WO2016000487A1 - 一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统 - Google Patents

一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统 Download PDF

Info

Publication number
WO2016000487A1
WO2016000487A1 PCT/CN2015/078026 CN2015078026W WO2016000487A1 WO 2016000487 A1 WO2016000487 A1 WO 2016000487A1 CN 2015078026 W CN2015078026 W CN 2015078026W WO 2016000487 A1 WO2016000487 A1 WO 2016000487A1
Authority
WO
WIPO (PCT)
Prior art keywords
dirac
target
moment
term
current
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2015/078026
Other languages
English (en)
French (fr)
Inventor
刘宗香
谢维信
李良群
李丽娟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
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 Shenzhen University filed Critical Shenzhen University
Publication of WO2016000487A1 publication Critical patent/WO2016000487A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16ZINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
    • G16Z99/00Subject matter not provided for in other main groups of this subclass

Definitions

  • the invention belongs to the field of multi-sensor information fusion technology, and in particular relates to a target tracking method and a tracking system based on a variable coefficient ⁇ - ⁇ filter.
  • the probability hypothesis density filtering method is a new method to solve the target detection and tracking. Its biggest advantage is that it reduces the integral operation in Bayesian filtering method, and can give the instantaneous target number estimation. It has been widely used in multi-target detection, location and tracking.
  • the probability hypothesis density filtering method is a data processing method that transmits the first moment of the joint posterior distribution. Due to the integral operation involved, the probability hypothesis density filter is usually implemented by particle filtering or Gaussian mixture method. Both the particle filter method and the Gaussian mixture method require that the distribution characteristics of the process noise during the target motion and the distribution characteristics of the observed noise during the sensor observation process are known. When the distribution characteristics of process noise are unknown, the existing probability hypothesis density filter and its approximate implementation method are difficult to work. Probabilistic Hypothesis The multi-target tracking problem of density filter is an important technical problem that needs to be explored and solved.
  • the technical problem to be solved by the present invention is to provide a target tracking method and tracking system based on a variable coefficient ⁇ - ⁇ filter, aiming at improving the new target detection capability and multi-target tracking accuracy of the ⁇ - ⁇ filter.
  • the present invention is implemented as follows:
  • a target tracking method based on a variable coefficient ⁇ - ⁇ filter comprising the following steps:
  • Step S1 predicting the target moment and the Dirac term of the target existing at the previous moment according to the target moment and the Dirac term at the previous moment, and assigning the corresponding target moment and the Dirac term to the new target at the current moment. ;
  • Step S2 determining, according to the predicted target moment at the previous moment, the target moment at the current moment, the Dirac term, the target moment of the new target at the current time, and the Dirac term and the current position measurement.
  • New Dirac item
  • Step S3 performing the reduction and merging of the updated Dirac items to obtain the target moment and the Dirac term at the current moment; and using the target moment of the current moment and the Dirac term as the recursive input of the next moment;
  • Step S4 extracting a Dirac term whose weight coefficient is greater than the first threshold as the output of the current time according to the target moment and the Dirac term at the current moment, and setting the target state value in the output Dirac term as a target state at the current time. value.
  • step S1 the previous moment is represented by k-1, and the current moment is represented by k;
  • the target moment of the previous moment is Where ⁇ represents the Dirac distribution and x represents the state of the target;
  • the step S1 specifically includes the following steps:
  • the Dirac item at the current moment i 1, 2,..., J k-1 ; among them,
  • the weighting factor representing the ith predicted Dirac term Indicates the target state value of the ith predicted Dirac term, The time at which the new Dirac term corresponding to the ith predicted Dirac term is generated, and Where F k-1 is the state transition matrix, p sk is the probability of the target surviving; the predicted target moment p k
  • the weighting factor representing the new Dirac term Indicates the target state value of the new Dirac item, Represents the time of the birth of the new Dirac item, J ⁇ k is the number of the new Dirac items, i is the index number, and t represents the time of the current time.
