WO2018098926A1 - 一种适用于闪烁噪声的多目标跟踪方法及系统 - Google Patents

一种适用于闪烁噪声的多目标跟踪方法及系统 Download PDF

Info

Publication number
WO2018098926A1
WO2018098926A1 PCT/CN2017/076776 CN2017076776W WO2018098926A1 WO 2018098926 A1 WO2018098926 A1 WO 2018098926A1 CN 2017076776 W CN2017076776 W CN 2017076776W WO 2018098926 A1 WO2018098926 A1 WO 2018098926A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
distribution
current time
contract
existence probability
Prior art date
Application number
PCT/CN2017/076776
Other languages
English (en)
French (fr)
Inventor
刘宗香
邹燕妮
谢维信
李良群
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 WO2018098926A1 publication Critical patent/WO2018098926A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/66Radar-tracking systems; Analogous systems

Definitions

  • the invention belongs to the field of multi-sensor information fusion technology, and in particular relates to a multi-target tracking method and system suitable for flicker noise.
  • the edge-distributed Bayesian filter is an effective method for multi-target tracking under clutter and noise, but the filter is only suitable for the case where the noise is Gaussian noise, and not for the case where the noise is flicker noise. How to solve the multi-target tracking problem under flicker noise is a key technical problem that needs to be explored and solved.
  • the technical problem to be solved by the present invention is to provide a multi-target tracking method and tracking system suitable for flicker noise, aiming at solving the tracking problem of nonlinear moving targets under flicker noise.
  • the present invention is implemented in such a manner that a multi-target tracking method suitable for scintillation noise includes:
  • the heuristic method is used to generate the shape parameters and scale parameters of the gamma distribution, and then the predicted contract distribution and prediction of each target at the current moment are obtained. Probability of existence;
  • the variation data Bayesian method is used to sequentially process the current time measurement data, and the update contract distribution and the update existence probability of each target at the current time are obtained;
  • the target having a probability greater than the second threshold, the contract distribution of the extracted target is used as the output of the current time, and the average of the output contract distribution is used as the state estimate of the current time target.
  • the invention also provides a multi-target tracking system suitable for scintillation noise, comprising:
  • the prediction module is configured to generate a shape parameter and a scale parameter of the gamma distribution according to the contract distribution and the existence probability of each target at the previous moment and the time difference between the current moment and the previous moment, thereby obtaining the target of each moment at the current moment.
  • Forecast contract distribution and forecast existence probability
  • An update module configured to sequentially process the current time measurement data according to the predicted contract distribution and the predicted existence probability of each target at the current time, and obtain the update contract distribution and update of each target at the current time by using the variational Bayesian method. Probability of existence;
  • a generating module configured to generate a contract distribution of the new target by using the current time measurement data, and specify an existence probability for the new target, and distribute the contract distribution and the existence probability of the new target to the update contract distribution of the current time respectively Update the existence probability to merge, and obtain the contract distribution and existence probability of each target at the current time;
  • An extracting module configured to cut off a target whose probability of existence is less than a first threshold from each target in the current time, and use the contract distribution and the existence probability of the remaining target as a next recursive input of the filter, after the clipping
  • the remaining targets extract the target whose probability of existence is greater than the second threshold, and the contract distribution of the extracted target is used as the output of the current time, and the average of the output contract distribution is used as the state estimation of the current time target.
  • the present invention has the beneficial effects that: in the embodiment of the present invention, the flicker noise is modeled by the t distribution, the closed expression is obtained by the variational Bayesian method, and the multivariate is approximated by the product of the edge distribution of each subvariate.
  • the contract distribution makes multivariate contract estimation into iterative estimation of the edge distribution of each sub-variable, which effectively solves the tracking problem of nonlinear moving targets under flicker noise and improves the tracking accuracy of multi-target.
  • FIG. 1 is a flowchart of a multi-target tracking method suitable for flicker noise according to an embodiment of the present invention
  • 2 is a measurement data of 70 scan cycles of a sensor according to an embodiment of the present invention
  • FIG. 3 is a result of processing by a multi-target tracking method according to an embodiment of the present invention under flicker noise;
  • Figure 4 is a result of processing according to the multi-target tracking method of the UK-PHD filter under flicker noise
  • FIG. 5 is a schematic diagram showing an average OFAC distance obtained by 100 experiments in a multi-target tracking method and a UK-PHD filtering method according to an embodiment of the present invention
  • FIG. 6 is a schematic structural diagram of a multi-target tracking system suitable for flicker noise according to an embodiment of the present invention.
  • the embodiment of the present invention predicts the contract distribution and the existence probability of each target of the new measurement data received at the current time according to the contract distribution and the existence probability of each target of the measurement data received at the previous moment; and utilizes the predicted contract distribution and the existence probability according to the predicted contract distribution and the existence probability
  • the variational Bayesian method sequentially processes each measurement at the current time to obtain the updated contract distribution and existence probability of each target; respectively, the updated contract distribution and existence probability are combined with the contract distribution and existence probability of the new target to generate each current moment.
  • the contract distribution and the probability of existence of the target so that the embodiment of the invention can well solve the tracking problem of the nonlinear moving target under the flicker noise.
  • an embodiment of the present invention provides a multi-target tracking method suitable for flicker noise as shown in FIG. 1, which includes:
  • S103 Generate a contract distribution of the new target by using the current time measurement data, and specify an existence probability for the new target, and update the contract distribution and the existence probability of the new target with the current time update contract distribution and update existence probability respectively. Consolidate to obtain the contract distribution and existence probability of each target at the current time;
  • step S101 to k-1 represents the previous time, k represents the current time, the time T k-1 represents a previous time, t k represents a measuring time of the current time, the current time is subject to noise ⁇
