WO2020173105A1 - 机动目标跟踪方法及装置 - Google Patents

机动目标跟踪方法及装置 Download PDF

Info

Publication number
WO2020173105A1
WO2020173105A1 PCT/CN2019/112696 CN2019112696W WO2020173105A1 WO 2020173105 A1 WO2020173105 A1 WO 2020173105A1 CN 2019112696 W CN2019112696 W CN 2019112696W WO 2020173105 A1 WO2020173105 A1 WO 2020173105A1
Authority
WO
WIPO (PCT)
Prior art keywords
sub
target
model
state
models
Prior art date
Application number
PCT/CN2019/112696
Other languages
English (en)
French (fr)
Inventor
李良群
王小梨
谢维信
刘宗香
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Publication of WO2020173105A1 publication Critical patent/WO2020173105A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01CMEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/20Instruments for performing navigational calculations
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/23Clustering techniques
    • G06F18/232Non-hierarchical techniques
    • G06F18/2321Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/20Analysis of motion
    • G06T7/277Analysis of motion involving stochastic approaches, e.g. using Kalman filters

Definitions

  • the invention relates to the technical field of target tracking, in particular to a method and device for tracking a mobile target.
  • Target tracking is to accurately predict and estimate the target's future trajectory based on the target's past state and observations.
  • the Interacting Multiple Model (IMM) algorithm selects multiple parallel models, and switches between the parallel models according to the Markov probability transition matrix to effectively predict and estimate the trajectory of the target.
  • the IMM algorithm is likely to cause large errors during the switching process.
  • the standard IMM algorithm further reduces the accuracy of model matching and state estimation. Therefore, the key to the IMM algorithm is the choice of filtering methods.
  • the multi-model algorithm based on non-linear filtering is currently the most widely used multi-model algorithm, and the more popular non-linear filtering method is the Extended Kalman Filter (Extended Kalman Filter, EKF).
  • EKF Extended Kalman Filter
  • the performance of the EKF drops sharply, which causes the problem of low target tracking accuracy.
  • the main purpose of the embodiments of the present invention is to provide a method and device for tracking a maneuvering target, which can improve the accuracy of target tracking.
  • the first aspect of the embodiments of the present invention provides a method for tracking a maneuvering target.
  • the method includes: expressing target feature information of a maneuvering target as a plurality of semantic fuzzy sets, and constructing a TS semantic fuzzy multiple set based on the plurality of semantic fuzzy sets.
  • the TS semantic fuzzy multi-model includes a plurality of sub-models; based on the unscented Kalman filter algorithm for subsequent parameter identification to determine the target state and target state covariance of each sub-model; based on the fuzzy C regression clustering algorithm
  • the antecedent parameter identification determines the target fuzzy membership function of the antecedent parameters of each sub-model; according to the target state, the covariance of the target state and the target fuzzy membership function of the antecedent parameters of each of the sub-models, the State estimation and covariance estimation of the maneuvering target; predict the trajectory of the maneuvering target according to the state estimation and the covariance estimation of the maneuvering target.
  • a second aspect of the embodiments of the present invention provides a maneuvering target tracking device.
  • the device includes: a building module for representing target feature information of a maneuvering target as multiple semantic fuzzy sets, and based on the multiple semantic fuzzy sets Construct a TS semantic fuzzy multi-model, the TS semantic fuzzy multi-model includes multiple sub-models; the determination module is used to identify the subsequent parameters based on the unscented Kalman filter algorithm, and determine the target state and target state coordination of each sub-model Variance; the determination module is also used to identify the antecedent parameters based on the fuzzy C regression clustering algorithm, and determine the target fuzzy membership function of the antecedent parameters of each of the sub-models; the calculation module is used to determine the target fuzzy membership function of each of the sub-model The target state, the covariance of the target state and the target fuzzy membership function of the antecedent parameters of the model are used to obtain the state estimation and covariance estimation of the maneuvering target; the prediction module is used to estimate the state of the maneuvering target and the covariance Variance estimation,
  • a T-S semantic fuzzy multi-model is constructed, thereby approximating the dynamic model with high precision.
  • the fuzzy C-based regression clustering algorithm is used to realize the identification of the antecedent parameters.
  • the unscented Kalman filter algorithm is introduced to identify the subsequent parameters, thereby effectively Targets are accurately tracked.
  • FIG. 1 is a schematic diagram of the implementation process of the maneuvering target tracking method in the first embodiment of the present invention
  • FIG. 2 is a schematic diagram of the implementation process of the maneuvering target tracking method in the second embodiment of the present invention.
  • step 202 is a schematic flowchart of the detailed steps of step 202 in the second embodiment of the present invention.
  • step 203 is a schematic flowchart of the detailed steps of step 203 in the second embodiment of the present invention.
  • FIG. 5 is a framework diagram of a method for tracking a maneuvering target in a second embodiment of the present invention.
  • FIG. 6 is a comparison diagram of application trajectories of the maneuvering target tracking method in the second embodiment of the present invention.
  • FIG. 7 is a comparison diagram of the application root mean square error of the maneuvering target tracking method in the second embodiment of the present invention.
  • Fig. 8 is a schematic structural diagram of a mobile target tracking device in a third embodiment of the present invention.
  • FIG. 1 is a schematic diagram of the implementation process of the maneuvering target tracking method in the first embodiment of the present invention. As shown in Figure 1, the method mainly includes the following steps:
  • T-S semantic fuzzy multi-model based on the multiple semantic fuzzy sets, where the T-S semantic fuzzy multi-model includes multiple sub-models.
  • the Takagi-Sugeno (TS) fuzzy model is a model of a nonlinear system described by a set of "IF-THEN" fuzzy rules. Each fuzzy rule corresponds to a sub-model.
  • the TS fuzzy model can be represented by multiple linear sub-models. Arbitrary precision nonlinear system.
  • the target characteristic information of the maneuvering target can include distance, speed, innovation or heading angle difference.
  • the innovation expression may be expressed as two semantic fuzzy sets (Small, Large)
  • the heading angle difference may be expressed as three semantic fuzzy sets (Negative Large, Small, Positive Large).
  • each fuzzy rule is defined as follows:
  • the state equation is a linear model
  • the observation equation is a nonlinear model
  • the unscented transformation is obtained by calculating the statistical properties of the function random quantity after the statistical properties of a certain known random variable. First estimate the probability density function of the state vector through the nonlinear state equation, take out a set of determined sampling points, then calculate the sampling points according to the unscented transformation, and obtain the corresponding posterior statistical characteristics through nonlinear calculations, and finally use the linear regression method Get posterior statistics.
  • the Unscented Kalman Filter (UKF) does not need to linearize the nonlinear system, and is easily applied to the state estimation of the nonlinear system.
  • UKF mainly uses unscented transformation to obtain Sigma particles with different weights.
  • the clustering algorithm of unsupervised learning in pattern recognition is usually used.
  • the fuzzy C-Means Fuzzy C-Means, FCM
  • FCM fuzzy C-Means
  • FCRM fuzzy C-Recursive Model
  • target tracking refers to the technology of accurately predicting and estimating the future trajectory of the target based on the past state and observation of the target. Therefore, the target state, the covariance of the target state and the target blurring of the antecedent parameters obtained from the identification of the sub-models
  • the membership function estimates the state of the maneuvering target, obtains the state estimation and covariance estimation of the maneuvering target, and then performs trajectory prediction.
  • a T-S semantic fuzzy multi-model is constructed, thereby approximating the dynamic model with high accuracy.
  • the FCRM algorithm is used to realize the identification of the antecedent parameters.
  • the UKF algorithm is introduced to identify the subsequent parameters, and then the target can be accurately tracked.
  • FIG. 2 is a schematic diagram of the implementation process of the maneuvering target tracking method in the second embodiment of the present invention. As shown in Figure 2, the method mainly includes the following steps:
  • target feature information of a maneuvering target as multiple semantic fuzzy sets, and construct a T-S semantic fuzzy multi-model based on the multiple semantic fuzzy sets, and the T-S semantic fuzzy multi-model includes multiple sub-models.
  • step 202 specifically includes:
  • a preset sampling rule determine a plurality of first sampling points and a weighted value of each of the first sampling points from the initial state of each sub-model.
  • sampling rules 2n X +1 sampling points and corresponding weighting values are determined.
  • represents the scale parameter, which can be any value of n X + ⁇ 0, for The jth column of the root mean square, and Respectively represent the first sampling points corresponding to the 0th column, 1 ⁇ n X columns and n X +1 ⁇ 2n X +1 columns, W 0 , W j and Respectively represent the weighted values of the first sampling points corresponding to the 0th column, 1 to n X columns, and n X +1 to 2n X +1, and n X is the dimension of the state vector x.
  • the following formula is used to perform one-step prediction according to each of the first sampling points and the weighted values of each of the first sampling points to obtain the update state and update state covariance of each sub-model:
  • Q represents process noise Covariance
  • sampling rule determine a plurality of second sampling points and a weighted value of each second sampling point from the update state of each sub-model.
  • steps 304 and 305 are implemented by the following formula:
  • the target state and target state covariance of each sub-model are determined according to the update state and update state covariance of each sub-model, as well as the update observation and the update observation variance:
  • z k represents the preset observation set of the maneuvering target, Represents the target state of the i-th model at time k, Represents the target state covariance of the i-th model at k time.
  • step 203 specifically includes the following steps:
  • the preset observation set z k is expressed as:
  • z k,l represents l th observation, while Indicates the predicted observation based on the fuzzy rule i th at time k.
  • the distance measurement function is expressed as follows:
  • Is the distance measurement function Indicates a given goal state
  • the observation z k,l likelihood function Represents the innovation covariance matrix.
  • FCRM general objective function
  • m refers to the weight index, which is generally 2.
  • the defined objective function is:
  • ⁇ k is the Lagrange multiplier vector, Is the distance measurement function.
  • the fuzzy membership of observation l at time k is:
  • the fuzzy membership function of the antecedent parameters of the sub-model is set as a Gaussian function:
  • model probability of each sub-model is obtained according to the target fuzzy membership function of the antecedent parameters of each sub-model through the following formula:
  • the standardized model probability of each sub-model is calculated according to the model probability of each sub-model through the following formula:
  • the state estimation of the maneuvering target is obtained through the following formula:
  • the algorithm implementation framework in this embodiment is shown in Figure 5, which mainly includes three parts: the identification of subsequent parameters based on UKF, the identification of previous parameters based on FRCM, and the fusion of the sub-models of the TS fuzzy model .
  • a simulation radar maneuvering target tracking problem is analyzed.
  • This problem has certain guiding significance for prevention and control applications.
  • the algorithm in this embodiment is simultaneously compared with the traditional IMM algorithm and the IMM-UKF algorithm, and all experiments are performed 100 times Monte Carlo simulation.
  • N f represents the total number of fuzzy rules, Represents the state vector, x k represents the target x-axis coordinate, y k represents the target y-axis coordinate, with They respectively indicate the speed of the target in the x-axis and y-axis coordinates.
  • the observation noise v k is non-Gaussian distributed noise.
  • the non-Gaussian noise in the simulation is mainly generated by the superposition of two Gaussian noises, where R is similar to the covariance matrix of Gaussian noise:
  • 0 x 0 , P 0
  • 0 diag (0.15 2 , 0.01, 0.15 2 , 0.01).
  • innovation and heading angle difference are selected as target feature information because these information can effectively reflect the movement state of the target.
  • innovation can reflect whether the target motion model is appropriate. When the innovation is large, it indicates the target The motion model is not in line with the current motion state, and the weight of each model is adjusted according to the innovation to obtain a more accurate motion model.
  • the initial values of the membership functions of the two fuzzy sets (Small, Large) and the three fuzzy sets of the heading angle difference (Negative Large, Small, Positive Large) are respectively The initial covariance is At the same time, the position of the sensor is at the origin of coordinates. Is the state transition matrix, and its representation method is as follows:
  • the turning rate ⁇ i is determined by the TS fuzzy model.
  • Table 1 shows that for different input variables And ⁇ k , the turning rate ⁇ i and the process noise standard deviation
  • Figure 6 shows the target trajectory and the estimated trajectory of the TS-UKF algorithm.
  • the tracking effect of the algorithm in this embodiment is basically the same as the trajectory of the simulation, and there is no obvious tracking loss.
  • the target maneuvering it shows good robustness, indicating the The algorithm can efficiently process uncertain information in nonlinear systems.
  • Fig. 7(ac) respectively describes the root mean square error of the target position, the root mean square error in the x-axis direction and the root mean square error in the y-axis direction.
  • the algorithm in this embodiment is used when the target is maneuvering.
  • the tracking effect is better than the other two algorithms, showing relatively stable tracking performance.
  • the IMM and IMM-UKF algorithms have large errors when the target is turning. The main reason is that the model set used in the IMM algorithm may not be large enough.
  • the TS-UKF algorithm can construct a target motion model based on multiple semantic information represented by the target's spatial feature information.
  • Table 2 shows the statistical results of the root mean square error of the three algorithms. From the data in the table, the tracking accuracy of TS-UKF is 30.32% and 2.17% higher than that of IMM and IMM-UKF in the position root mean square error, respectively, reflecting a more accurate tracking effect.
  • a T-S fuzzy multi-model is constructed by using multiple semantic fuzzy sets to fuzzyly represent the target feature information, thereby approximating the dynamic model with high precision.
  • the membership function of the antecedent parameters in the T-S fuzzy multi-model is used to adaptively adjust the weight of each rule to further improve the accuracy of the target motion model, thereby improving the accuracy of target tracking.
  • the UKF algorithm is introduced to identify the subsequent parameters, which improves the filtering accuracy.
  • FIG. 8 is a schematic structural diagram of a maneuvering target tracking device in a third embodiment of the present invention. As shown in Figure 8, the device mainly includes:
  • the construction module 501 is used to express the target feature information of the maneuvering target as multiple semantic fuzzy sets, and construct a T-S semantic fuzzy multi-model based on the multiple semantic fuzzy sets.
  • the T-S semantic fuzzy multi-model includes multiple sub-models.
  • the determining module 502 is used to identify the subsequent parameters based on the unscented Kalman filter algorithm, and determine the target state and the target state covariance of each sub-model.
  • the determining module 502 is also used to identify the antecedent parameters based on the fuzzy C regression clustering algorithm, and determine the target fuzzy membership function of the antecedent parameters of each sub-model.
  • the calculation module 503 is used to obtain the state estimation and the covariance estimation of the maneuvering target according to the target state, the covariance of the target state and the target fuzzy membership function of the antecedent parameters of each sub-model.
  • the prediction module 504 is used to predict the trajectory of the maneuvering target according to the state estimation and the covariance estimation of the maneuvering target.
  • calculation module 503 is also used to calculate the standardized model probability of each sub-model according to the target fuzzy membership function of the antecedent parameters of each sub-model.
  • the calculation module 503 is also used to obtain the state estimation of the maneuvering target according to the target state of each sub-model and the standardized model probability.
  • the calculation module 503 is also used to obtain the covariance estimate of the maneuvering target according to the target state covariance of each sub-model and the standardized model probability.
  • the determining module 502 is also used to determine the initial state and initial observation of each sub-model based on the discrete dynamic system.
  • the determining module is also used to determine a plurality of first sampling points and a weighted value of each first sampling point from the initial state of each sub-model according to a preset sampling rule.
  • the determining module 502 is further configured to perform a one-step prediction according to each first sampling point and the weighted value of each first sampling point to obtain the update state and update state covariance of each sub-model.
  • the determining module 502 is further configured to determine multiple second sampling points and the weighted value of each second sampling point from the update state of each sub-model according to the sampling rule.
  • the determining module 502 is further configured to obtain updated observations and updated observation variances according to the initial state, the update state and update state covariance of each sub-model, and the weighted values of each second sampling point and each second sampling point.
  • the determining module 502 is further configured to determine the target state and target state covariance of each sub-model according to the update state and update state covariance of each sub-model, and the update observation and update observation variance.
  • the determining module 502 is further configured to construct a distance measurement function according to the preset observation set, the preset prediction observation set, and the target state of each sub-model.
  • the determining module 502 is also used to define the objective function according to the distance measurement function and the constraint condition of the fuzzy membership function of the antecedent parameters of each sub-model.
  • the determining module 502 is also used to obtain partial derivatives of the fuzzy membership functions of the antecedent parameters of each sub-model according to the objective function, and obtain the updated fuzzy membership functions of the antecedent parameters of each sub-model.
  • the determining module 502 is also used to obtain the membership matrix according to the updated fuzzy membership function of the antecedent parameters of each sub-model.
  • the determining module 502 is also used to set the fuzzy membership function of the antecedent parameters of the sub-models to a Gaussian function, and determine the target fuzzy membership function of the antecedent parameters of each sub-model according to the membership matrix.
  • calculation module 503 is also used to obtain the model probability of each sub-model according to the target fuzzy membership function of the antecedent parameters of each sub-model.
  • the calculation module 503 is also used to calculate the standardized model probability of each sub-model according to the model probability of each sub-model.
  • a T-S fuzzy multi-model is constructed by using multiple semantic fuzzy sets to fuzzyly represent the target feature information, thereby approximating the dynamic model with high precision.
  • the membership function of the antecedent parameters in the T-S fuzzy multi-model is used to adaptively adjust the weight of each rule to further improve the accuracy of the target motion model, thereby improving the accuracy of target tracking.
  • the UKF algorithm is introduced to identify the subsequent parameters, which improves the filtering accuracy.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Data Mining & Analysis (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Automation & Control Theory (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