  • step S2 is specifically:
  • Dirac items at the current moment based on the predicted existing target at the previous moment And the target moment p k
  • the time generated for the updated Dirac term, n k is the number of current time position measurements
  • the position measurement of the current time includes the position measurement generated by the current time target and the position measurement generated by the current time clutter
  • i and j is the index number, and when j ⁇ n k ,
  • t represents the time of the current time
  • K i is the gain matrix
  • z j represents the jth position measurement in the n k position measurement at the current time
  • H k is the observation matrix
  • R j is the variance matrix of the observed noise
  • c is The amplification factor
  • R 0 is the covariance set for detecting the new target
  • p Dk is the detection probability of the target
  • K i takes the gain matrix of the variable coefficient ⁇ - ⁇ filter, and Where T is the time difference between the current time and the previous time, and ⁇ k and ⁇ k are two time-varying coefficients, and
  • step S3 is specifically:
  • the target state value for the qth Dirac item The time produced for the new Dirac term corresponding to the qth Dirac term, and J k
  • Dirac item updated from the current moment q 1,...,J k
  • the method of merging multiple Dirac items is: Where L is the set formed by the merged Dirac term, and the superscripts T and b respectively represent the transposed and merged Dirac terms of the matrix; the time of the merged Dirac term Taking the time of the largest Dirac term of the weight coefficient in the Dirac item before the merger;
  • a target tracking system based on a variable coefficient ⁇ - ⁇ filter comprising:
  • the prediction module is configured to predict the target moment and the Dirac term of the target existing at the previous moment according to the target moment and the Dirac term at the previous moment, and specify the corresponding target moment and the target for the new target at the current moment. Lack item
  • An update module configured to determine an updated Dirac item according to the target moment and the Dirac term of the target at the current moment predicted, the target moment of the new target at the current moment, and the Dirac term and the current time position measurement;
  • the reduction and merging module is used for cutting and merging the updated Dirac items to obtain the target moment and the Dirac term at the current moment; and using the target moment of the current moment and the Dirac term as the recursive input of the next moment;
  • a target state extraction module configured to extract, according to a target moment and a Dirac term at the current moment, a Dirac term with a weight coefficient greater than a first threshold as an output of the current moment, and set a target state value in the output Dirac term as a current moment The status value of a target.
  • k-1 represents a previous moment
  • k represents a current moment
  • the weighting factor representing the ith predicted Dirac term Indicates the target state value of the ith predicted Dirac term, The time at which the new Dirac term corresponding to the ith predicted Dirac term is generated, and Where F k-1 is the state transition matrix, p sk is the probability of the target surviving; the predicted target moment p k
  • the weighting factor representing the new Dirac term Indicates the target state value of the new Dirac item, Represents the time of the birth of the new Dirac item, J ⁇ k is the number of the new Dirac items, i is the index number, and t represents the time of the current time.
  • the update module is configured to use the Dirac item at the current moment according to the predicted target that has existed at the previous moment. And the target moment p k
  • t represents the time of the current time
  • K i is the gain matrix
  • z j represents the jth position measurement in the n k position measurement at the current time
  • H k is the observation matrix
  • R j is the variance matrix of the observed noise
  • c is The amplification factor
  • R 0 is the covariance set for detecting the new target
  • p Dk is the detection probability of the target
  • K i takes the gain matrix of the variable coefficient ⁇ - ⁇ filter, and Where T is the time difference between the current time and the previous time, and ⁇ k and ⁇ k are two time-varying coefficients, and
  • the target state value for the qth Dirac item The time produced for the new Dirac term corresponding to the qth Dirac term, and J k
  • Dirac item updated from the current moment q 1,...,J k
  • the method of merging multiple Dirac items is: Where L is the set formed by the merged Dirac term, and the superscripts T and b respectively represent the transposed and merged Dirac terms of the matrix; the time of the merged Dirac term Taking the time of the largest Dirac term of the weight coefficient in the Dirac item before the merger;
  • each Dirac item has a time stamp for recording the time of the corresponding new Dirac item.