  • the t distribution of the dimension, the probability density function measured by the current time is represented by S(z k ; H k x k , R k , r k ), where H k x k represents the mean of the t distribution, and R k represents the precision matrix, r k represents the degree of freedom of the t distribution, and
  • the multivariate contract distribution of target i at the previous moment is The existence probability of the target i is ⁇ i,k-1 , where N represents a Gaussian distribution, g represents a gamma distribution, and x i,k-1 represents the state vector of the i-th contract distribution at the previous moment, m i,k- 1 represents the mean of the Gaussian distribution in the i-th contract distribution at the previous
  • ⁇ i,k-1 represent the shape parameters of the gamma distribution in the i-th contract distribution at the previous moment
  • ⁇ i,k-1 represent the scale parameter of the gamma distribution in the i-th contract distribution at the previous moment
  • the predicted existence probability of each target at the current moment is ⁇ i,k
  • k-1 P s,k (t k -t k-1 ) ⁇ i,k-1 ;
  • Sigma point x i,0 m i,k-1
  • Sigma point weight l 1,..., ⁇
  • Survival probability for the target ⁇ i,k
  • k-1 ⁇ ⁇ ⁇ i,k-1 is the shape parameter of the gamma distribution in the i-th contract distribution at the current time
  • the contract distribution of each target when updating with the jth measurement is among them, Represents the gamma function, tr represents the trace of the matrix, Means the mean vector, Representing a covariance matrix, Represents the filter gain; Sigma point
  • the shape parameter of the gamma distribution is
  • the scale parameter of the gamma distribution is H k is the observation matrix, R k is the observed noise variance matrix, P D,k is the detection probability of the target, ⁇ c,k is the clutter density, I is the unit matrix, y j,k is the jth received at the current time Measurement data, the superscript T is represented as a transpose of a matrix or a vector, and ⁇ is the total dimension of the state vector;
  • step S103 it is set that the update contract distribution of each target at the current time is
  • the target of nonlinear motion in a two-dimensional space [-1000 m, 1000 m] ⁇ [-1000 m, 1000 m] is considered.
  • the observation equation of the radar is:
  • the simulated observation data of one experiment is shown in Fig. 2.
  • the simulation data of FIG. 2 is processed by the UK-PHD filter under the existing flicker noise in the embodiment of the present invention, and the average OFAP (Optimal Subpattern Assignment) distance is obtained by 100 Monte Carlo experiments. Shown.
  • the existing multi-target tracking method based on flicker noise can more accurately track the nonlinear moving target with uncertain correlation and detection uncertainty under scintillation noise.
  • reliable target state estimation its OSPA distance is smaller than the existing SPAC distance obtained by this method.
  • the present invention also provides an embodiment as shown in FIG. 6, a multi-target tracking system, comprising:
  • the prediction module 601 is configured to generate a shape parameter and a scale parameter of the gamma distribution according to the contract distribution and the existence probability of each target at the previous moment and the time difference between the current time and the previous moment, thereby obtaining each target at the current moment. Predicted contract distribution and predicted existence probability;
  • the update module 602 is configured to sequentially process the measurement data of the current time according to the predicted contract distribution and the predicted existence probability of each target at the current time, and obtain the update contract distribution and update existence of each target at the current time by using the variational Bayesian method. Probability
  • the generating module 603 is configured to generate a contract distribution of the new target by using the measurement data of the current time, and specify an existence probability for the new target, and distribute the contract distribution and the existence probability of the new target to the updated contract distribution of the current time respectively. And updating the existence probability to merge, and obtaining the contract distribution and the existence probability of each target at the current time;
  • the extracting module 604 is configured to: cut off the target whose probability of existence is less than the first threshold from each target in the current time, and use the contract distribution and the existence probability of the remaining target as the input of the filter next recursion, from the cut.
  • the remaining target is extracted from the target whose probability of existence is greater than the second threshold, and the contract distribution of the extracted target is used as the output of the current time, and the average value of the output contract distribution is used as the state estimation of the current time target.
  • prediction module 601 is specifically configured to:
  • the current time is represented by k-1, k represents the current time, t k-1 represents the time of the previous time, t k represents the time of the current time, and the measurement noise of the current time obeys the t distribution of the dimension, with S ( z k ; H k x k , R k , r k ) represent the probability density function measured at the current time, where H k x k represents the mean of the t distribution, R k represents the precision matrix, and r k represents the degree of freedom of the t distribution
  • the multivariate contract distribution of target i at the previous moment is The existence probability of the target i is ⁇ i,k-1 , where N represents a Gaussian distribution, g represents a gamma distribution, and x i,k-1 represents the state vector of the i-th contract distribution at the previous moment, m i,k- 1 represents the mean of the Gaussian distribution in the i-th contract distribution at the previous moment, and P
  • ⁇ i,k-1 represent the shape parameters of the gamma distribution in the i-th contract distribution at the previous moment
  • ⁇ i,k-1 represent the scale parameter of the gamma distribution in the i-th contract distribution at the previous moment
  • the predicted existence probability of each target at the current moment is ⁇ i,k
  • k-1 P s,k (t k -t k-1 ) ⁇ i,k-1 ;
  • Sigma point x i,0 m i,k-1
  • Sigma point weight l 1,..., ⁇
  • Survival probability for the target ⁇ i,k
  • k-1 ⁇ ⁇ ⁇ i,k-1 is the shape parameter of the gamma distribution in the i-th contract distribution at the current time
  • update module 602 is specifically configured to:
  • the first to the Mth measurement data are sequentially processed sequentially by using the variational Bayesian method
  • the contract distribution of each target when updating with the jth measurement is among them, Represents the gamma function, tr represents the trace of the matrix, Means the mean vector, Representing a covariance matrix, Represents the filter gain; Sigma point
  • the shape parameter of the gamma distribution is
  • the scale parameter of the gamma distribution is H k is the observation matrix, R k is the observed noise variance matrix, P D,k is the detection probability of the target, ⁇ c,k is the clutter density, I is the unit matrix, y j,k is the jth received at the current time Measurement data, the superscript T is represented as a transpose of a matrix or a vector, and ⁇ is the total dimension of the state vector;
  • the contract distribution and the existence probability of each target after processing the Mth measurement data are respectively used as the update contract distribution of each target at the current time, thereby obtaining the update contract distribution of each target at the current time.
  • the generating module 603 is specifically configured to:

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种适用于闪烁噪声的多目标跟踪方法与系统,多目标跟踪方法包括:预测步骤、更新步骤、生成步骤以及提取步骤。在保证数据处理实时性的同时,有效地解决了闪烁噪声下非线性运动目标的跟踪问题。

Description

一种适用于闪烁噪声的多目标跟踪方法及系统 技术领域
本发明属于多传感器信息融合技术领域,尤其涉及一种适用于闪烁噪声的多目标跟踪方法及系统。
背景技术
边缘分布贝叶斯滤波器是适用于杂波和噪声下多目标跟踪的有效方法,但滤波器仅适用于噪声为高斯噪声的情况,不适用于噪声为闪烁噪声的情况。如何解决闪烁噪声下多目标跟踪问题是需要探索和解决的关键技术问题。
发明内容
本发明所要解决的技术问题在于提供一种适用于闪烁噪声的多目标跟踪方法与跟踪系统,旨在解决闪烁噪声下非线性运动目标的跟踪问题。
本发明是这样实现的,一种适用于闪烁噪声的多目标跟踪方法,包括:
根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率;
根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率;
利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率;
从所述当前时刻各个目标中裁减掉存在概率小于第一阈值的目标,并将裁减后余下目标的合同分布和存在概率作为滤波器下一次递归的输入,从所述裁减后余下的目标中提取存在概率大于第二阈值的目标,所提取出的目标的合同分布作为所述当前时刻的输出,所输出的合同分布的均值作为当前时刻目标的状态估计。
本发明还提供了一种适用于闪烁噪声的多目标跟踪系统,包括:
预测模块,用于根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率;
更新模块,用于根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新 存在概率;
生成模块,用于利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率;
提取模块,用于从所述当前时刻各个目标中裁减掉存在概率小于第一阈值的目标,并将裁减后余下目标的合同分布和存在概率作为滤波器下一次递归的输入,从所述裁减后余下的目标中提取存在概率大于第二阈值的目标,所提取出的目标的合同分布作为所述当前时刻的输出,所输出的合同分布的均值作为当前时刻目标的状态估计。
本发明与现有技术相比,有益效果在于:本发明实施例通过用t分布建模闪烁噪声,利用变分贝叶斯方法获得闭合表达式,用各个分变量边缘分布的乘积来逼近多变量的合同分布,从而将多变量的合同估计化为对各个分变量边缘分布的迭代估计,有效解决了闪烁噪声下非线性运动目标的跟踪问题,提高多目标的跟踪精度。
附图说明
图1是本发明实施例提供的一种适用于闪烁噪声的多目标跟踪方法的流程图;
图2是本发明实施例提供的传感器70个扫描周期的测量数据;
图3是在闪烁噪声下按照本发明实施例提供的多目标跟踪方法处理得到的结果;
图4是在闪烁噪声下根据UK-PHD滤波器的多目标跟踪方法处理得到的结果;
图5是按照本发明实施例提供多目标跟踪方法与UK-PHD滤波方法在经过100次实验得到的平均OSPA距离示意图;
图6是本发明实施例提供一种适用于闪烁噪声的多目标跟踪系统的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本发明进行进一步详细说明。应当理解,此处所描述的具体实施例仅仅用以解释本发明,并不用于限定本发明。
本发明实施例根据前一时刻接收的测量数据的各个目标的合同分布及存在概率预测当前时刻接收的新的测量数据的各个目标的合同分布及存在概率;根据预测的合同分布及存在概率,利用变分贝叶斯方法序贯处理当前时刻的各测量得到各个目标的更新合同分布及存在概率;分别将更新的合同分布及存在概率与新目标的合同分布及存在概率进行合并,生成当前时刻各个目标的合同分布及存在概率,这样使得本发明实施例能够很好的解决闪烁噪声下非线性运动目标的跟踪问题。
基于上述原理,本发明实施例提供了如图1所示的一种适用于闪烁噪声的多目标跟踪方法,包括:
S101,根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间 差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率;
S102,根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率;
S103,利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率;
S104,从所述当前时刻各个目标中裁减掉存在概率小于第一阈值的目标,并将裁减后余下目标的合同分布和存在概率作为滤波器下一次递归的输入,从所述裁减后余下的目标中提取存在概率大于第二阈值的目标,所提取出的目标的合同分布作为所述当前时刻的输出,所输出的合同分布的均值作为当前时刻目标的状态估计。
具体地,在步骤S101中,以k-1表示前一时刻,k表示当前时刻,tk-1表示前一时刻的时间,tk表示当前时刻的时间,所述当前时刻的测量噪声服从ξ维的t分布,以S(zk;Hkxk,Rk,rk)表示所述当前时刻测量的概率密度函数,其中Hkxk表示t分布的均值,Rk表示精度矩阵,rk表示t分布的自由度,且
Figure PCTCN2017076776-appb-000001
前一时刻目标i的多变量合同分布为
Figure PCTCN2017076776-appb-000002
目标i的存在概率为ρi,k-1,其中,N表示高斯分布,g表示伽玛分布,xi,k-1表示前一时刻第i个合同分布的状态向量,mi,k-1表示前一时刻第i个合同分布中高斯分布的均值,Pi,k-1表示前一时刻第i个合同分布中高斯分布的方差,
Figure PCTCN2017076776-appb-000003
表示Rk的对角线元素,
Figure PCTCN2017076776-appb-000004
和γi,k-1表示前一时刻第i个合同分布中伽玛分布的形状参数,
Figure PCTCN2017076776-appb-000005
和ηi,k-1表示前一时刻第i个合同分布中伽玛分布的尺度参数,ξ为状态向量的维数,i=1,2,…,Nk-1,Nk-1为前一时刻目标的总数;
根据前一时刻各个目标的合同分布和存在概率、当前时刻与前一时刻的时间差,得到当前时刻各个目标的预测合同分布
Figure PCTCN2017076776-appb-000006
当前时刻各个目标的预测存在概率为ρi,k|k-1=Ps,k(tk-tk-1i,k-1;其中,i=1,2,…,Nk-1
Figure PCTCN2017076776-appb-000007
为当前时刻第i个合同分布中高斯分布的均值,
Figure PCTCN2017076776-appb-000008
为当前时刻第i个合同分布中高斯分布的方差,Sigma点xi,0=mi,k-1
Figure PCTCN2017076776-appb-000009
Sigma点的权重
Figure PCTCN2017076776-appb-000010
Figure PCTCN2017076776-appb-000011
l=1,…,ξ,
Figure PCTCN2017076776-appb-000012
为目标的幸存概率,
Figure PCTCN2017076776-appb-000013
γi,k|k-1=ργγi,k-1为当前时刻第i个合同分布中伽玛分布的形状参数,
Figure PCTCN2017076776-appb-000014
ηi,k|k-1=ρηηi,k-1为当前时刻第i个合同分布中伽玛分布的尺度参数,f为非线性函数,Qk-1为所述接收时刻的过程噪声方差矩阵,上标T表示矩阵或向量的转置,T为采样周期,δ为已知的常数,ρα,ρβ,ργ,ρη为传播因子,取值范围为(0,1],rk表示自由度,为已知常数,ξ为状态向量的维数,k为一尺度参数。
在步骤S102中,设当前时刻接收到的观测集为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到测量数据总数,则所述根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率,包括:
S1021,以当前时刻各个目标的预测合同分布和预测存在概率作为当前时刻各个目标的初始合同分布和初始存在概率,即初始合同分布取为
Figure PCTCN2017076776-appb-000015
初始存在概率取为
Figure PCTCN2017076776-appb-000016
其中i=1,2,…,Nk-1
Figure PCTCN2017076776-appb-000017
Figure PCTCN2017076776-appb-000018
S1022,利用变分贝叶斯方法对第1个至第M个测量数据依次进行序贯处理;
设第j个测量数据处理前各个目标的合同分布及存在概率分别为
Figure PCTCN2017076776-appb-000019
Figure PCTCN2017076776-appb-000020
其中,i=1,2,…,Nk-1,1≤j≤M;由
Figure PCTCN2017076776-appb-000021
Figure PCTCN2017076776-appb-000022
求得用第j个测量更新时各个目标的存在概率为