一种机动目标跟踪方法及装置,应用于目标跟踪技术领域。该方法包括:将机动目标的目标特征信息表示为多个语义模糊集,并根据多个语义模糊集构建T-S语义模糊多模型,T-S语义模糊多模型包括多个子模型(101)。基于无迹卡尔曼滤波算法进行后件参数辨识,确定各子模型的目标状态和目标状态协方差(102)。基于模糊C回归聚类算法进行前件参数辨识,确定各子模型的前件参数的目标模糊隶属度函数(103)。根据各子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到机动目标的状态估计和协方差估计(104)。根据机动目标的状态估计和协方差估计,预测机动目标的轨迹(105)。该方法可提高目标跟踪的精确度。

Description

机动目标跟踪方法及装置 技术领域
本发明涉及目标跟踪技术领域,尤其涉及一种机动目标跟踪方法及装置。
背景技术
目标跟踪是根据目标的过去状态和观测,精确地预测和估计目标的未来轨迹。为了获取目标的精确位置、速度和加速度,设计目标的精确模型是非常关键的。其中,交互多模型(Interacting Multiple Model,IMM)算法选择多个并行模型,并根据马尔可夫概率转移矩阵在各并行模型之间进行切换来有效地预测和估计目标的轨迹。然而,由于模型切换机制的存在,使得IMM算法在切换过程中容易造成较大的误差。特别是对于不确定性模型的估计,标准的IMM算法更加降低了模型匹配和状态估计的精度。因此,IMM算法的关键是滤波方法的选择。
近年来,为了提高IMM算法的估计精度,基于非线性滤波的多模型算法是目前应用最广泛的多模型算法,比较流行的非线性滤波方法是扩展卡尔曼滤波(ExtendedKalmanFilter,EKF)。然而,随着动态系统非线性的加剧,EKF的性能急剧下降,这就存在目标跟踪的精确度较低的问题。
技术问题
本发明实施例的主要目的在于提供机动目标跟踪方法及装置,可提高目标跟踪的精确度。
技术解决方案
本发明实施例第一方面提供了一种机动目标跟踪方法,所述方法包括:将机动目标的目标特征信息表示为多个语义模糊集,并根据多个所述语义模糊集构建T-S语义模糊多模型,所述T-S语义模糊多模型包括多个子模型;基于无迹卡尔曼滤波算法进行后件参数辨识,确定各所述子模型的目标状态和目标状态协方差;基于模糊C回归聚类算法进行前件参数辨识,确定各所述子模型的前件参数的目标模糊隶属度函数;根据各所述子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到所述机动目标的状态估计和协方差估计;根据所述机动目标的状态估计和所述协方差估计,预测所述机动目标的轨迹。
本发明实施例第二方面提供了一种机动目标跟踪装置,所述装置包括:构建模块,用于将机动目标的目标特征信息表示为多个语义模糊集,并根据多个所述语义模糊集构建T-S语义模糊多模型,所述T-S语义模糊多模型包括多个子模型;确定模块,用于基于无迹卡尔曼滤波算法进行后件参数辨识,确定各所述子模型的目标状态和目标状态协方差;所述确定模块,还用于基于模糊C回归聚类算法进行前件参数辨识,确定各所述子模型的前件参数的目标模糊隶属度函数;计算模块,用于根据各所述子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到所述机动目标的状态估计和协方差估计;预测模块,用于根据所述机动目标的状态估计和所述协方差估计,预测所述机动目标的轨迹。
有益效果
从上述实施例可知,通过利用多个语义模糊集对目标特征信息进行模糊表示,构建T-S语义模糊多模型,从而高精度地逼近动态模型。 另外,在T-S语义模糊多模型中,利用基于模糊C回归聚类算法实现对前件参数的辨识,同时,为了实现非线性特征,引入无迹卡尔曼滤波算法辨识后件参数,进而有效地对目标进行精确跟踪。
附图说明
图1是本发明提供的第一实施例中的机动目标跟踪方法的实现流程示意图;
图2是本发明提供的第二实施例中的机动目标跟踪方法的实现流程示意图;
图3是本发明提供的第二实施例中的步骤202的细化步骤的流程示意图;
图4是本发明提供的第二实施例中的步骤203的细化步骤的流程示意图;
图5是本发明提供的第二实施例中的机动目标跟踪方法的框架图;
图6是本发明提供的第二实施例中的机动目标跟踪方法的应用轨迹对比图;
图7是本发明提供的第二实施例中的机动目标跟踪方法的应用均方根误差对比图;
图8是本发明提供的第三实施例中的机动目标跟踪装置的结构示意图。
本发明的实施方式
为使得本发明的发明目的、特征、优点能够更加的明显和易懂,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而非全部实施例。基于本发明中的实施例,本领域技术人员在 没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
请参阅图1,图1是本发明提供的第一实施例中的机动目标跟踪方法的实现流程示意图。如图1所示,该方法主要包括以下步骤:
101、将机动目标的目标特征信息表示为多个语义模糊集,并根据多个该语义模糊集构建T-S语义模糊多模型,该T-S语义模糊多模型包括多个子模型。
具体的,为了获得精确的目标跟踪性能,模糊逻辑作为动态系统参数辨识和综合控制规律的一种通用技术而被广泛应用开来。其中,Takagi–Sugeno(T-S)模糊模型是由一组“IF-THEN”模糊规则来描述非线性系统的模型,每条模糊规则对应一个子模型,T-S模糊模型可利用多个线性子模型来表示任意精度的非线性系统。
在实际应用过程中,机动目标的目标特征信息可包括距离、速度、新息或航向角差。示例性的,新息表示可表示为两个语义模糊集(Small,Large),航向角差可表示为三个语义模糊集(Negative Large,Small,Positive Large)。
其中,设定模糊规则的总数目,每条模糊规则定义如下:
Figure PCTCN2019112696-appb-000001
上式中,
Figure PCTCN2019112696-appb-000002
表示k时刻机动目标的第G个目标特征信息,
Figure PCTCN2019112696-appb-000003
表示第G个目标特征信息的模糊隶属度函数,
Figure PCTCN2019112696-appb-000004
表示k时刻第i个子模型的状态转移矩阵,
Figure PCTCN2019112696-appb-000005
表示k时刻第i个子模型的观测函数,
Figure PCTCN2019112696-appb-000006
表示第i个子模型的过程噪声和观测噪声,
Figure PCTCN2019112696-appb-000007
表示k-1 时刻第i个子模型的状态,
Figure PCTCN2019112696-appb-000008
表示k时刻第i个子模型的状态,
Figure PCTCN2019112696-appb-000009
表示k时刻第i个子模型的观测,N f表示模糊规则总数目。
其中,状态方程为线性模型,观测方程为非线性模型。
102、基于无迹卡尔曼滤波算法进行后件参数辨识,确定各该子模型的目标状态和目标状态协方差。
具体的,无迹变换是通过对一定的已知随机变量统计特性后求其函数随机量的统计特性而得出。首先通过非线性状态方程来估算状态向量的概率密度函数,取出一组确定的采样点,然后根据无迹变换计算采样点,并通过非线性计算得到相应后验统计特征,最后用线性回归的方法得到后验统计。其中,无迹卡尔曼滤波(Unscented Kalman Filter,UKF)不需要对非线性系统进行线性化,并便于地应用于非线性系统的状态估计。其中,UKF主要是利用无迹变换获取带有不同权值的Sigma粒子。
103、基于模糊C回归聚类算法进行前件参数辨识,确定各该子模型的前件参数的目标模糊隶属度函数。
具体的,为了识别前件参数,通常使用模式识别中的无监督学习的聚类算法。其中,模糊C均值聚类(Fuzzy C-Means,FCM)算法应用最为广泛的模糊聚类算法,但FCM算法只适用于超球面的数据分类中,而本实施例中T-S模糊模型的模糊规则对应超平面状态,因此,采用适用于超平面聚类算法的模糊C回归聚类(Fuzzy C-Recursive Model,FCRM)算法,来确定各子模型的前件参数的目标模糊隶属度函数。
104、根据各该子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到该机动目标的状态估计和协方差估计。
105、根据该机动目标的状态估计和该协方差估计,预测该机动目标的轨迹。
具体的,目标跟踪是指根据目标的过去状态和观测,精准地预测和估计目标的未来轨迹的技术,因此根据辨识得到的各子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,对机动目标的状态进行估计,得到机动目标的状态估计和协方差估计,进而进行轨迹预测。
在本发明实施例中,通过利用多个语义模糊集对目标特征信息进行模糊表示,构建T-S语义模糊多模型,从而高精度地逼近动态模型。另外,在T-S语义模糊多模型中,利用FCRM算法实现对前件参数的辨识,同时,为了实现非线性特征,引入UKF算法辨识后件参数,进而有效地对目标进行精确跟踪。
请参阅图2,图2是本发明提供的第二实施例中的机动目标跟踪方法的实现流程示意图。如图2所示,该方法主要包括以下步骤:
201、将机动目标的目标特征信息表示为多个语义模糊集,并根据多个该语义模糊集构建T-S语义模糊多模型,该T-S语义模糊多模型包括多个子模型。
202、基于无迹卡尔曼滤波算法进行后件参数辨识,确定各该子模型的目标状态和目标状态协方差。
具体的,如图3所示,步骤202具体包括:
301、基于离散动态系统,确定各该子模型的初始状态和初始观测。