  • the time stamp in the Dirac term is used for the calculation of the filter gain of the alpha-beta filter.
  • the filtering gain of the ⁇ - ⁇ filter is large, so that the present invention can quickly detect and track the nascent target; for the existing target, the filtering gain of the ⁇ - ⁇ filter is small, thus ensuring the present invention. High tracking accuracy for existing targets.
  • FIG. 1 is a schematic flow chart of a target tracking method based on a variable coefficient ⁇ - ⁇ filter according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a target tracking system based on a variable coefficient ⁇ - ⁇ filter according to an embodiment of the present invention. Schematic;
  • Figure 3 Simulation measurement data used in an embodiment of the present invention
  • Figure 4 Comparison of the average OSAC distances of the present invention with existing target tracking methods based on constant coefficient alpha-beta filters.
  • the filter gain different from the constant coefficient ⁇ - ⁇ filter is fixed, and the filter gain of the ⁇ - ⁇ filter in the present invention is time-varying.
  • the use of time-varying filter gain improves the detection capability and multi-target tracking accuracy of the present invention for new targets.
  • a target tracking method based on a variable coefficient ⁇ - ⁇ filter includes the following steps:
  • Step S1 predicting the target moment and the Dirac term of the target existing at the previous moment according to the target moment and the Dirac term at the previous moment, and assigning the corresponding target moment and the Dirac term to the new target at the current moment. ;
  • Step S2 determining an updated Dirac term according to the target moment and the Dirac term of the target existing at the previous moment predicted, the target moment of the new target at the current moment, and the Dirac term and the current position measurement;
  • Step S3 performing the reduction and merging of the updated Dirac items to obtain the target moment and the Dirac term at the current moment; and using the target moment of the current moment and the Dirac term as the recursive input of the next moment;
  • Step S4 extracting a Dirac term whose weight coefficient is greater than the first threshold as the output of the current time according to the target moment and the Dirac term at the current moment, and setting the target state value in the output Dirac term as a target state at the current time. value.
  • step S1 the previous moment is represented by k-1, and the current moment is represented by k;
  • the target moment of the previous moment is Where ⁇ represents the Dirac distribution and x represents the state of the target;
  • the step S1 specifically includes the following steps:
  • the Dirac item at the current moment i 1, 2,..., J k-1 ; among them,
  • the weighting factor representing the ith predicted Dirac term Indicates the target state value of the ith predicted Dirac term, The time at which the new Dirac term corresponding to the ith predicted Dirac term is generated, and Where F k-1 is the state transition matrix, p sk is the probability of the target surviving; the predicted target moment p k
  • the weighting factor representing the new Dirac term Indicates the target state value of the new Dirac item, Represents the time of the birth of the new Dirac item, J ⁇ k is the number of the new Dirac items, i is the index number, and t represents the time of the current time.