Figure PCTCN2017076776-appb-000023
其中
Figure PCTCN2017076776-appb-000024
求得用第j个测量更新时各个目标的合同分布为
Figure PCTCN2017076776-appb-000025
其中,
Figure PCTCN2017076776-appb-000026
表示伽玛函数,tr表示矩阵的迹,
Figure PCTCN2017076776-appb-000027
表示均值向量,
Figure PCTCN2017076776-appb-000028
表示协方差矩阵,
Figure PCTCN2017076776-appb-000029
表示滤波器增益;其中
Figure PCTCN2017076776-appb-000030
Figure PCTCN2017076776-appb-000031
Figure PCTCN2017076776-appb-000032
Sigma点
Figure PCTCN2017076776-appb-000033
Figure PCTCN2017076776-appb-000034
伽玛分布的形状参数为
Figure PCTCN2017076776-appb-000035
伽玛分布的尺度参数为
Figure PCTCN2017076776-appb-000036
Figure PCTCN2017076776-appb-000037
Figure PCTCN2017076776-appb-000038
Hk为观测矩阵,Rk为观测噪声方差矩阵,PD,k为目标的检测概率,λc,k为杂波密度,I表示单位矩阵,yj,k为当前时刻接收到的第j个测量数据,上标T表示为矩阵或向量的转置,ξ为状态向量的总维数;
Figure PCTCN2017076776-appb-000039
则第j个测量数据处理后目标i的合同分布为
Figure PCTCN2017076776-appb-000040
Figure PCTCN2017076776-appb-000041
其存在概率为
Figure PCTCN2017076776-appb-000042
其中
Figure PCTCN2017076776-appb-000043
Figure PCTCN2017076776-appb-000044
Figure PCTCN2017076776-appb-000045
则第j个测量数据处理后目标i的合同分布为
Figure PCTCN2017076776-appb-000046
Figure PCTCN2017076776-appb-000047
其存在概率为
Figure PCTCN2017076776-appb-000048
其中
Figure PCTCN2017076776-appb-000049
Figure PCTCN2017076776-appb-000050
S1023,第M个测量数据处理后的各个目标的合同分布及存在概率分别为
Figure PCTCN2017076776-appb-000051
Figure PCTCN2017076776-appb-000052
其中i=1,2,…,Nk-1;将第M个测量数据处理后各个目标的合同分布及存在概率分别作为当前时刻各个目标的更新合同分布,由此得到当前时刻各个目标的更新合同分布为
Figure PCTCN2017076776-appb-000053
Figure PCTCN2017076776-appb-000054
及当前时刻各个目标的更新存在概率
Figure PCTCN2017076776-appb-000055
其中i=1,…,Nk-1
Figure PCTCN2017076776-appb-000056
Figure PCTCN2017076776-appb-000057
在步骤S103中,设所述当前时刻各个目标的更新合同分布为
Figure PCTCN2017076776-appb-000058
各个目标的存在概率为ρi,k;其中i=1,2,…,Nk-1,利用所述当前时刻的M个测量数据生成当前时刻新生目标的合同分布为
Figure PCTCN2017076776-appb-000059
并指定当前时刻各新生目标的存在概率为
Figure PCTCN2017076776-appb-000060
其中j=1,2,…,M,
Figure PCTCN2017076776-appb-000061
为给定的第j个新生目标的合同分布中高斯分布的协方差,
Figure PCTCN2017076776-appb-000062
为第j个新生目标的合同分布中高斯分布的均值,
Figure PCTCN2017076776-appb-000063
由所述当前时刻第j个测量数据yj,k=[xj,k yj,k]T产生,并且
Figure PCTCN2017076776-appb-000064
Figure PCTCN2017076776-appb-000065
为所述当前时刻第j个新生目标的合同分布中伽玛分布的形状参数,
Figure PCTCN2017076776-appb-000066
Figure PCTCN2017076776-appb-000067
为所述当前时刻第j个新生目标的合同分布中伽玛分布的尺度参数。
将所述当前时刻各个目标的更新合同分布与所述当前时刻新生目标的合同分布进行合并,得到当前时刻各个目标的合同分布为
Figure PCTCN2017076776-appb-000068
将所述当前时刻各个目标的存在概率与所述当前时刻新生目标的存在概率进行合并,得到所述当前时刻各个目标的存在概率为
Figure PCTCN2017076776-appb-000069
其中Nk=Nk-1+M。
下面结合图2至图5对本实施例进行进一步地解释:
在本实施例中,考虑二维空间[-1000m,1000m]×[-1000m,1000m]中非线性运动的目标。目标的状态由位置、速度和转弯率构成,表示为x=[x &x y &y ω],其中x和y分别表示位置分量,&x和&y分别表示速度分量,ω表示转弯率,上标T表示向量的转置,状 态转移矩阵为
Figure PCTCN2017076776-appb-000070
过程噪声方差矩阵为
Figure PCTCN2017076776-appb-000071
Δtk=tk-tk-1为当前时刻与前一时刻的时间差,σv和σw为过程噪声标准差;雷达的观测方程为:
Figure PCTCN2017076776-appb-000072
观测噪声方差矩阵
Figure PCTCN2017076776-appb-000073
σr和σθ为观测噪声标准差,测量噪声vk假设服从rk=10的t分布。
为了产生仿真数据,取幸存概率pS,k=1.0、检测概率pD,k=0.9、过程噪声的标准差σv,=1ms-2,σw,=0.1rads-2和观测噪声的标准差σr=2m,σθ=0.0003rad。一次实验的仿真观测数据如图2所示。为了处理仿真数据,将本发明实施例与闪烁噪声下的无迹卡尔曼高斯混合概率假设密度滤波器的相关参数设置为pS,k=1.0、pD,k=0.9、σv=1ms-2、σw=0.1rads-2,σr=2m,σθ=0.0003rad、第一阈值为10-3、第二阈值为0.5,传播因子ρα=ρβ=ργ=ρη=0.75,伽玛分布形状参数初始值
Figure PCTCN2017076776-appb-000074
伽玛分布尺度参数初始值
Figure PCTCN2017076776-appb-000075
Figure PCTCN2017076776-appb-000076
闪烁噪声下UK-PHD滤波器新生目标的权重wγ=0.1、本发明实施例的新生目标的存在概率ργ=0.1、新产生目标的协方差
Figure PCTCN2017076776-appb-000077
图3和图4为对比滤波器与本发明实施例提供的多目标跟踪方法产生的结果。将本发明实施例与现有的闪烁噪声下的UK-PHD滤波器对图2的仿真数据进行处理,100次Monte Carlo实验得到平均OSPA(Optimal Subpattern Assignment,最优亚模式分配)距离如图5所示。将现有的基于闪烁噪声的UK-PHD滤波器与本发明相比,本发明的多目标跟踪方法在闪烁噪声下对于关联不确定、检测不确定的非线性运动目标的跟踪可获得更为精确和可靠的目标状态估计、其OSPA距离比现有的这种方法得到的OSPA距离要小。
本发明还提供了如图6所示的实施例,一种多目标跟踪系统,包括:
预测模块601,用于根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率;
更新模块602,用于根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率;
生成模块603,用于利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率;
提取模块604,用于从所述当前时刻各个目标中裁减掉存在概率小于第一阈值的目标,并将裁减后余下目标的合同分布和存在概率作为滤波器下一次递归的输入,从所述裁减后余下的目标中提取存在概率大于第二阈值的目标,所提取出的目标的合同分布作为所述当前时刻的输出,所输出的合同分布的均值作为当前时刻目标的状态估计。
进一步地,预测模块601具体用于:
以k-1表示前一时刻,k表示当前时刻,tk-1表示前一时刻的时间,tk表示当前时刻的时间,所述当前时刻的测量噪声服从ξ维的t分布,以S(zk;Hkxk,Rk,rk)表示所述当前时刻测量的概率密度函数,其中Hkxk表示t分布的均值,Rk表示精度矩阵,rk表示t分布的自由度,且
Figure PCTCN2017076776-appb-000078