其中,考虑离散动态系统,各子模型的初始状态和初始观测为:
Figure PCTCN2019112696-appb-000010
Figure PCTCN2019112696-appb-000011
上式中,
Figure PCTCN2019112696-appb-000012
Figure PCTCN2019112696-appb-000013
是已知函数,
Figure PCTCN2019112696-appb-000014
表示k时刻第i个子模型的状态,
Figure PCTCN2019112696-appb-000015
表示k时刻第i个子模型的观测,
Figure PCTCN2019112696-appb-000016
Figure PCTCN2019112696-appb-000017
时刻第i个子模型的过程噪声和测量噪声。
302、根据预设采样规则,从各该子模型的初始状态中确定多个第一采样点及各该第一采样点的加权值。
其中,根据采样规则,确定2n X+1个采样点以及相应的加权值。
Figure PCTCN2019112696-appb-000018
Figure PCTCN2019112696-appb-000019
Figure PCTCN2019112696-appb-000020
W 0=λ/(n X+λ)   j=0
W j=1/2(n X+λ)   j=1,…,n X
Figure PCTCN2019112696-appb-000021
上式中,λ表示尺度参数,可以为n X+λ≠0的任意值,
Figure PCTCN2019112696-appb-000022
Figure PCTCN2019112696-appb-000023
均方根的第j列,
Figure PCTCN2019112696-appb-000024
Figure PCTCN2019112696-appb-000025
分别表示第0列、1~n X列及n X+1~2n X+1列对应的第一采样点,W 0、W j
Figure PCTCN2019112696-appb-000026
分别表示第0列、1~n X列及n X+1~2n X+1列对应的第一采样点的加权值,n X为状态向量x的维数。
303、根据各该第一采样点及各该第一采样点的加权值进行一步预测,得到各该子模型的更新状态和更新状态协方差。
其中,通过如下公式,根据各该第一采样点及各该第一采样点的加权值进行一步预测,得到各该子模型的更新状态和更新状态协方差:
Figure PCTCN2019112696-appb-000027
Figure PCTCN2019112696-appb-000028
Figure PCTCN2019112696-appb-000029
上式中,Q表示过程噪声
Figure PCTCN2019112696-appb-000030
的协方差,
Figure PCTCN2019112696-appb-000031
表示k时刻第i个模型的更新状态,
Figure PCTCN2019112696-appb-000032
表示k时刻第i个模型的更新状态协方差。
304、根据该采样规则,从各该子模型的更新状态中确定多个第二采样点及各该第二采样点的加权值。
305、根据该初始状态、各该子模型的更新状态和更新状态协方差以及各该第二采样点及各该第二采样点的加权值,得到更新观测和更新观测方差。
其中,通过如下公式,实施步骤304和305:
Figure PCTCN2019112696-appb-000033
Figure PCTCN2019112696-appb-000034
Figure PCTCN2019112696-appb-000035
W 0=λ/(n X+λ)   j=0
W j=1/2(n X+λ)   j=1,…,n X
Figure PCTCN2019112696-appb-000036
Figure PCTCN2019112696-appb-000037
Figure PCTCN2019112696-appb-000038
Figure PCTCN2019112696-appb-000039
Figure PCTCN2019112696-appb-000040
上式中,
Figure PCTCN2019112696-appb-000041
Figure PCTCN2019112696-appb-000042
分别表示第0列、1~n X列及n X+1~2n X+1列对应的第二采样点,W 0、W j
Figure PCTCN2019112696-appb-000043
分别表示第0列、1~n X列及n X+1~2n X+1列对应的第二采样点的加权值,
Figure PCTCN2019112696-appb-000044
表示k时刻第i个模型的更新观测,
Figure PCTCN2019112696-appb-000045
Figure PCTCN2019112696-appb-000046
均表示k时刻第i个模型的更新观测方差,R表示过程噪声v k的协方差。
306、根据各该子模型的更新状态和更新状态协方差、以及该更新观测和该更新观测方差,确定各该子模型的目标状态和目标状态协方差。
其中,通过如下公式,根据各该子模型的更新状态和更新状态协方差、以及该更新观测和该更新观测方差,确定各该子模型的目标状态和目标状态协方差:
Figure PCTCN2019112696-appb-000047
Figure PCTCN2019112696-appb-000048
Figure PCTCN2019112696-appb-000049
上式中,z k表示机动目标的预设观测集,
Figure PCTCN2019112696-appb-000050
表示k时刻第i个模型的目标状态,
Figure PCTCN2019112696-appb-000051
表示k时刻第i个模型的目标状态协方差。
203、基于模糊C回归聚类算法进行前件参数辨识,确定各该子模型的前件参数的目标模糊隶属度函数。
具体的,如图4所示,步骤203具体包括如下步骤:
401、根据预设观测集、预设预测观测集和各该子模型的目标状态,构建距离测量函数。
其中,预设观测集z k表示为:
Figure PCTCN2019112696-appb-000052
预设预测观测集
Figure PCTCN2019112696-appb-000053
表示为:
Figure PCTCN2019112696-appb-000054
上式中,z k,l表示l th观测,同时
Figure PCTCN2019112696-appb-000055
表示k时刻基于模糊规则i th的预测观测。
距离测量函数表述如下:
Figure PCTCN2019112696-appb-000056
Figure PCTCN2019112696-appb-000057
上式中,
Figure PCTCN2019112696-appb-000058
为距离测量函数,
Figure PCTCN2019112696-appb-000059
表示给定目标状态
Figure PCTCN2019112696-appb-000060
的观测z k,l似然函数,
Figure PCTCN2019112696-appb-000061
表示新息协方差矩阵。
402、根据该距离测量函数和各该子模型的前件参数的模糊隶属度函数的约束条件,定义目标函数。
其中,FCRM算法的通用目标函数为:
Figure PCTCN2019112696-appb-000062
上式中,m是指权重指数,一般情况下为2,
Figure PCTCN2019112696-appb-000063
表示模糊规则i th的观测与输出之间的距离测量函数,
Figure PCTCN2019112696-appb-000064
表示第i个模型k时刻l观测的模糊隶属度函数。
各子模型的模糊隶属度函数的约束条件为:
Figure PCTCN2019112696-appb-000065
定义的目标函数为:
Figure PCTCN2019112696-appb-000066
上式中,λ k为拉格朗日乘子向量,
Figure PCTCN2019112696-appb-000067
为距离测量函数。
403、根据该目标函数对各该子模型的前件参数的模糊隶属度函数求偏导,得到各该子模型的前件参数的更新模糊隶属度函数。
其中,各该子模型的前件参数的更新模糊隶属度函数表示为:
Figure PCTCN2019112696-appb-000068
404、根据各该子模型的前件参数的更新模糊隶属度函数,得到隶属度矩阵。
其中,对观测l在时刻k上的模糊隶属度为:
Figure PCTCN2019112696-appb-000069
根据u k,l确定隶属度矩阵U。
405、将该子模型的前件参数的模糊隶属度函数设定为高斯型函数,并根据该隶属度矩阵确定各该子模型的前件参数的目标模糊隶属度函数。
其中,将该子模型的前件参数的模糊隶属度函数设定为高斯型函数:
Figure PCTCN2019112696-appb-000070
上式中,
Figure PCTCN2019112696-appb-000071
为高斯型函数的均值,
Figure PCTCN2019112696-appb-000072
为高斯型函数的均方根误差。
通过如下公式,根据该隶属度矩阵确定参数识别表达式:
Figure PCTCN2019112696-appb-000073
根据上述参数识别表达式确定该子模型的前件参数的目标模糊隶属度函数:
204、根据各该子模型的前件参数的目标模糊隶属度函数,得到各该子模型的模型概率。
具体的,通过如下公式,根据各该子模型的前件参数的目标模糊隶属度函数,得到各该子模型的模型概率:
Figure PCTCN2019112696-appb-000074
205、根据各该子模型的模型概率,计算得到各该子模型的标准化模型概率。
具体的,通过如下公式,根据各该子模型的模型概率,计算得到各该子模型的标准化模型概率:
Figure PCTCN2019112696-appb-000075
206、根据各该子模型的目标状态和标准化模型概率,得到该机动目标的状态估计。
具体的,通过如下公式,根据各该子模型的目标状态和标准化模型概率,得到机动目标的状态估计:
Figure PCTCN2019112696-appb-000076
207、根据各该子模型的目标状态协方差和标准化模型概率,得到该机动目标的协方差估计。
具体的,通过如下公式,根据各该子模型的目标状态协方差和标 准化模型概率,得到该机动目标的协方差估计:
Figure PCTCN2019112696-appb-000077
208、根据该机动目标的状态估计和该协方差估计,预测该机动目标的轨迹。
在实际应用过程中,本实施例中的算法实施框架如图5所示,主要包括三部分:基于UKF的后件参数辨识,基于FRCM的前件参数辨识以及T-S模糊模型的各子模型的融合。
示例性地,为验证本实施例中算法的跟踪性能,对一种仿真雷达机动目标跟踪问题进行分析。该问题对于防控应用具有一定的指导意义,本实施例中算法同时对比于传统的IMM算法、IMM-UKF算法,所有实验进行100次蒙特卡洛仿真。
本实施例中算法中机动目标的状态方程和测量方程如下所示:
Figure PCTCN2019112696-appb-000078
Figure PCTCN2019112696-appb-000079
其中,N f表示模糊规则的总数目,
Figure PCTCN2019112696-appb-000080
表示状态向量,x k表示目标x轴坐标,y k表示目标y轴坐标,
Figure PCTCN2019112696-appb-000081
Figure PCTCN2019112696-appb-000082
分别表示目标在x轴和y轴坐标对应的速度。假设过程噪声e k是服从零均值和均方根为σ i,e的高斯噪声,其中过程噪声协方差矩阵Q是一个4×4矩阵(Q ij=0,for i≠j,Q=diag(σ i,ei,e))。本次实验中假设观测噪声v k为非高斯分布噪声,仿真中的非高斯噪声主要有两个高斯噪声叠加产生,其中R类似于高斯噪声的协方差矩阵:
Figure PCTCN2019112696-appb-000083
初始状态x 0由目标初始位置决定x 0=[2km,0.15km/s,8km,0.26km/s] T,主要描述目标的位置和速度,假设先验概率密度函数服从高斯分布,其中x 0|0=x 0,P 0|0=diag(0.15 2,0.01,0.15 2,0.01)。
在本示例中选择新息和航向角差作为目标特征信息是因为这些信息能够有效地体现出目标的运动状态,比如新息可反映出目标运动模型是否合适,当新息较大时,说明目标运动模型不太符合当前的运动状态,根据新息对每个模型的权值进行一定的调整,从而得到一个更加准确的运动模型。新息两个模糊集(Small,Large)和航向角差三个模糊集(Negative Large,Small,Positive Large)隶属函数的均值初始值分别为
Figure PCTCN2019112696-appb-000084
初始的协方差为
Figure PCTCN2019112696-appb-000085
同时传感器的位置在坐标原点。
Figure PCTCN2019112696-appb-000086
是状态转移矩阵,它的表示方法如下:
Figure PCTCN2019112696-appb-000087
其中,转弯率ω i由T-S模糊模型决定,表1给出了对于不同输入变量
Figure PCTCN2019112696-appb-000088
和Δν k的的转弯率ω i和过程噪声标准差
Figure PCTCN2019112696-appb-000089
过程噪声标准差设置为0时是一种理想状态,在实际情况下,过程噪声设置为0存在概率是非常低的,但是为了模拟更多的运动场景,因此将其考虑进来,如果真正的实验环境噪声不为0,那么这条规则的权重则会很低。另外,如果ω i=0,模糊线性模型就变成了常速度 线性模型。
表1
Figure PCTCN2019112696-appb-000090
对于IMM和IMM-UKF都采样三个运动模型:一个常速度运动模型和两个转弯模型(转弯率分别是w=0.0325&-0.0325)。图6给出了目标运动轨迹和TS-UKF算法的估计轨迹图。如图6所示,本实施例中的算法跟踪效果和模拟仿真的轨迹基本一致,没有出现明显的跟丢现象,尤其是在目标机动的情况下,表现出很好的鲁棒性,说明该算法在非线性系统中能够高效地处理不确定信息。
图7(a-c)分别描述了目标的位置均方根误差,x轴方向的均方根误差和y轴方向的均方根误差,如图7所示,本实施例中的算法在目标机动时跟踪效果优于其他两种算法,表现出相对稳定的跟踪性能。而IMM和IMM-UKF算法在目标转弯时误差较大,主要原因是IMM算法中使用的模型集可能不够大。当目标机动时,所选择的模型集不能有效地匹配目标的运动状态。而TS-UKF算法可以根据目标的空间特征信息表示的多个语义信息构建目标运动模型,同时,利用T-S模糊模型中前件参数的隶属函数自适应地调整各规则的权重,最终目标运动模型的准确率得到更进一步的提升。并且使用无迹卡尔曼滤波算法对后件参数进行识别,提高了滤波精度。
表2给出了三种算法的均方根误差统计结果。从表中数据得出,TS-UKF的跟踪精度在位置均方根误差上比IMM和IMM-UKF分别提高 了30.32%和2.17%,体现了更精确的跟踪效果。
表2
Figure PCTCN2019112696-appb-000091
在本发明实施例中,通过利用多个语义模糊集对目标特征信息进行模糊表示,构建T-S模糊多模型,从而高精度地逼近动态模型。另外,利用T-S模糊多模型中前件参数的隶属函数自适应地调整各规则的权重,进一步提升目标运动模型的准确率,进而提高目标跟踪的精确度。同时,为了实现非线性特征,引入UKF算法辨识后件参数,提高了滤波精度。
请参阅图8,图8是本发明提供的第三实施例中的机动目标跟踪装置的结构示意图。如图8所示,该装置主要包括:
构建模块501,用于将机动目标的目标特征信息表示为多个语义模糊集,并根据多个语义模糊集构建T-S语义模糊多模型,T-S语义模糊多模型包括多个子模型。
确定模块502,用于基于无迹卡尔曼滤波算法进行后件参数辨识,确定各子模型的目标状态和目标状态协方差。
确定模块502,还用于基于模糊C回归聚类算法进行前件参数辨识,确定各子模型的前件参数的目标模糊隶属度函数。
计算模块503,用于根据各子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到机动目标的状态估计和协方差估计。
预测模块504,用于根据机动目标的状态估计和协方差估计,预测机动目标的轨迹。
进一步地,计算模块503,还用于根据各子模型的前件参数的目标模糊隶属度函数,计算得到各子模型的标准化模型概率。
计算模块503,还用于根据各子模型的目标状态和标准化模型概率,得到机动目标的状态估计。
计算模块503,还用于根据各子模型的目标状态协方差和标准化模型概率,得到机动目标的协方差估计。
进一步地,确定模块502,还用于基于离散动态系统,确定各子模型的初始状态和初始观测。
确定模块,还用于根据预设采样规则,从各子模型的初始状态中确定多个第一采样点及各第一采样点的加权值。
确定模块502,还用于根据各第一采样点及各第一采样点的加权值进行一步预测,得到各子模型的更新状态和更新状态协方差。
确定模块502,还用于根据采样规则,从各子模型的更新状态中确定多个第二采样点及各第二采样点的加权值。
确定模块502,还用于根据初始状态、各子模型的更新状态和更新状态协方差以及各第二采样点及各第二采样点的加权值,得到更新观测和更新观测方差。
确定模块502,还用于根据各子模型的更新状态和更新状态协方差、以及更新观测和更新观测方差,确定各子模型的目标状态和目标状态协方差。
进一步地,确定模块502,还用于根据预设观测集、预设预测观测集和各子模型的目标状态,构建距离测量函数。
确定模块502,还用于根据距离测量函数和各子模型的前件参数的模糊隶属度函数的约束条件,定义目标函数。
确定模块502,还用于根据目标函数对各子模型的前件参数的模糊隶属度函数求偏导,得到各子模型的前件参数的更新模糊隶属度函数。
确定模块502,还用于根据各子模型的前件参数的更新模糊隶属度函数,得到隶属度矩阵。
确定模块502,还用于将子模型的前件参数的模糊隶属度函数设定为高斯型函数,并根据隶属度矩阵确定各子模型的前件参数的目标模糊隶属度函数。
进一步地,计算模块503,还用于根据各子模型的前件参数的目标模糊隶属度函数,得到各子模型的模型概率。
计算模块503,还用于根据各子模型的模型概率,计算得到各子模型的标准化模型概率。
在本发明实施例中,通过利用多个语义模糊集对目标特征信息进行模糊表示,构建T-S模糊多模型,从而高精度地逼近动态模型。另外,利用T-S模糊多模型中前件参数的隶属函数自适应地调整各规则的权重,进一步提升目标运动模型的准确率,进而提高目标跟踪的精确度。同时,为了实现非线性特征,引入UKF算法辨识后件参数,提高了滤波精度。
在上述实施例中,对各个实施例的描述都各有侧重,某个实施例中没有详述的部分,可以参见其他实施例的相关描述。
以上为本发明所提供的机动目标跟踪方法及装置的描述,对于本领域的一般技术人员,依据本发明实施例的思想,在具体实施方式及 应用范围上均有改变之处,综上,本说明书内容不应理解为对本发明的限制。

