introducing auxiliary variables that satisfy an exponential distribution, and updates particle weights by combining the likelihood function of the auxiliary particle filter to form a modified likelihood function. This enables the estimation of target states under soft constraints. At the same time, it uses a genetic resampling method to resample particles to solve the particle degradation problem. This invention integrates known nonlinear inequality soft constraints in the tracking of ground moving targets, improving the estimation accuracy of ground targets under such soft constraints. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart of the soft-constraint target tracking method based on genetic resampling provided by the present invention;

[0056] Figure 2 A schematic diagram of the road constraint scenario under consideration.

[0057] Figure 3 A simulation diagram for tracking the scene;

[0058] Figure 4 This is a diagram illustrating the comparison of position estimation errors.

[0059] Figure 5 This is a diagram illustrating the comparison of velocity estimation errors. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] The purpose of this invention is to provide a soft-constraint target tracking method and system based on genetic resampling to improve the tracking accuracy of ground targets.

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] Example 1

[0064] like Figure 1 As shown, the soft-constraint target tracking method based on genetic resampling provided by this invention includes the following steps:

[0065] S1: Construct a target tracking model based on the tracked target; the target tracking model includes state equations, measurement equations, and soft constraint models.

[0066] Equations of state:

[0067] X k =F k X k-1 +wk-1 (1)

[0068] Measurement equation:

[0069] z k =H k X k +v k (2)

[0070] In the formula, This is a 4-dimensional state vector, representing the position and velocity of the ground target (i.e., the tracked target) in the x and y directions. The superscript T represents matrix transpose, and F... k Given the state transition matrix, w k-1 Let the variance be Q k-1 4D process noise, v k Let the variance be R k The 2D Gaussian measurement noise, with subscript k, k-1 representing a discrete time series. k =[r k θ k ] T It is a 2D measurement vector containing the relative distance measured by radar. and relative azimuth H k This is the measurement matrix.

[0071] Assume the ground target moves according to a two-dimensional cooperative turning model, i.e., the state transition matrix F k satisfy:

[0072]

[0073] Where Ω represents the turning speed.

[0074] In addition, the measurement matrix H k as follows:

[0075]

[0076] Consider a ground target moving on a circular road of known width; the road inequality constraint acting on the target can be described as follows:

[0077] R min ≤r k ≤R max (5)

[0078] R min R max These represent the upper and lower boundaries of the road, respectively. The road constraint scenarios considered in this invention are as follows: Figure 2 As shown.

[0079] Decompose equation (5) into two one-sided inequality constraints:

[0080]

[0081]

[0082] Among them, g1(X) k g2(X) k G1 and G2 are both intermediate variables.

[0083] Since soft constraints do not strictly limit the target state, a random variable Γ is introduced to establish a soft constraint model, namely:

[0084]

[0085] Wherein, the random variable Γ follows an exponential distribution with expectation μ, that is:

[0086]

[0087] p(·) represents the probability density function. It can be seen that by introducing a random variable Γ, the target state can exceed the constraint limit. However, the probability of the event occurring decreases exponentially with the degree of exceeding the constraint boundary under the influence of the exponential distribution.

[0088] S2: Generate multiple particles required for particle filtering based on the initial state value of the tracked target.

[0089] First, at the initial time, i.e., when k=0, N initial particles required for particle filtering are generated based on the initial value of the target state. and corresponding weights

[0090] The selection method is: from the initial state estimation Let P be the mean and the initial state covariance. 0|0 N particles are randomly sampled from a Gaussian distribution with variance to form the initial sampling points for particle filtering. And assign a weight to each particle. For other times such as k=1,..., perform the following steps S3 to S7.

[0091] S3: Based on the target tracking model, update the time of each particle and predict the state and weight of each particle. Specifically, this includes: updating the time of each particle using the state equation and predicting the state of each particle; dividing each particle into particles that satisfy the soft constraints and particles that do not satisfy the soft constraints based on the soft constraint model; optimizing the particles that do not satisfy the soft constraints using the quasi-Newton method to obtain the optimization results in the region that satisfies the soft constraints; replacing the particles that do not satisfy the soft constraints with the optimization results, and predicting the weight of each particle based on the likelihood function; the likelihood function is calculated based on the measurement equation.

[0092] S31: Perform a one-step state prediction for each particle and classify them according to whether they meet the soft constraint conditions.

[0093] For each particle, the state equation is used to perform a time update, resulting in a one-step state prediction estimate for each particle.