前一时刻目标i的多变量合同分布为
Figure PCTCN2017076776-appb-000079
目标i的存在概率为ρi,k-1,其中,N表示高斯分布,g表示伽玛分布,xi,k-1表示前一时刻第i个合同分布的状态向量,mi,k-1表示前一时刻第i个合同分布中高斯分布的均值,Pi,k-1表示前一时刻第i个合同分布中高斯分布的方差,
Figure PCTCN2017076776-appb-000080
表示Rk的对角线元素,
Figure PCTCN2017076776-appb-000081
和γi,k-1表示前一时刻第i个合同分布中伽玛分布的形状参数,
Figure PCTCN2017076776-appb-000082
和ηi,k-1表示前一时刻第i个合同分布中伽玛分布的尺度参数,ξ为状态向量的维数,i=1,2,…,Nk-1,Nk-1为前一时刻目标的总数;
根据前一时刻各个目标的合同分布和存在概率、当前时刻与前一时刻的时间差,得到当前时刻各个目标的预测合同分布
Figure PCTCN2017076776-appb-000083
当前时刻各个目标的预测存在概率为ρi,k|k-1=Ps,k(tk-tk-1i,k-1;其中,i=1,2,…,Nk-1
Figure PCTCN2017076776-appb-000084
为当前时刻第i个合同分布中高斯分布的均值,
Figure PCTCN2017076776-appb-000085
为当前时刻第i个合同分布中高斯分布的方差,Sigma点xi,0=mi,k-1
Figure PCTCN2017076776-appb-000086
Sigma点的权重
Figure PCTCN2017076776-appb-000087
Figure PCTCN2017076776-appb-000088
l=1,…,ξ,
Figure PCTCN2017076776-appb-000089
为目标的幸存概率,
Figure PCTCN2017076776-appb-000090
γi,k|k-1=ργγi,k-1为当前时刻第i个合同分布中伽玛分布的形状参数,
Figure PCTCN2017076776-appb-000091
ηi,k|k-1=ρηηi,k-1为当前时刻第i个合同分布中伽玛分布的尺度参数,f为非线性函数,Qk-1为所述接收时刻的过程噪声方差矩阵,上标T表示矩阵或向量的转置,T为采样周期,δ为已知的常数,ρα,ρβ,ργ,ρη为传播因子,取值范围为(0,1],rk表示自由度,为已知常数,ξ为状态向量的维数,k为一尺度参数。
进一步地,更新模块602具体用于:
以当前时刻各个目标的预测合同分布和预测存在概率作为当前时刻各个目标的初始合同分布和初始存在概率,即初始合同分布取为
Figure PCTCN2017076776-appb-000092
初始存在概率取为
Figure PCTCN2017076776-appb-000093
其中i=1,2,…,Nk-1
Figure PCTCN2017076776-appb-000094
Figure PCTCN2017076776-appb-000095
利用变分贝叶斯方法对第1个至第M个测量数据依次进行序贯处理;
设第j个测量数据处理前各个目标的合同分布及存在概率分别为
Figure PCTCN2017076776-appb-000096
Figure PCTCN2017076776-appb-000097
其中,i=1,2,…,Nk-1,1≤j≤M;由
Figure PCTCN2017076776-appb-000098
Figure PCTCN2017076776-appb-000099
求得用第j个测量更新时各个目标的存在概率为
Figure PCTCN2017076776-appb-000100
其中
Figure PCTCN2017076776-appb-000101
求得用第j个测量更新时各个目标的合同分布为
Figure PCTCN2017076776-appb-000102
其中,
Figure PCTCN2017076776-appb-000103
表示伽玛函数,tr表示矩阵的迹,
Figure PCTCN2017076776-appb-000104
表示均值向量,
Figure PCTCN2017076776-appb-000105
表示协方差矩阵,
Figure PCTCN2017076776-appb-000106
表示滤波器增益;其中
Figure PCTCN2017076776-appb-000107
Figure PCTCN2017076776-appb-000108
Figure PCTCN2017076776-appb-000109
Sigma点
Figure PCTCN2017076776-appb-000110
Figure PCTCN2017076776-appb-000111
伽玛分布的形状参数为
Figure PCTCN2017076776-appb-000112
伽玛分布的尺度参数为
Figure PCTCN2017076776-appb-000113
Figure PCTCN2017076776-appb-000114
Figure PCTCN2017076776-appb-000115
Hk为观测矩阵,Rk为观测噪声方差矩阵,PD,k为目标的检测概率,λc,k为杂波密度,I表示单位矩阵,yj,k为当前时刻接收到的第j个测量数据,上标T表示为矩阵或向量的转置,ξ为状态向量的总维数;
Figure PCTCN2017076776-appb-000116
则第j个测量数据处理后目标i的合同分布为
Figure PCTCN2017076776-appb-000117
Figure PCTCN2017076776-appb-000118
其存在概率为
Figure PCTCN2017076776-appb-000119
其中
Figure PCTCN2017076776-appb-000120
Figure PCTCN2017076776-appb-000121
Figure PCTCN2017076776-appb-000122
则第j个测量数据处理后目标i的合同分布为
Figure PCTCN2017076776-appb-000123
Figure PCTCN2017076776-appb-000124
其存在概率为
Figure PCTCN2017076776-appb-000125
其中
Figure PCTCN2017076776-appb-000126
Figure PCTCN2017076776-appb-000127
第M个测量数据处理后的各个目标的合同分布及存在概率分别为
Figure PCTCN2017076776-appb-000128
Figure PCTCN2017076776-appb-000129
其中i=1,2,…,Nk-1
将第M个测量数据处理后各个目标的合同分布及存在概率分别作为当前时刻各个目标的更新合同分布,由此得到当前时刻各个目标的更新合同分布为
Figure PCTCN2017076776-appb-000130
Figure PCTCN2017076776-appb-000131
及当前时刻各个目标的更新存在概率
Figure PCTCN2017076776-appb-000132
其中i=1,…,Nk-1
Figure PCTCN2017076776-appb-000133
Figure PCTCN2017076776-appb-000134
进一步地,生成模块603具体用于:
设所述当前时刻各个目标的更新合同分布为
Figure PCTCN2017076776-appb-000135
各个目标的存在概率为ρi,k;其中i=1,2,…,Nk-1,利用所述当前时刻的M个测量数据生成当前时刻新生目标的合同分布为
Figure PCTCN2017076776-appb-000136
并指定当前时刻各新生目标的存在概率为
Figure PCTCN2017076776-appb-000137
其中j=1,2,…,M,
Figure PCTCN2017076776-appb-000138
为给定的第j个新生目标的合同分布中高斯分布的协方差,
Figure PCTCN2017076776-appb-000139
为第j个新生目标的合同分布中高斯分布的均值,
Figure PCTCN2017076776-appb-000140
由所述当前时刻第j个测量数据yj,k=[xj,k yj,k]T产生,并且
Figure PCTCN2017076776-appb-000141
Figure PCTCN2017076776-appb-000142
为所述当前时刻第j个新生目标的合同分布中伽玛分布的形状参数,
Figure PCTCN2017076776-appb-000143
Figure PCTCN2017076776-appb-000144
为所述当前时刻第j个新生目标的合同分布中伽玛分布的尺度参数。
将所述当前时刻各个目标的更新合同分布与所述当前时刻新生目标的合同分布进行合并,得到当前时刻各个目标的合同分布为
Figure PCTCN2017076776-appb-000145
将所述当前时刻各个目标的存在概率与所述当前时刻新生目标的存在概率进行合并,得到所述当前时刻各个目标的存在概率为
Figure PCTCN2017076776-appb-000146
其中Nk=Nk-1+M。
以上所述仅为本发明的较佳实施例而已,并不用以限制本发明,凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等,均应包含在本发明的保护范围之内。