  • Step S2 is specifically:
  • Dirac items at the current moment based on the predicted existing target at the previous moment And the target moment p k
  • the time generated for the updated Dirac term, n k is the number of current time position measurements
  • the position measurement of the current time includes the position measurement generated by the current time target and the position measurement generated by the current time clutter
  • i and j is the index number, and when j ⁇ n k ,
  • t represents the time of the current time
  • K i is the gain matrix
  • z j represents the jth position measurement in the n k position measurement at the current time
  • H k is the observation matrix
  • R j is the variance matrix of the observed noise
  • c is The amplification factor
  • R 0 is the covariance set for detecting the new target
  • p Dk is the detection probability of the target
  • K i takes the gain matrix of the variable coefficient ⁇ - ⁇ filter, and Where T is the time difference between the current time and the previous time, and ⁇ k and ⁇ k are two time-varying coefficients, and
  • Step S3 is specifically as follows:
  • the target state value for the qth Dirac item The time produced for the new Dirac term corresponding to the qth Dirac term, and J k
  • Dirac item updated from the current moment q 1,...,J k
  • the method of merging multiple Dirac items is: Where L is the set formed by the merged Dirac term, and the superscripts T and b respectively represent the transposed and merged Dirac terms of the matrix; the time of the merged Dirac term Taking the time of the largest Dirac term of the weight coefficient in the Dirac item before the merger;
  • the present invention also provides a target tracking system based on a variable coefficient ⁇ - ⁇ filter, including:
  • the prediction module 1 is configured to predict a target moment and a Dirac term of the target existing at the previous moment according to the target moment and the Dirac term at the previous moment, and specify a corresponding target moment for the new target at the current moment and Dirac item;
  • the updating module 2 is configured to determine the updated Dirac item according to the target moment and the Dirac item of the current moment at the predicted previous moment, the target moment of the new target at the current moment, and the Dirac item and the current time position measurement. ;
  • the reduction and merging module 3 is configured to cut and merge the updated Dirac items to obtain the target moment and the Dirac term at the current moment; and use the target moment and the Dirac term at the current moment as the recursive input of the next moment;
  • the target state extraction module 4 is configured to extract, according to the target moment and the Dirac term at the current moment, a Dirac term whose weight coefficient is greater than the first threshold as the output of the current moment, and the output in the Dirac term
  • the target status value is the status value of a target at the current time.
  • the previous time is represented by k-1, and the current time is represented by k;
  • the weighting factor representing the ith predicted Dirac term Indicates the target state value of the ith predicted Dirac term, The time at which the new Dirac term corresponding to the ith predicted Dirac term is generated, and Where F k-1 is the state transition matrix, p sk is the probability of the target surviving; the predicted target moment p k
  • the weighting factor representing the new Dirac term Indicates the target state value of the new Dirac item, Represents the time of the birth of the new Dirac item, J ⁇ k is the number of the new Dirac items, i is the index number, and t represents the time of the current time.
  • the update module 2 is used for the Dirac item at the current moment according to the predicted target that has existed at the previous moment. And the target moment p k
  • t represents the time of the current time
  • K i is the gain matrix
  • z j represents the jth position measurement in the n k position measurement at the current time
  • H k is the observation matrix
  • R j is the variance matrix of the observed noise
  • c is The amplification factor
  • R 0 is the covariance set for detecting the new target
  • p Dk is the detection probability of the target
  • K i takes the gain matrix of the variable coefficient ⁇ - ⁇ filter, and Where T is the time difference between the current time and the previous time, and ⁇ k and ⁇ k are two time-varying coefficients, and
  • the target state value for the qth Dirac item The time produced for the new Dirac term corresponding to the qth Dirac term, and J k
  • Dirac item updated from the current moment q 1,...,J k
  • the method of merging multiple Dirac items is: Where L is the set formed by the merged Dirac term, and the superscripts T and b respectively represent the transposed and merged Dirac terms of the matrix; the time of the merged Dirac term Taking the time of the largest Dirac term of the weight coefficient in the Dirac item before the merger;
  • the target tracking method based on the variable coefficient ⁇ - ⁇ filter according to the present invention improves the detection capability of the new target and the multi-target tracking accuracy by using the time-varying filter gain.
  • the clutter density ⁇ c 5 ⁇ 10 -6 m -2
  • the target state Where x and y represent position components, respectively.
  • T transpose
  • Survival probability p sk 1.0
  • target detection probability p Dk 1.0
  • R 0 (diag([45(m) 45(m))))) 2
  • amplification factor c 3
  • first threshold is taken as 0.5
  • second The threshold is taken as 10 -3 and the third threshold is taken as 2m.