[0094]

[0095] Estimation based on the one-step predicted state of each particle Whether soft constraints are satisfied is divided into two categories: those containing K. a A naturally constrained particle λ i and weight θ i set {λ i ,θ i}, i=1,...,K a and K b A particle η that does not satisfy the soft constraints i and weight gather Among them, K a =NK b ,Right now

[0096]

[0097] in, One-step prediction estimation for particles The first and third dimensions represent the position in the x and y directions.

[0098] S32: Optimize particles that do not meet the soft constraints using the quasi-Newton method, and obtain their optimization results within the region that meets the soft constraints.

[0099]

[0100] Among them, X ini To utilize the set of particles {λ} that naturally satisfy soft constraints i ,θi}, i=1,...,K a Determine the initial point for state optimization.

[0101]

[0102] Where, d i For the search direction, satisfying:

[0103]

[0104] In the formula, J represents the gradient calculation operation. i The representative index function, due to process noise w k-1 The distribution has a mean of 0 and a variance of Q. k-1 Since the random variable Γ in the soft-constraint model follows a Gaussian distribution with parameter μ, the index function can be written as:

[0105]

[0106] In the formula, N(X) k η i Q k-1 ) indicates that the mean is η i The variance is Q k-1 Given the Gaussian distribution, it can be seen that the purpose of this step is to obtain the location where the mean of each predicted particle is most likely to occur, considering the presence of soft constraints.

[0107] α i is the scaling factor, which is the minimum value in the following three cases:

[0108]

[0109] S33: Replace the particles that did not meet the constraints with the optimization results of each particle, and calculate the predicted particle weights in one step based on the likelihood function.

[0110] After step S32, all particles are either within the region satisfying the soft constraints, or have been optimized to reach the boundary range satisfying the soft constraints. The new set of particles and their corresponding weights is then: Calculate the likelihood function of each particle to obtain the one-step predicted particle weight update.

[0111] Calculation measurement one-step prediction

[0112]

[0113] Calculate new information

[0114]

[0115] Calculate the likelihood function ρ i :

[0116]

[0117] The normalized one-step prediction weights are then updated as follows:

[0118]

[0119] S4: Resample each particle using a genetic algorithm.

[0120] Because particle filtering suffers from particle degradation, after a period of operation, a small number of particles have weights close to 1 while most tend to 0, resulting in a loss of particle diversity and causing the filter to diverge. To solve the particle degradation problem, a genetic algorithm is used to resample the particles based on genetic resampling technology.

[0121] The main idea of ​​the genetic resampling process is to select, crossover, and mutate the sampled particles to obtain a new set of particles, thereby performing state estimation.

[0122] S41: Set the selection probability P s Crossover probability P c And the probability of mutation P m Generally speaking, P m =0.01~0.1, satisfying P s +P c +P m =1.

[0123] S42: Selection. Using the polynomial resampling method, select N with larger weights. s =P s ×N particles are used as part of a new particle swarm.

[0124]

[0125] S43: Crossover. Using the arithmetic crossover method, two randomly selected particles... By crossing over, we obtain the crossed-over particles. Finally, we get N. c =P c ×N cross-particles.

[0126]

[0127] In the formula, α c A random number between 0 and 1.

[0128] S44: Mutation. N is obtained using the method of real number mutation.m =P m ×N mutated particles, the process is as follows:

[0129]

[0130] In the formula, α m A random number between 0 and 1; β m It is a random positive real number between 0 and 5.

[0131] S45: Merge particles. Select cross Mutated particles The group is merged as a new particle swarm

[0132] S5: Calculate the corrected likelihood function of the resampled particles based on the measurement equation. Specifically, this includes: calculating the measurement vector of the resampled particles based on the measurement equation; calculating the new information after resampling based on the measurement vector of the resampled particles; and calculating the corrected likelihood function based on the new information after resampling.

[0133] Calculate the corrected likelihood function for each particle based on genetic resampling to obtain the corrected particle weight update.

[0134] Calculate the one-step prediction of the resampled measurement

[0135]

[0136] Calculate the new information after resampling

[0137]

[0138] Calculate the modified likelihood function ρ considering soft constraints. r,i :

[0139]

[0140] S6: Adjust the predicted weights based on the modified likelihood function.

[0141] The normalized corrected particle weights are updated as follows:

[0142]

[0143] S7: Update the predicted state according to the corrected weights to obtain the final estimated state of the tracked target.