Claims (8)

  1. 一种适用于闪烁噪声的多目标跟踪方法,其特征在于,包括:
    根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率;
    根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率;
    利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率;
    从所述当前时刻各个目标中裁减掉存在概率小于第一阈值的目标,并将裁减后余下目标的合同分布和存在概率作为滤波器下一次递归的输入,从所述裁减后余下的目标中提取存在概率大于第二阈值的目标,所提取出的目标的合同分布作为所述当前时刻的输出,所输出的合同分布的均值作为当前时刻目标的状态估计。
  2. 如权利要求1所述的多目标跟踪方法,其特征在于,所述根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率,包括:
    以k-1表示前一时刻,k表示当前时刻,tk-1表示前一时刻的时间,tk表示当前时刻的时间,所述当前时刻的测量噪声服从ξ维的t分布,以S(zk;Hkxk,Rk,rk)表示所述当前时刻测量的概率密度函数,其中Hkxk表示t分布的均值,Rk表示精度矩阵,rk表示t分布的自由度,且
    Figure PCTCN2017076776-appb-100001
    前一时刻目标i的多变量合同分布为
    Figure PCTCN2017076776-appb-100002
    目标i的存在概率为ρi,k-1,其中,N表示高斯分布,g表示伽玛分布,xi,k-1表示前一时刻第i个合同分布的状态向量,mi,k-1表示前一时刻第i个合同分布中高斯分布的均值,Pi,k-1表示前一时刻第i个合同分布中高斯分布的方差,
    Figure PCTCN2017076776-appb-100003
    表示Rk的对角线元素,
    Figure PCTCN2017076776-appb-100004
    和γi,k-1表示前一时刻第i个合同分布中伽玛分布的形状参数,
    Figure PCTCN2017076776-appb-100005
    和ηi,k-1表示前一时刻第i个合同分布中伽玛分布的尺度参数,ξ为状态向量的维数,i=1,2,...,Nk-1,Nk-1为前一时刻目标的总数;
    根据前一时刻各个目标的合同分布和存在概率、当前时刻与前一时刻的时间差,得到当前时刻各个目标的预测合同分布
    Figure PCTCN2017076776-appb-100006
    当前时刻各个目标的预测存在 概率为ρi,k|k-1=Ps,k(tk-tk-1i,k-1;其中,i=1,2,...,Nk-1
    Figure PCTCN2017076776-appb-100007
    为当前时刻第i个合同分布中高斯分布的均值,
    Figure PCTCN2017076776-appb-100008
    为当前时刻第i个合同分布中高斯分布的方差,Sigma点xi,0=mi,k.1
    Figure PCTCN2017076776-appb-100009
    Sigma点的权重
    Figure PCTCN2017076776-appb-100010
    Figure PCTCN2017076776-appb-100011
    l=1,...,ξ,
    Figure PCTCN2017076776-appb-100012
    为目标的幸存概率,
    Figure PCTCN2017076776-appb-100013
    γi,k|k-1=ργγi,k-1为当前时刻第i个合同分布中伽玛分布的形状参数,
    Figure PCTCN2017076776-appb-100014
    ηi,k|k-1=ρηηi,k-1为当前时刻第i个合同分布中伽玛分布的尺度参数,f为非线性函数,Qk-1为所述接收时刻的过程噪声方差矩阵,上标T表示矩阵或向量的转置,T为采样周期,δ为已知的常数,ρα,ρβ,ργ,ρη为传播因子,取值范围为(0,1],rk表示自由度,为已知常数,ξ为状态向量的维数,k为一尺度参数。
  3. 如权利要求2所述的多目标跟踪方法,其特征在于,设当前时刻接收到的观测集为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到测量数据总数,则所述根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率,包括:
    以当前时刻各个目标的预测合同分布和预测存在概率作为当前时刻各个目标的初始合同分布和初始存在概率,即初始合同分布取为
    Figure PCTCN2017076776-appb-100015
    Figure PCTCN2017076776-appb-100016
    初始存在概率取为
    Figure PCTCN2017076776-appb-100017
    其中i=1,2,...,Nk-1
    Figure PCTCN2017076776-appb-100018
    Figure PCTCN2017076776-appb-100019
    利用变分贝叶斯方法对第1个至第M个测量数据依次进行序贯处理;
    设第j个测量数据处理前各个目标的合同分布及存在概率分别为
    Figure PCTCN2017076776-appb-100020
    Figure PCTCN2017076776-appb-100021
    其中,i=1,2,...,Nk-1,1≤j≤M;由
    Figure PCTCN2017076776-appb-100022
    Figure PCTCN2017076776-appb-100023
    求得用第j个测量更新时各个目标的存在概率为
    Figure PCTCN2017076776-appb-100024
    其中
    Figure PCTCN2017076776-appb-100025
    求得用第j个测量更新时各个目标的合同分布为
    Figure PCTCN2017076776-appb-100026
    其中,
    Figure PCTCN2017076776-appb-100027
    表示伽玛函数,tr表示矩阵的迹,
    Figure PCTCN2017076776-appb-100028
    表示均值向量,
    Figure PCTCN2017076776-appb-100029
    表示协方差矩阵,
    Figure PCTCN2017076776-appb-100030
    表示滤波器增益;其中
    Figure PCTCN2017076776-appb-100031
    Figure PCTCN2017076776-appb-100032
    Figure PCTCN2017076776-appb-100033
    Sigma点
    Figure PCTCN2017076776-appb-100034
    Figure PCTCN2017076776-appb-100035
    伽玛分布的形状参数为
    Figure PCTCN2017076776-appb-100036
    伽玛分布的尺度参数为
    Figure PCTCN2017076776-appb-100037
    Figure PCTCN2017076776-appb-100038
    Figure PCTCN2017076776-appb-100039
    Hk为观测矩阵,Rk为观测噪声方差矩阵,PD,k为目标的检测概率,λc,k为杂波密度,I表示单位矩阵,yj,k为当前时刻接收到的第j个测量数据,上标T表示为矩阵或向量的转置,ξ为状态向量的总维数;
    Figure PCTCN2017076776-appb-100040
    则第j个测量数据处理后目标i的合同分布为
    Figure PCTCN2017076776-appb-100041
    Figure PCTCN2017076776-appb-100042
    其存在概率为
    Figure PCTCN2017076776-appb-100043
    其中
    Figure PCTCN2017076776-appb-100044
    Figure PCTCN2017076776-appb-100045
    Figure PCTCN2017076776-appb-100046
    则第j个测量数据处理后目标i的合同分布为
    Figure PCTCN2017076776-appb-100047
    Figure PCTCN2017076776-appb-100048
    其存在概率为
    Figure PCTCN2017076776-appb-100049
    其中
    Figure PCTCN2017076776-appb-100050
    Figure PCTCN2017076776-appb-100051
    第M个测量数据处理后的各个目标的合同分布及存在概率分别为
    Figure PCTCN2017076776-appb-100052
    Figure PCTCN2017076776-appb-100053
    其中i=1,2,...,Nk-1
    将第M个测量数据处理后各个目标的合同分布及存在概率分别作为当前时刻各个目标的更新合同分布,由此得到当前时刻各个目标的更新合同分布为
    Figure PCTCN2017076776-appb-100054
    Figure PCTCN2017076776-appb-100055
    及当前时刻各个目标的更新存在概率
    Figure PCTCN2017076776-appb-100056
    其中i=1,…,Nk-1
    Figure PCTCN2017076776-appb-100057
    Figure PCTCN2017076776-appb-100058
  4. 如权利要求1所述的多目标跟踪方法,其特征在于,设所述当前时刻接收到的观测集为yk=(y1,k,…,yM,k),其中,M为所述当前时刻接收到所述新的测量数据总数,所述利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率包括:
    设所述当前时刻各个目标的更新合同分布为
    Figure PCTCN2017076776-appb-100059
    各个目标的存在概率为ρi,k;其中i=1,2,...,Nk-1,利用所述当前时刻的M个测量数据生成当前时刻新生目标的合同分布为
    Figure PCTCN2017076776-appb-100060
    并指定当前时刻各新生目标的存在概率为
    Figure PCTCN2017076776-appb-100061
    其中j=1,2,...,M,
    Figure PCTCN2017076776-appb-100062
    为给定的第j个新生目标的合同分布中高斯分布的协方差,
    Figure PCTCN2017076776-appb-100063
    为第j个新生目标的合同分布中高斯分布的均值,
    Figure PCTCN2017076776-appb-100064
    由所述当前时刻第j个测量数据yj,k=[xj,k yj,k]T产生,并且
    Figure PCTCN2017076776-appb-100065
    Figure PCTCN2017076776-appb-100066
    为所述当前时刻第j个新生目标的合同分布中伽玛分布的形状参数,
    Figure PCTCN2017076776-appb-100067
    Figure PCTCN2017076776-appb-100068
    为所述当前时刻第j个新生目标的合同分布中伽玛分布的尺度参数;
    将所述当前时刻各个目标的更新合同分布与所述当前时刻新生目标的合同分布进行合并,得到当前时刻各个目标的合同分布为
    Figure PCTCN2017076776-appb-100069
    将所述当前时刻各个目标的存在概率与所述当前时刻新生目标的存在概率进行合并,得到所 述当前时刻各个目标的存在概率为
    Figure PCTCN2017076776-appb-100070
    其中Nk=Nk-1+M。
  5. 一种适用于闪烁噪声的多目标跟踪系统,其特征在于,包括:
    预测模块,用于根据前一时刻各个目标的合同分布和存在概率以及当前时刻与前一时刻的时间差,采用启发式的方法产生伽马分布的形状参数和尺度参数,进而得到当前时刻各个目标的预测合同分布和预测存在概率;
    更新模块,用于根据所述当前时刻各个目标的预测合同分布和预测存在概率,利用变分贝叶斯方法序贯处理当前时刻的测量数据,得到当前时刻各个目标的更新合同分布和更新存在概率;
    生成模块,用于利用当前时刻的测量数据生成新生目标的合同分布,并为所述新生目标指定存在概率,将所述新生目标的合同分布及存在概率分别与所述当前时刻的更新合同分布及更新存在概率进行合并,得到当前时刻各个目标的合同分布和存在概率;
    提取模块,用于从所述当前时刻各个目标中裁减掉存在概率小于第一阈值的目标,并将裁减后余下目标的合同分布和存在概率作为滤波器下一次递归的输入,从所述裁减后余下的目标中提取存在概率大于第二阈值的目标,所提取出的目标的合同分布作为所述当前时刻的输出,所输出的合同分布的均值作为当前时刻目标的状态估计。
  6. 如权利要求5所述的多目标跟踪系统,其特征在于,所述预测模块具体用于:
    以k-1表示前一时刻,k表示当前时刻,tk-1表示前一时刻的时间,tk表示当前时刻的时间,所述当前时刻的测量噪声服从ξ维的t分布,以S(zk;Hkxk,Rk,rk)表示所述当前时刻测量的概率密度函数,其中Hkxk表示t分布的均值,Rk表示精度矩阵,rk表示t分布的自由度,且
    Figure PCTCN2017076776-appb-100071
    前一时刻目标i的多变量合同分布为
    Figure PCTCN2017076776-appb-100072
    目标i的存在概率为ρi,k-1,其中,N表示高斯分布,g表示伽玛分布,xi,k-1表示前一时刻第i个合同分布的状态向量,mi,k-1表示前一时刻第i个合同分布中高斯分布的均值,Pi,k-1表示前一时刻第i个合同分布中高斯分布的方差,
    Figure PCTCN2017076776-appb-100073
    表示Rk的对角线元素,
    Figure PCTCN2017076776-appb-100074
    和γi,k-1表示前一时刻第i个合同分布中伽玛分布的形状参数,
    Figure PCTCN2017076776-appb-100075
    和ηi,k-1表示前一时刻第i个合同分布中伽玛分布的尺度参数,ξ为状态向量的维数,i=1,2,...,Nk-1,Nk-1为前一时刻目标的总数;
    根据前一时刻各个目标的合同分布和存在概率、当前时刻与前一时刻的时间差,得到当前时刻各个目标的预测合同分布
    Figure PCTCN2017076776-appb-100076
    当前时刻各个目标的预测存在概率为ρi,k|k-1=Ps,k(tk-tk-1i,k-1;其中,i=1,2,...,Nk-1
    Figure PCTCN2017076776-appb-100077
    为当前时 刻第i个合同分布中高斯分布的均值,
    Figure PCTCN2017076776-appb-100078
    为当前时刻第i个合同分布中高斯分布的方差,Sigma点xi,0=mi,k-1
    Figure PCTCN2017076776-appb-100079
    Sigma点的权重
    Figure PCTCN2017076776-appb-100080
    Figure PCTCN2017076776-appb-100081
    l=1,...,ξ,
    Figure PCTCN2017076776-appb-100082
    为目标的幸存概率,
    Figure PCTCN2017076776-appb-100083
    γi,k|k-1=ργγi,k-1为当前时刻第i个合同分布中伽玛分布的形状参数,
    Figure PCTCN2017076776-appb-100084
    ηi,k|k-1=ρηηi,k-1为当前时刻第i个合同分布中伽玛分布的尺度参数,f为非线性函数,Qk-1为所述接收时刻的过程噪声方差矩阵,上标T表示矩阵或向量的转置,T为采样周期,δ为已知的常数,ρα,ρβ,ργ,ρη为传播因子,取值范围为(0,1],rk表示自由度,为已知常数,ξ为状态向量的维数,k为一尺度参数。
  7. 如权利要求6所述的多目标跟踪方法,其特征在于,所述更新模块具体用于:
    以当前时刻各个目标的预测合同分布和预测存在概率作为当前时刻各个目标的初始合同分布和初始存在概率,即初始合同分布取为
    Figure PCTCN2017076776-appb-100085
    Figure PCTCN2017076776-appb-100086
    初始存在概率取为
    Figure PCTCN2017076776-appb-100087
    其中i=1,2,...,Nk-1
    Figure PCTCN2017076776-appb-100088
    Figure PCTCN2017076776-appb-100089
    利用变分贝叶斯方法对第1个至第M个测量数据依次进行序贯处理;
    设第j个测量数据处理前各个目标的合同分布及存在概率分别为
    Figure PCTCN2017076776-appb-100090
    Figure PCTCN2017076776-appb-100091
    其中,i=1,2,...,Nk-1,1≤j≤M;由
    Figure PCTCN2017076776-appb-100092
    Figure PCTCN2017076776-appb-100093
    求得用第j个测量更新时各个目标的存在概率为
    Figure PCTCN2017076776-appb-100094
    其中
    Figure PCTCN2017076776-appb-100095
    求得用第j个测量更新时各个目标的合同分布为
    Figure PCTCN2017076776-appb-100096
    其中,
    Figure PCTCN2017076776-appb-100097
    表示伽玛函数,tr表示矩阵的迹,
    Figure PCTCN2017076776-appb-100098
    表示均值向量,
    Figure PCTCN2017076776-appb-100099
    表示协方差矩阵,
    Figure PCTCN2017076776-appb-100100
    表示滤波器增益;其中
    Figure PCTCN2017076776-appb-100101
    Figure PCTCN2017076776-appb-100102
    Figure PCTCN2017076776-appb-100103
    Sigma点
    Figure PCTCN2017076776-appb-100104
    Figure PCTCN2017076776-appb-100105
    伽玛分布的形状参数为
    Figure PCTCN2017076776-appb-100106
    伽玛分布的尺度参数为
    Figure PCTCN2017076776-appb-100107
    Figure PCTCN2017076776-appb-100108
    Figure PCTCN2017076776-appb-100109
    Hk为观测矩阵,Rk为观测噪声方差矩阵,PD,k为目标的检测概率,λc,k为杂波密度,I表示单位矩阵,yj,k为当前时刻接收到的第j个测量数据,上标T表示为矩阵或向量的转置,ξ为状态向量的总维数;
    Figure PCTCN2017076776-appb-100110
    则第j个测量数据处理后目标i的合同分布为
    Figure PCTCN2017076776-appb-100111
    Figure PCTCN2017076776-appb-100112
    其存在概率为
    Figure PCTCN2017076776-appb-100113
    其中
    Figure PCTCN2017076776-appb-100114
    Figure PCTCN2017076776-appb-100115
    Figure PCTCN2017076776-appb-100116
    则第j个测量数据处理后目标i的合同分布为
    Figure PCTCN2017076776-appb-100117
    Figure PCTCN2017076776-appb-100118
    其存在概率为
    Figure PCTCN2017076776-appb-100119
    其中
    Figure PCTCN2017076776-appb-100120
    Figure PCTCN2017076776-appb-100121
    第M个测量数据处理后的各个目标的合同分布及存在概率分别为
    Figure PCTCN2017076776-appb-100122
    Figure PCTCN2017076776-appb-100123
    其中i=1,2,...,Nk-1
    将第M个测量数据处理后各个目标的合同分布及存在概率分别作为当前时刻各个目标的更新合同分布,由此得到当前时刻各个目标的更新合同分布为
    Figure PCTCN2017076776-appb-100124
    Figure PCTCN2017076776-appb-100125
    及当前时刻各个目标的更新存在概率
    Figure PCTCN2017076776-appb-100126
    其中i=1,…,Nk-1
    Figure PCTCN2017076776-appb-100127
    Figure PCTCN2017076776-appb-100128
  8. 如权利要求5所述的多目标跟踪系统,其特征在于,所述更新模块还用于:
    设所述当前时刻各个目标的更新合同分布为
    Figure PCTCN2017076776-appb-100129
    各个目标的存在概率为ρi,k;其中i=1,2,...,Nk-1,利用所述当前时刻的M个测量数据生成当前时刻新生目标的合同分布为
    Figure PCTCN2017076776-appb-100130
    并指定当前时刻各新生目标的存在概率为
    Figure PCTCN2017076776-appb-100131
    其中j=1,2,...,M,
    Figure PCTCN2017076776-appb-100132
    为给定的第j个新生目标的合同分布中高斯分布的协方差,
    Figure PCTCN2017076776-appb-100133
    为第j个新生目标的合同分布中高斯分布的均值,
    Figure PCTCN2017076776-appb-100134
    由所述当前时刻第j个测量数据yj,k=[xj,k yj,k]T产生,并且
    Figure PCTCN2017076776-appb-100135
    Figure PCTCN2017076776-appb-100136
    为所述当前时刻第j个新生目标的合同分布中伽玛分布的形状参数,
    Figure PCTCN2017076776-appb-100137
    Figure PCTCN2017076776-appb-100138
    为所述当前时刻第j个新生目标的合同分布中伽玛分布的尺度参数;
    将所述当前时刻各个目标的更新合同分布与所述当前时刻新生目标的合同分布进行合并,得到当前时刻各个目标的合同分布为
    Figure PCTCN2017076776-appb-100139
    将所述当前时刻各个目标的存在概率与所述当前时刻新生目标的存在概率进行合并,得到所述当前时刻各个目标的存在概率为
    Figure PCTCN2017076776-appb-100140
    其中Nk=Nk-1+M。
PCT/CN2017/076776 2016-11-29 2017-03-15 一种适用于闪烁噪声的多目标跟踪方法及系统 WO2018098926A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611090305.1A CN106772353B (zh) 2016-11-29 2016-11-29 一种适用于闪烁噪声的多目标跟踪方法及系统
CN201611090305.1 2016-11-29