Claims (10)

  1. 一种机动目标跟踪方法,其特征在于,所述方法包括:
    将机动目标的目标特征信息表示为多个语义模糊集,并根据多个所述语义模糊集构建T-S语义模糊多模型,所述T-S语义模糊多模型包括多个子模型;
    基于无迹卡尔曼滤波算法进行后件参数辨识,确定各所述子模型的目标状态和目标状态协方差;
    基于模糊C回归聚类算法进行前件参数辨识,确定各所述子模型的前件参数的目标模糊隶属度函数;
    根据各所述子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到所述机动目标的状态估计和协方差估计;
    根据所述机动目标的状态估计和所述协方差估计,预测所述机动目标的轨迹。
  2. 如权利要求1所述的机动目标跟踪方法,其特征在于,所述根据各所述子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到所述机动目标的状态估计和协方差估计包括:
    根据各所述子模型的前件参数的目标模糊隶属度函数,计算得到各所述子模型的标准化模型概率;
    根据各所述子模型的目标状态和标准化模型概率,得到所述机动目标的状态估计;
    根据各所述子模型的目标状态协方差和标准化模型概率,得到所述机动目标的协方差估计。
  3. 如权利要求2所述的机动目标跟踪方法,其特征在于,所述基于无迹卡尔曼滤波算法进行后件参数辨识,确定各所述子模型的目 标状态和目标状态协方差包括:
    基于离散动态系统,确定各所述子模型的初始状态和初始观测;
    根据预设采样规则,从各所述子模型的初始状态中确定多个第一采样点及各所述第一采样点的加权值;
    根据各所述第一采样点及各所述第一采样点的加权值进行一步预测,得到各所述子模型的更新状态和更新状态协方差;
    根据所述采样规则,从各所述子模型的更新状态中确定多个第二采样点及各所述第二采样点的加权值;
    根据所述初始状态、各所述子模型的更新状态和更新状态协方差以及各所述第二采样点及各所述第二采样点的加权值,得到更新观测和更新观测方差;
    根据各所述子模型的更新状态和更新状态协方差、以及所述更新观测和所述更新观测方差,确定各所述子模型的目标状态和目标状态协方差。
  4. 如权利要求3所述的机动目标跟踪方法,其特征在于,所述基于模糊C回归聚类算法进行前件参数辨识,确定各所述子模型的前件参数的目标模糊隶属度函数包括:
    根据预设观测集、预设预测观测集和各所述子模型的目标状态,构建距离测量函数;
    根据所述距离测量函数和各所述子模型的前件参数的模糊隶属度函数的约束条件,定义目标函数;
    根据所述目标函数对各所述子模型的前件参数的模糊隶属度函数求偏导,得到各所述子模型的前件参数的更新模糊隶属度函数;
    根据各所述子模型的前件参数的更新模糊隶属度函数,得到隶属 度矩阵;
    将所述子模型的前件参数的模糊隶属度函数设定为高斯型函数,并根据所述隶属度矩阵确定各所述子模型的前件参数的目标模糊隶属度函数。
  5. 如权利要求2所述的机动目标跟踪方法,其特征在于,所述根据各所述子模型的前件参数的目标模糊隶属度函数,计算得到各所述子模型的标准化模型概率包括:
    根据各所述子模型的前件参数的目标模糊隶属度函数,得到各所述子模型的模型概率;
    根据各所述子模型的模型概率,计算得到各所述子模型的标准化模型概率。
  6. 一种机动目标跟踪装置,其特征在于,所述装置包括:
    构建模块,用于将机动目标的目标特征信息表示为多个语义模糊集,并根据多个所述语义模糊集构建T-S语义模糊多模型,所述T-S语义模糊多模型包括多个子模型;
    确定模块,用于基于无迹卡尔曼滤波算法进行后件参数辨识,确定各所述子模型的目标状态和目标状态协方差;
    所述确定模块,还用于基于模糊C回归聚类算法进行前件参数辨识,确定各所述子模型的前件参数的目标模糊隶属度函数;
    计算模块,用于根据各所述子模型的目标状态、目标状态协方差以及前件参数的目标模糊隶属度函数,得到所述机动目标的状态估计和协方差估计;
    预测模块,用于根据所述机动目标的状态估计和所述协方差估计,预测所述机动目标的轨迹。
  7. 如权利要求6所述的机动目标跟踪装置,其特征在于,
    所述计算模块,还用于根据各所述子模型的前件参数的目标模糊隶属度函数,计算得到各所述子模型的标准化模型概率;
    所述计算模块,还用于根据各所述子模型的目标状态和标准化模型概率,得到所述机动目标的状态估计;
    所述计算模块,还用于根据各所述子模型的目标状态协方差和标准化模型概率,得到所述机动目标的协方差估计。
  8. 如权利要求7所述的机动目标跟踪装置,其特征在于,
    所述确定模块,还用于基于离散动态系统,确定各所述子模型的初始状态和初始观测;
    所述确定模块,还用于根据预设采样规则,从各所述子模型的初始状态中确定多个第一采样点及各所述第一采样点的加权值;
    所述确定模块,还用于根据各所述第一采样点及各所述第一采样点的加权值进行一步预测,得到各所述子模型的更新状态和更新状态协方差;
    所述确定模块,还用于根据所述采样规则,从各所述子模型的更新状态中确定多个第二采样点及各所述第二采样点的加权值;
    所述确定模块,还用于根据所述初始状态、各所述子模型的更新状态和更新状态协方差以及各所述第二采样点及各所述第二采样点的加权值,得到更新观测和更新观测方差;
    所述确定模块,还用于根据各所述子模型的更新状态和更新状态协方差、以及所述更新观测和所述更新观测方差,确定各所述子模型的目标状态和目标状态协方差。
  9. 如权利要求8所述的机动目标跟踪装置,其特征在于,
    所述确定模块,还用于根据预设观测集、预设预测观测集和各所述子模型的目标状态,构建距离测量函数;
    所述确定模块,还用于根据所述距离测量函数和各所述子模型的前件参数的模糊隶属度函数的约束条件,定义目标函数;
    所述确定模块,还用于根据所述目标函数对各所述子模型的前件参数的模糊隶属度函数求偏导,得到各所述子模型的前件参数的更新模糊隶属度函数;
    所述确定模块,还用于根据各所述子模型的前件参数的更新模糊隶属度函数,得到隶属度矩阵;
    所述确定模块,还用于将所述子模型的前件参数的模糊隶属度函数设定为高斯型函数,并根据所述隶属度矩阵确定各所述子模型的前件参数的目标模糊隶属度函数。
  10. 如权利要求7所述的机动目标跟踪装置,其特征在于,
    所述计算模块,还用于根据各所述子模型的前件参数的目标模糊隶属度函数,得到各所述子模型的模型概率;
    所述计算模块,还用于根据各所述子模型的模型概率,计算得到各所述子模型的标准化模型概率。
PCT/CN2019/112696 2019-02-28 2019-10-23 机动目标跟踪方法及装置 WO2020173105A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910150221.XA CN109990786B (zh) 2019-02-28 2019-02-28 机动目标跟踪方法及装置
CN201910150221.X 2019-02-28