  • the target state values are [-900(m), 0(ms -1 ), - 900(m), 0(ms -1 )] T , [-900(m), 0(ms -1 ), 900(m), 0(ms -1 )] T , [-900(m),0 (ms -1 ), 200 (m), 0 (ms -1 )] T and [-900 (m), 0 (ms -1 ), -400 (m), 0 (ms -1 )] T , 4
  • the weight coefficient of the Dirac term of a new target is 0.1; in the case where both the new target appears and the existing target disappears, and the process noise covariance of the target is unknown, the present invention and the existing constant coefficient ⁇ -
  • the target tracking method of the ⁇ filter is the average OSA (Optimal Subpattern Assignment) obtained by 100 Monte Carlo experiments when the simulation data shown in Figure 3 (the simulation)
  • the multi-target tracking accuracy of the present invention is better than the existing method, and the OSPA distance is obtained by the existing method.
  • the OSPA distance is small.

Landscapes

  • Radar Systems Or Details Thereof (AREA)
  • Image Analysis (AREA)

Abstract

一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统,适用于多传感器信息融合领域。所述方法步骤如下:首先根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项,再结合当前时刻的位置测量并利用变系数α-β滤波器确定更新的狄拉克项,再对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项,当前时刻的目标矩及狄拉克项作为下一时刻递归的输入,最后提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,所输出狄拉克项中的目标状态值为当前时刻一个目标的状态值。

Description

一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统 技术领域
本发明属于多传感器信息融合技术领域,尤其涉及一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统。
背景技术
概率假设密度滤波方法是解决目标检测和跟踪的新方法。其最大的优点是减少了贝叶斯滤波方法中的积分运算,并能给出瞬时目标数估计,在多目标的检测、定位与跟踪中已取得了比较广泛的应用。
概率假设密度滤波方法是一种传递联合后验分布一阶矩的数据处理方法。由于涉及积分运算,概率假设密度滤波器通常采用粒子滤波法或高斯混和法近似实现。不论是粒子滤波法还是高斯混和法均要求目标运动过程中过程噪声的分布特征、传感器观测过程中观测噪声的分布特征均是已知的。当过程噪声的分布特征未知时,现有的概率假设密度滤波器及其近似实现方法难以工作。过程噪声分布特征未知情况下概率假设密度滤波器的多目标跟踪问题是需要探索和解决的一个关键技术问题。
为解决过程噪声分布特征或目标数未知情况下的多目标跟踪问题,我们提出了一种基于常系数α-β滤波器的目标跟踪方法,但该方法存在新目标探测能力差,多目标跟踪精度低等问题,如何提高α-β滤波器的新目标探测能力和多目标跟踪精度又是一个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基于变系数α-β滤波器的目标跟踪方法与跟踪系统,旨在提高α-β滤波器的新目标探测能力和多目标跟踪精度。本发明是这样实现的:
一种基于变系数α-β滤波器的目标跟踪方法,包括以下步骤:
步骤S1:根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项;
步骤S2:根据预测的前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项、当前时刻新生目标的目标矩及狄拉克项以及当前时刻的位置测量确定更 新的狄拉克项;
步骤S3:对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项;并将当前时刻的目标矩及狄拉克项作为下一时刻递归的输入;
步骤S4:根据当前时刻的目标矩及狄拉克项,提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,并将所输出狄拉克项中的目标状态值为当前时刻一个目标的状态值。
进一步地,所述步骤S1中,以k-1表示前一时刻,以k表示当前时刻;
前一时刻的狄拉克项为
Figure PCTCN2015078026-appb-000001
i=1,2,…,Jk-1,其中,
Figure PCTCN2015078026-appb-000002