Publications (1)

Publication Number Publication Date
WO2018098926A1 true WO2018098926A1 (zh) 2018-06-07

Family

ID=58915525

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/076776 WO2018098926A1 (zh) 2016-11-29 2017-03-15 一种适用于闪烁噪声的多目标跟踪方法及系统

Country Status (2)

Country Link
CN (1) CN106772353B (zh)
WO (1) WO2018098926A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110309950A (zh) * 2019-05-16 2019-10-08 中国人民解放军海军工程大学 模块化贮存模式下初始贮存方案的优化方法及装置
CN110376582A (zh) * 2019-01-24 2019-10-25 西安电子科技大学 自适应gm-phd的机动目标跟踪方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107462882B (zh) * 2017-09-08 2020-06-02 深圳大学 一种适用于闪烁噪声的多机动目标跟踪方法及系统
CN109474892B (zh) * 2018-11-05 2020-07-10 浙江工商大学 基于信息形式的强鲁棒传感器网络目标跟踪方法
CN109800721B (zh) * 2019-01-24 2020-10-23 深圳大学 一种多目标跟踪方法及系统
CN110390684A (zh) * 2019-07-16 2019-10-29 深圳大学 一种闪烁噪声下的多目标跟踪方法及系统
CN110503071B (zh) * 2019-08-29 2021-11-23 电子科技大学 基于变分贝叶斯标签多伯努利叠加模型的多目标跟踪方法
CN111325776A (zh) * 2020-02-19 2020-06-23 江苏理工学院 一种基于变分贝叶斯t分布卡尔曼滤波的phd多目标跟踪方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130006576A1 (en) * 2010-03-15 2013-01-03 Bae Systems Plc Target tracking
CN103424742A (zh) * 2013-07-24 2013-12-04 深圳大学 一种序贯处理测量数据的目标跟踪方法与目标跟踪系统
CN103778320A (zh) * 2013-12-30 2014-05-07 杭州电子科技大学 一种基于变分贝叶斯多传感器量化融合目标跟踪方法
CN103902829A (zh) * 2014-04-11 2014-07-02 深圳大学 传递边缘分布及存在概率的目标跟踪方法与目标跟踪系统
CN105719312A (zh) * 2016-01-19 2016-06-29 深圳大学 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104318059B (zh) * 2014-09-24 2018-02-02 深圳大学 用于非线性高斯系统的目标跟踪方法和跟踪系统
CN104867163A (zh) * 2015-05-28 2015-08-26 深圳大学 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130006576A1 (en) * 2010-03-15 2013-01-03 Bae Systems Plc Target tracking
CN103424742A (zh) * 2013-07-24 2013-12-04 深圳大学 一种序贯处理测量数据的目标跟踪方法与目标跟踪系统
CN103778320A (zh) * 2013-12-30 2014-05-07 杭州电子科技大学 一种基于变分贝叶斯多传感器量化融合目标跟踪方法
CN103902829A (zh) * 2014-04-11 2014-07-02 深圳大学 传递边缘分布及存在概率的目标跟踪方法与目标跟踪系统
CN105719312A (zh) * 2016-01-19 2016-06-29 深圳大学 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110376582A (zh) * 2019-01-24 2019-10-25 西安电子科技大学 自适应gm-phd的机动目标跟踪方法
CN110376582B (zh) * 2019-01-24 2022-10-04 西安电子科技大学 自适应gm-phd的机动目标跟踪方法
CN110309950A (zh) * 2019-05-16 2019-10-08 中国人民解放军海军工程大学 模块化贮存模式下初始贮存方案的优化方法及装置

Also Published As

Publication number Publication date
CN106772353B (zh) 2019-04-05
CN106772353A (zh) 2017-05-31

Similar Documents

Publication Publication Date Title
WO2018098926A1 (zh) 一种适用于闪烁噪声的多目标跟踪方法及系统
WO2017124299A1 (zh) 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN107462882B (zh) 一种适用于闪烁噪声的多机动目标跟踪方法及系统
CN101975575B (zh) 基于粒子滤波的被动传感器多目标跟踪方法
CN105719312B (zh) 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN104318059B (zh) 用于非线性高斯系统的目标跟踪方法和跟踪系统
Vo et al. A random finite set conjugate prior and application to multi-target tracking
US10935653B2 (en) Multi-target tracking method and tracking system applicable to clutter environment
Li et al. PHD filter for multi-target tracking with glint noise
WO2018010099A1 (zh) 一种用于跟踪转弯机动目标的方法及其系统
CN104730511A (zh) 星凸模型下的势概率假设密度多扩展目标跟踪方法
CN104794735A (zh) 基于变分贝叶斯期望最大化的扩展目标跟踪方法
Li et al. State estimation for jump Markov linear systems by variational Bayesian approximation
CN111722214A (zh) 雷达多目标跟踪phd实现方法
CN106168943A (zh) 一种用于跟踪转弯机动目标的方法及其系统
Li et al. Distributed fusion with PHD filter for multi-target tracking in asynchronous radar system
Jeon et al. Distributed asynchronous multiple sensor fusion with nonlinear multiple models
Gehly et al. An AEGIS-CPHD filter to maintain custody of GEO space objects with limited tracking data
Úbeda-Medina et al. Multiple target tracking in the fully adaptive radar framework
CN111340853B (zh) 基于ospa迭代的多传感器gmphd自适应融合方法
WO2016000487A1 (zh) 一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统
Nandakumaran et al. Improved multitarget tracking using probability hypothesis density smoothing
Liu et al. Sequential measurement-driven multi-target Bayesian filter
Wang et al. Particle Filter with Hybrid Proposal Distribution for Nonlinear State Estimation.
Fu et al. Observation bootstrapping CDK-GMPHD filter based on consensus fusion strategy

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: 17877130

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 25/09/2019)

122 Ep: pct application non-entry in european phase

Ref document number: 17877130

Country of ref document: EP

Kind code of ref document: A1