Publications (1)

Publication Number Publication Date
WO2020173105A1 true WO2020173105A1 (zh) 2020-09-03

Family

ID=67130551

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/112696 WO2020173105A1 (zh) 2019-02-28 2019-10-23 机动目标跟踪方法及装置

Country Status (2)

Country Link
CN (1) CN109990786B (zh)
WO (1) WO2020173105A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112328965A (zh) * 2020-10-14 2021-02-05 南京航空航天大学 使用声矢量传感器阵列的多机动信号源doa跟踪的方法
CN112465034A (zh) * 2020-11-30 2021-03-09 中国长江电力股份有限公司 一种基于水轮发电机的t-s模糊模型的建立方法及系统
CN113326651A (zh) * 2021-05-10 2021-08-31 北京建筑大学 基于t-s模糊模型的制冷站负荷和能效比动态建模方法
CN114237053A (zh) * 2021-12-17 2022-03-25 西安交通大学 一种基于交互式多模型的捷联导引头视线角速度估计方法
CN114339595A (zh) * 2021-12-24 2022-04-12 北京理工大学重庆创新中心 一种基于多模型预测的超宽带动态反演定位方法
CN114371232A (zh) * 2021-12-22 2022-04-19 天津国科医工科技发展有限公司 基于卡尔曼滤波算法的色谱滤波方法、装置、介质、系统
CN114415157A (zh) * 2021-12-30 2022-04-29 西北工业大学 一种基于水声传感器网络的水下目标多模型跟踪方法
CN115308704A (zh) * 2022-07-17 2022-11-08 西北工业大学 基于交互式多模型和最大熵模糊聚类的多机动目标跟踪方法
CN115952930A (zh) * 2023-03-14 2023-04-11 中国人民解放军国防科技大学 一种基于imm-gmr模型的社会行为体位置预测方法
CN117251748A (zh) * 2023-10-10 2023-12-19 中国船舶集团有限公司第七〇九研究所 一种基于历史规律挖掘的航迹预测方法、设备及存储介质

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109990786B (zh) * 2019-02-28 2020-10-13 深圳大学 机动目标跟踪方法及装置
CN111291312A (zh) * 2020-02-28 2020-06-16 大连海事大学 一种基于当前统计模型的模糊自适应算法的机动目标跟踪方法
CN111474538A (zh) * 2020-04-28 2020-07-31 北京理工大学 一种基于模糊逻辑推理的目标分类方法
CN117990112A (zh) * 2024-04-03 2024-05-07 中国人民解放军海军工程大学 基于鲁棒无迹卡尔曼滤波的无人机光电平台目标定位方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102438257A (zh) * 2011-09-01 2012-05-02 哈尔滨工业大学 模糊支持向量机在话务量预测中的应用方法
CN106443661A (zh) * 2016-09-08 2017-02-22 河南科技大学 基于无迹卡尔曼滤波的机动扩展目标跟踪方法
CN108061887A (zh) * 2016-11-09 2018-05-22 北京电子工程总体研究所(航天科工防御技术研究开发中心) 一种基于模糊交互式多模型算法的临近空间目标跟踪方法
CN109325128A (zh) * 2018-12-03 2019-02-12 深圳大学 一种机动目标的跟踪方法及系统
CN109990786A (zh) * 2019-02-28 2019-07-09 深圳大学 机动目标跟踪方法及装置

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102568004A (zh) * 2011-12-22 2012-07-11 南昌航空大学 一种高机动目标跟踪算法
CN104020480B (zh) * 2014-06-17 2016-07-06 北京理工大学 一种带自适应因子的交互式多模型ukf的卫星导航方法
CN104252178B (zh) * 2014-09-12 2017-11-03 西安电子科技大学 一种基于强机动的目标跟踪方法
CN104504728B (zh) * 2014-09-16 2016-05-04 深圳大学 多机动目标跟踪方法、系统及其广义联合概率数据关联器
CN105447574B (zh) * 2015-11-10 2018-07-03 深圳大学 一种辅助截断粒子滤波方法、装置及目标跟踪方法及装置
CN105975747A (zh) * 2016-04-27 2016-09-28 渤海大学 一种基于无迹卡尔曼滤波算法的cstr模型参数辨识方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102438257A (zh) * 2011-09-01 2012-05-02 哈尔滨工业大学 模糊支持向量机在话务量预测中的应用方法
CN106443661A (zh) * 2016-09-08 2017-02-22 河南科技大学 基于无迹卡尔曼滤波的机动扩展目标跟踪方法
CN108061887A (zh) * 2016-11-09 2018-05-22 北京电子工程总体研究所(航天科工防御技术研究开发中心) 一种基于模糊交互式多模型算法的临近空间目标跟踪方法
CN109325128A (zh) * 2018-12-03 2019-02-12 深圳大学 一种机动目标的跟踪方法及系统
CN109990786A (zh) * 2019-02-28 2019-07-09 深圳大学 机动目标跟踪方法及装置