表示前一时刻第i个狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000003
表示前一时刻第i个狄拉克项的目标状态值,Jk-1表示前一时刻狄拉克项的数目,i为索引号,取值为1至Jk-1
Figure PCTCN2015078026-appb-000004
表示第i个狄拉克项对应的新生狄拉克项产生的时间;
前一时刻的目标矩为
Figure PCTCN2015078026-appb-000005
其中,δ表示狄拉克分布,x表示目标的状态;
所述步骤S1具体包括下述步骤:
根据前一时刻目标矩pk-1(x)及狄拉克项
Figure PCTCN2015078026-appb-000006
预测前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000007
i=1,2,…,Jk-1;其中,
Figure PCTCN2015078026-appb-000008
表示第i个预测的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000009
表示第i个预测的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000010
表示第i个预测的狄拉克项对应的新生狄拉克项产生的时间,并且
Figure PCTCN2015078026-appb-000011
其中,Fk-1为状态转移矩阵,psk为目标幸存的概率;预测的目标矩pk|k-1(x)为
Figure PCTCN2015078026-appb-000012
指定当前时刻新生目标的狄拉克项为
Figure PCTCN2015078026-appb-000013
i=1,2,…,Jγk,新生目标的目标矩为
Figure PCTCN2015078026-appb-000014
其中,
Figure PCTCN2015078026-appb-000015
表示新生的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000016
表示新生的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000017
表示新生狄拉克项产生的时间,Jγk为新生狄拉克项的个数,i为索引号,并且
Figure PCTCN2015078026-appb-000018
t表示当前时刻的时间。
进一步地,所述步骤S2具体为:
根据预测的前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000019
及目标矩pk|k-1(x),以及指定的当前时刻新生目标的狄拉克项
Figure PCTCN2015078026-appb-000020
及目标矩pγk(x),以及当前时刻的位置测量,利用变系数α-β滤波器确定出当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000021
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1,其中,
Figure PCTCN2015078026-appb-000022
为更新的狄拉克项的权重系数、
Figure PCTCN2015078026-appb-000023
为更新的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000024
为更新的狄拉克项产生的时间,nk为当前时刻位置测量的个数,所述当前时刻的位置测量包含由当前时刻目标产生的位置测量和由当前时刻杂波产生的位置测量,i和j是索引号,并且当j≤nk时,
Figure PCTCN2015078026-appb-000025
Figure PCTCN2015078026-appb-000026
其中,t表示当前时刻的时间,Ki为增益矩阵,zj表示当前时刻nk个位置测量中的第j个位置测量,Hk为观测矩阵,Rj为观测噪声的方差矩阵,c为放大因子,R0是为探测新目标而设置的协方差,pDk为目标的检测概率,λC为当前时刻观测空间中杂波的密度,当j=nk+1时,
Figure PCTCN2015078026-appb-000027
Figure PCTCN2015078026-appb-000028
Ki取变系数α-β滤波器的增益矩阵,并且
Figure PCTCN2015078026-appb-000029
其中,T为当前时刻与前一时刻的时间差,αk和βk是两个时变的系数,并且
Figure PCTCN2015078026-appb-000030
进一步地,所述步骤S3具体为:
将当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000031
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1表述为
Figure PCTCN2015078026-appb-000032
q=1,…,Jk|k;其中,q为索引号,取值为1至Jk|k,Jk|k为更新的狄拉克项的个数,
Figure PCTCN2015078026-appb-000033
为第q个狄拉克项的权重,
Figure PCTCN2015078026-appb-000034
为第q个狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000035
为第q个狄拉克项对应的新生狄拉克项产生的时间,并且Jk|k=(nk+1)(Jk-1+Jγk),q=(Jk-1+Jγk)(j-1)+i,
Figure PCTCN2015078026-appb-000036
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1;