[0144] The final state estimate considering soft constraints is as follows:

[0145]

[0146] To verify the effectiveness of the proposed method, a tracking scenario was first constructed and simulated. A ground target performs a two-dimensional cooperative turning motion with an angular velocity of -3° / s along a road from 0 to 30 seconds. The target's initial position is x0 = 0m, y0 = 2000m, and its initial velocity is V. x0 =104 m / s, V y0 =0m / s, the radar is located at the origin of the coordinate system, and the upper and lower boundaries of the road are R. min =1980m,R max =2020m, simulation scenario as follows Figure 3 As shown.

[0147] As can be seen, although the target remains within the constraints on both sides of the road for most of the time, there are periods when its position exceeds the constraint boundaries. Tracking simulations were performed using Extended Kalman Filter (EKF), Auxiliary Particle Filter (CAPF) considering hard constraints, and the proposed algorithm based on genetic resampling auxiliary particle filtering considering soft constraints. A two-dimensional cooperative turning model was selected as the target motion model. The simulation step size was T = 0.1 s, and the sensor measurement noise was v. t It follows a Gaussian distribution with variance R. k =diag[100 2 2 2 The initial state error covariance is P0 = diag[(150m)]. 2 (10m) 2 (15m / s) 2 (10m / s) 2 ].

[0148] Tracking results for example Figure 4 and Figure 5 As shown, traditional unconstrained extended Kalman filtering cannot consider the motion characteristics of the target motion. Therefore, it does not restrict the boundary of the target motion state during estimation, which means that the target motion may greatly exceed the boundary. Due to the lack of consideration of constraint prior information, the estimation accuracy is the worst. Hard constraints, on the other hand, strictly restrict the target to the road, even if the target occasionally exceeds the constraint boundary. The algorithm proposed in this invention models the soft constraints as random variables with exponentially distributed boundary conditions. Based on the auxiliary particle filter, it models the probability of the target exceeding the constraint boundary. Therefore, the estimated result considers the constraint conditions and allows the target to exceed the constraint boundary to a small extent, resulting in the highest estimation accuracy and achieving high-precision tracking of targets considering soft constraints.

[0149] Example 2

[0150] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a soft-constraint target tracking system based on genetic resampling is provided below.

[0151] The system includes:

[0152] The target tracking model construction module is used to construct a target tracking model based on the tracked target; the target tracking model includes a state equation, a measurement equation, and a soft constraint model.

[0153] The particle generation module is used to generate multiple particles required for particle filtering based on the initial state value of the tracked target.

[0154] The prediction module is used to update each particle over time based on the target tracking model and predict the state and weight of each particle.

[0155] The resampling module is used to resample each particle using a genetic algorithm.

[0156] The modified likelihood function calculation module is used to calculate the modified likelihood function of the resampled particles based on the measurement equation.

[0157] The weight correction module is used to correct the predicted weights based on the corrected likelihood function.

[0158] The final estimated state determination module updates the predicted state according to the corrected weights to obtain the final estimated state of the tracked target.

[0159] Furthermore, the prediction module specifically includes:

[0160] The state prediction unit is used to update the state of each particle over time using the state equation and predict the state of each particle.

[0161] A partitioning unit is used to divide each particle into particles that satisfy the soft constraints and particles that do not satisfy the soft constraints based on the soft constraint model.

[0162] An optimization unit is used to optimize the particles that do not meet the soft constraint conditions using a quasi-Newton method, and to obtain optimization results in the region that meets the soft constraint conditions.

[0163] The weight prediction unit is used to replace the particles that do not meet the soft constraints with the optimization results, and predict the weight of each particle based on the likelihood function; the likelihood function is calculated based on the measurement equation.

[0164] Furthermore, the modified likelihood function calculation module specifically includes:

[0165] The measurement information calculation unit is used to calculate the measurement vector of the resampled particles based on the measurement equation.

[0166] The innovation calculation unit is used to calculate the resampled innovation based on the measurement vector of the resampled particles.

[0167] The modified likelihood function calculation unit is used to calculate the modified likelihood function based on the resampled information.