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
CHANG, CHIAWEN ET AL.: "A Novel Approach to Implement Takagi-Sugeno Fuzzy Models", IEEE TRANSACTIONS ON CYBERNETICS, vol. 47, no. 09, 16 May 2017 (2017-05-16), pages 2353 - 2361, XP011658337, DOI: 20200125184404A *
GAO, ZILIN; XIONG, JIANG; PAN, YONG; LI, HONGBING; LUO, WEIMIN: "Adaptive Sliding Tracking Control and Simulation for a Class of Nonlinear Systems Based on T-S Model", JOURNAL OF CHONGQING NORMAL UNIVERSITY (NATURAL SCIENCE), vol. 33, no. 05, 30 September 2016 (2016-09-30), pages 128 - 132, XP009522841, ISSN: 1672-6693, DOI: 10.11721/cqnuj20160503 *
MIZUMOTO, I. ET AL.: "Adaptive Output Feedback Based Output Tracking Control for Uncertain Nonlinear Systems via T-S Fuzzy Model", 2017 IEEE 56TH ANNUAL CONFERENCE ON DECISION AND CONTROL (CDC), 15 December 2017 (2017-12-15), XP033303947, DOI: 20200125184240A *
WANG, XIAOLI; LI, LIANGQUN; XIE, WEIXIN: "T-S Fuzzy Multiple Model Target Tracking Algorithm with UKF Parameter Identification", JOURNAL OF SIGNAL PROCESSING, vol. 35, no. 03, 25 March 2019 (2019-03-25), pages 361 - 368, XP009522842, ISSN: 1003-0530, DOI: 10.16798/j.issn.1003-0530.2019.03.006 *

Cited By (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112328965A (zh) * 2020-10-14 2021-02-05 南京航空航天大学 使用声矢量传感器阵列的多机动信号源doa跟踪的方法
CN112328965B (zh) * 2020-10-14 2024-02-20 南京航空航天大学 使用声矢量传感器阵列的多机动信号源doa跟踪的方法
CN112465034A (zh) * 2020-11-30 2021-03-09 中国长江电力股份有限公司 一种基于水轮发电机的t-s模糊模型的建立方法及系统
CN112465034B (zh) * 2020-11-30 2023-06-27 中国长江电力股份有限公司 一种基于水轮发电机的t-s模糊模型的建立方法及系统
CN113326651A (zh) * 2021-05-10 2021-08-31 北京建筑大学 基于t-s模糊模型的制冷站负荷和能效比动态建模方法
CN113326651B (zh) * 2021-05-10 2023-05-23 北京建筑大学 基于t-s模糊模型的制冷站负荷和能效比动态建模方法
CN114237053A (zh) * 2021-12-17 2022-03-25 西安交通大学 一种基于交互式多模型的捷联导引头视线角速度估计方法
CN114237053B (zh) * 2021-12-17 2023-07-04 西安交通大学 一种基于交互式多模型的捷联导引头视线角速度估计方法
CN114371232A (zh) * 2021-12-22 2022-04-19 天津国科医工科技发展有限公司 基于卡尔曼滤波算法的色谱滤波方法、装置、介质、系统
CN114371232B (zh) * 2021-12-22 2024-03-22 天津国科医工科技发展有限公司 基于卡尔曼滤波算法的色谱滤波方法、装置、介质、系统
CN114339595A (zh) * 2021-12-24 2022-04-12 北京理工大学重庆创新中心 一种基于多模型预测的超宽带动态反演定位方法
CN114339595B (zh) * 2021-12-24 2023-12-01 北京理工大学重庆创新中心 一种基于多模型预测的超宽带动态反演定位方法
CN114415157A (zh) * 2021-12-30 2022-04-29 西北工业大学 一种基于水声传感器网络的水下目标多模型跟踪方法
CN115308704A (zh) * 2022-07-17 2022-11-08 西北工业大学 基于交互式多模型和最大熵模糊聚类的多机动目标跟踪方法
CN115308704B (zh) * 2022-07-17 2024-04-26 西北工业大学 基于交互式多模型和最大熵模糊聚类的多机动目标跟踪方法
CN115952930A (zh) * 2023-03-14 2023-04-11 中国人民解放军国防科技大学 一种基于imm-gmr模型的社会行为体位置预测方法
CN115952930B (zh) * 2023-03-14 2023-08-22 中国人民解放军国防科技大学 一种基于imm-gmr模型的社会行为体位置预测方法
CN117251748A (zh) * 2023-10-10 2023-12-19 中国船舶集团有限公司第七〇九研究所 一种基于历史规律挖掘的航迹预测方法、设备及存储介质
CN117251748B (zh) * 2023-10-10 2024-04-19 中国船舶集团有限公司第七〇九研究所 一种基于历史规律挖掘的航迹预测方法、设备及存储介质

Also Published As

Publication number Publication date
CN109990786A (zh) 2019-07-09
CN109990786B (zh) 2020-10-13

Similar Documents

Publication Publication Date Title
WO2020173105A1 (zh) 机动目标跟踪方法及装置
Toussaint et al. Probabilistic inference for solving discrete and continuous state Markov Decision Processes
CN106054170B (zh) 一种约束条件下的机动目标跟踪方法
Huang et al. Robust student’s t-based stochastic cubature filter for nonlinear systems with heavy-tailed process and measurement noises
WO2017124299A1 (zh) 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN109325128B (zh) 一种机动目标的跟踪方法及系统
CN107462882B (zh) 一种适用于闪烁噪声的多机动目标跟踪方法及系统
Wang et al. Variational Bayesian IMM-filter for JMSs with unknown noise covariances
CN103326903A (zh) 基于隐马尔科夫的Internet网络时延预测方法
Feng et al. Overview of nonlinear Bayesian filtering algorithm
CN111291471B (zh) 一种基于l1正则无迹变换的约束多模型滤波方法
CN110111367A (zh) 模糊模型粒子滤波方法、装置、设备及存储介质
CN111711432B (zh) 一种基于ukf和pf混合滤波的目标跟踪算法
Wang et al. Interacting ts fuzzy particle filter algorithm for transfer probability matrix of adaptive online estimation model
CN115204212A (zh) 一种基于stm-pmbm滤波算法的多目标跟踪方法
CN111798494A (zh) 广义相关熵准则下的机动目标鲁棒跟踪方法
Zhao et al. Adaptive non-linear joint probabilistic data association for vehicle target tracking
CN105424043A (zh) 一种基于判断机动的运动状态估计方法
CN111291319A (zh) 一种应用于非高斯噪声环境下的移动机器人状态估计方法
CN113452349B (zh) 一种基于贝叶斯序贯重要性积分的卡尔曼滤波方法
Hao et al. Mode separability-based state estimation for uncertain constrained dynamic systems
Wu et al. State estimation for Markovian Jump Linear Systems with bounded disturbances
CN111262556B (zh) 一种同时估计未知高斯测量噪声统计量的多目标跟踪方法
CN105373805A (zh) 一种基于最大熵准则的多传感器机动目标跟踪方法
CN114445456B (zh) 基于部分模型的数据驱动智能机动目标跟踪方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19917009

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 21/01/2022

122 Ep: pct application non-entry in european phase

Ref document number: 19917009

Country of ref document: EP

Kind code of ref document: A1