从当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000037
q=1,…,Jk|k中裁减掉权重系数
Figure PCTCN2015078026-appb-000038
小于第二阈值的狄拉克项;
从裁减后余下的狄拉克项中将距离dij小于第三阈值的狄拉克项合并成一 个;其中,合并距离dij
Figure PCTCN2015078026-appb-000039
多个狄拉克项的合并方法为:
Figure PCTCN2015078026-appb-000040
其中,L为合并狄拉克项上标形成的集合,上标T和b分别表示矩阵的转置和合并后狄拉克项的索引号;合并狄拉克项的时间
Figure PCTCN2015078026-appb-000041
取合并前狄拉克项中权重系数最大狄拉克项的时间;
将裁减与合并后的狄拉克项
Figure PCTCN2015078026-appb-000042
i=1,…,Jk作为当前时刻的狄拉克项,并将裁减与合并后的狄拉克项的加权和作为当前时刻的目标矩
Figure PCTCN2015078026-appb-000043
其中,Jk为当前时刻狄拉克项的个数,i为索引号,取值为1至Jk
一种基于变系数α-β滤波器的目标跟踪系统,包括:
预测模块,用于根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项;
更新模块,用于根据预测的前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项、当前时刻新生目标的目标矩及狄拉克项以及当前时刻的位置测量确定更新的狄拉克项;
裁减与合并模块,用于对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项;并将当前时刻的目标矩及狄拉克项作为下一时刻递归的输入;
目标状态提取模块,用于根据当前时刻的目标矩及狄拉克项,提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,并将所输出狄拉克项中的目标状态值为当前时刻一个目标的状态值。
进一步地,在所述预测模块中,以k-1表示前一时刻,以k表示当前时刻;
前一时刻的狄拉克项为
Figure PCTCN2015078026-appb-000044
i=1,2,…,Jk-1,其中,
Figure PCTCN2015078026-appb-000045
表示前一时刻第i个狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000046
表示前一时刻第i个狄拉克项的目标状态值,Jk-1表示前一时刻狄拉克项的数目,i为索引号,取值为1至Jk-1
Figure PCTCN2015078026-appb-000047
表示第i个狄拉克项对应的新生狄拉克项产生的时间;
前一时刻的目标矩为
Figure PCTCN2015078026-appb-000048
其中,δ表示狄拉克分布,x表示目标的状态;所述预测模块用于根据前一时刻目标矩pk-1(x)及狄拉克项
Figure PCTCN2015078026-appb-000049
预测前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000050
i=1,2,…,Jk-1;其中,
Figure PCTCN2015078026-appb-000051
表示第i个预测的狄拉克项的权重系 数,
Figure PCTCN2015078026-appb-000052
表示第i个预测的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000053
表示第i个预测的狄拉克项对应的新生狄拉克项产生的时间,并且
Figure PCTCN2015078026-appb-000054
Figure PCTCN2015078026-appb-000055
其中,Fk-1为状态转移矩阵,psk为目标幸存的概率;预测的目标矩pk|k-1(x)为
Figure PCTCN2015078026-appb-000056
指定当前时刻新生目标的狄拉克项为
Figure PCTCN2015078026-appb-000057
i=1,2,…,Jγk,新生目标的目标矩为
Figure PCTCN2015078026-appb-000058
其中,
Figure PCTCN2015078026-appb-000059
表示新生的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000060
表示新生的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000061
表示新生狄拉克项产生的时间,Jγk为新生狄拉克项的个数,i为索引号,并且
Figure PCTCN2015078026-appb-000062
t表示当前时刻的时间。
进一步地,所述更新模块用于根据预测的前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000063
及目标矩pk|k-1(x),以及指定的当前时刻新生目标的狄拉克项
Figure PCTCN2015078026-appb-000064
及目标矩pγk(x),以及当前时刻的位置测量,利用变系数α-β滤波器确定出当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000065
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1,其中,为更新的狄拉克项的权重系数、
Figure PCTCN2015078026-appb-000067