[0168] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0169] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A soft constraint object tracking method based on genetic resampling, characterized in that, include: Build a target tracking model based on the target being tracked; The target tracking model includes a state equation, a measurement equation, and a soft constraint model; Generate multiple particles required for particle filtering based on the initial state value of the tracked target; Based on the target tracking model, each particle is updated over time to predict its state and weight. Resampling of each particle is performed using a genetic algorithm; The corrected likelihood function of the resampled particles is calculated based on the measurement equation. The predicted weights are adjusted based on the modified likelihood function; The final estimated state of the tracked target is obtained by updating the predicted state based on the corrected weights. The state equation is expressed as follows: The expression for the measurement equation is as follows: In the formula, This is a 4-dimensional state vector, representing the position and velocity of the tracked target in the x and y directions, with superscripts... Represents matrix transpose. Given the state transition matrix, The variance is 4D process noise, The variance is The 2D Gaussian measurement noise, where the subscript k, k-1 represents the discrete time series. It is a 2D measurement vector containing the relative distance measured by radar. and relative azimuth , For measurement matrix; Assuming the tracked target moves according to a two-dimensional cooperative turning model, the state transition matrix is... satisfy: in, Turning speed; The road inequality constraint on the tracked target is described as follows: These are the upper and lower boundaries of the road, respectively. Specifically, based on the target tracking model, the time updates of each particle are performed to predict the state and weight of each particle, including: The state equation is used to update the state of each particle over time, and the state of each particle is predicted. Based on the soft constraint model, each particle is divided into particles that satisfy the soft constraints and particles that do not satisfy the soft constraints. The quasi-Newton method is used to optimize particles that do not meet soft constraints, and the optimization results are obtained in the region where the soft constraints are met. : ;in, To utilize naturally satisfied soft-constraint particle sets Determine the initial point for state optimization; ;in, For the search direction, satisfying: In the formula, , This indicates the gradient calculation operation. The representative indicator function; the indicator function is written as: ; In the formula, The mean is The variance is Gaussian distribution; The optimization results are used to replace the particles that do not meet the soft constraints, and the weights of each particle are predicted based on the likelihood function, which is calculated based on the measurement equation.

2. The soft-constraint target tracking method based on genetic resampling according to claim 1, characterized in that, The expression for the soft constraint model is as follows: in, For random variables, and As an intermediate variable, , These represent the upper and lower boundaries of the road, respectively.

3. The soft-constraint target tracking method based on genetic resampling according to claim 1, characterized in that, The genetic algorithm is used to resample each particle, specifically including: Select, crossover, and mutation operations are performed on the particles to obtain resampled particles.

4. The soft-constraint target tracking method based on genetic resampling according to claim 1, characterized in that, The corrected likelihood function of the resampled particles is calculated based on the measurement equation, specifically including: The measurement vector of the resampled particle is calculated based on the measurement equation. The resampled information is calculated based on the measurement vector of the resampled particles; The corrected likelihood function is calculated based on the resampled information.

5. A soft-constraint target tracking system based on genetic resampling, characterized in that, The system is used to execute the soft-constraint target tracking method based on genetic resampling as described in any one of claims 1-4, the system comprising: The target tracking model construction module is used to construct a target tracking model based on the tracked target; the target tracking model includes a state equation, a measurement equation, and a soft constraint model. The particle generation module is used to generate multiple particles required for particle filtering based on the initial state value of the tracked target; The prediction module is used to update each particle over time based on the target tracking model and predict the state and weight of each particle. The resampling module is used to resample each particle using a genetic algorithm; The modified likelihood function calculation module is used to calculate the modified likelihood function of the resampled particles based on the measurement equation. The weight correction module is used to correct the predicted weights based on the corrected likelihood function. The final estimated state determination module updates the predicted state according to the corrected weights to obtain the final estimated state of the tracked target.

6. The soft-constraint target tracking system based on genetic resampling according to claim 5, characterized in that, The prediction module specifically includes: The state prediction unit is used to update the state of each particle over time using the state equation and predict the state of each particle. A partitioning unit is used to divide each particle into particles that satisfy the soft constraints and particles that do not satisfy the soft constraints based on the soft constraint model. An optimization unit is used to optimize the particles that do not meet the soft constraint conditions using a quasi-Newton method, and to obtain optimization results in the region that meets the soft constraint conditions. The weight prediction unit is used to replace the particles that do not meet the soft constraints with the optimization results, and predict the weight of each particle based on the likelihood function; the likelihood function is calculated based on the measurement equation.

7. The soft-constraint target tracking system based on genetic resampling according to claim 5, characterized in that, The modified likelihood function calculation module specifically includes: The measurement information calculation unit is used to calculate the measurement vector of the resampled particle based on the measurement equation. The innovation calculation unit is used to calculate the resampled innovation based on the measurement vector of the resampled particle; The modified likelihood function calculation unit is used to calculate the modified likelihood function based on the resampled information.

Images