为更新的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000068
为更新的狄拉克项产生的时间,nk为当前时刻位置测量的个数,所述当前时刻的位置测量包含由当前时刻目标产生的位置测量和由当前时刻杂波产生的位置测量,i和j是索引号,并且当j≤nk时,
Figure PCTCN2015078026-appb-000069
Figure PCTCN2015078026-appb-000070
其中,t表示当前时刻的时间,Ki为增益矩阵,zj表示当前时刻nk个位置测量中的第j个位置测量,Hk为观测矩阵,Rj为观测噪声的方差矩阵,c为放大因子,R0是为探测新目标而设置的协方差,pDk为目标的检测概率,λC为当前时刻观测空间中杂波的密度,当j=nk+1时,
Figure PCTCN2015078026-appb-000071
Figure PCTCN2015078026-appb-000072
Ki取变系数α-β滤波器的增益矩阵,并且
Figure PCTCN2015078026-appb-000073
其中,T为当前时刻与前一时刻的时间差,αk和βk是两个时变的系数,并且
Figure PCTCN2015078026-appb-000074
进一步地,所述裁减与合并模块用于将当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000075
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1表述为
Figure PCTCN2015078026-appb-000076
q=1,…,Jk|k;其中,q为索引号,取值为1至Jk|k,Jk|k为更新的狄拉克项的个数,
Figure PCTCN2015078026-appb-000077
为第q个狄拉克项的权重,
Figure PCTCN2015078026-appb-000078
为第q个狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000079
为第q个狄拉克项对应的新生狄拉克项产生的时间,并且Jk|k=(nk+1)(Jk-1+Jγk),q=(Jk-1+Jγk)(j-1)+i,
Figure PCTCN2015078026-appb-000080
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1;
从当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000081
q=1,…,Jk|k中裁减掉权重系数
Figure PCTCN2015078026-appb-000082
小于第二阈值的狄拉克项;
从裁减后余下的狄拉克项中将距离dij小于第三阈值的狄拉克项合并成一个;其中,合并距离dij
Figure PCTCN2015078026-appb-000083
多个狄拉克项的合并方法为:
Figure PCTCN2015078026-appb-000084
其中,L为合并狄拉克项上标形成的集合,上标T和b分别表示矩阵的转置和合并后狄拉克项的索引号;合并狄拉克项的时间
Figure PCTCN2015078026-appb-000085
取合并前狄拉克项中权重系数最大狄拉克项的时间;
将裁减与合并后的狄拉克项
Figure PCTCN2015078026-appb-000086
i=1,…,Jk作为当前时刻的狄拉克项,并将裁减与合并后的狄拉克项的加权和作为当前时刻的目标矩
Figure PCTCN2015078026-appb-000087
其中,Jk为当前时刻狄拉克项的个数,i为索引号,取值为1至Jk
本发明与基于常系数α-β滤波器的多目标跟踪方法相比,常系数α-β滤波器的滤波增益是固定不变的,而本发明中α-β滤波器的滤波增益是时变的。在本发明中,每一个狄拉克项都有一个时间标识,用以记录与之相应的新生狄拉克项产生的时间。狄拉克项中的时间标识被用于α-β滤波器的滤波增益的计算。对于新生的目标,α-β滤波器的滤波增益大,从而使得本发明能快速地探测与跟踪新生的目标;对于已存在的目标,α-β滤波器的滤波增益小,这样能保证本发明对已存在的目标具有较高的跟踪精度。
附图说明
图1:本发明实施例提供的一种基于变系数α-β滤波器的目标跟踪方法的流程示意图;
图2:本发明实施例提供的一种基于变系数α-β滤波器的目标跟踪系统的 结构示意图;
图3:本发明实施例所使用的仿真测量数据;
图4:本发明与现有的基于常系数α-β滤波器的目标跟踪方法的平均OSPA距离对比图。
具体实施方式
为使本发明的目的、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。
不同于常系数α-β滤波器的滤波增益固定不变,本发明中α-β滤波器的滤波增益是时变的。利用时变的滤波增益提高了本发明对新目标的探测能力及多目标跟踪精度。
如图1所示,一种基于变系数α-β滤波器的目标跟踪方法,包括以下步骤:
步骤S1:根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项;
步骤S2:根据预测的前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项、当前时刻新生目标的目标矩及狄拉克项以及当前时刻的位置测量确定更新的狄拉克项;
步骤S3:对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项;并将当前时刻的目标矩及狄拉克项作为下一时刻递归的输入;
步骤S4:根据当前时刻的目标矩及狄拉克项,提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,并将所输出狄拉克项中的目标状态值为当前时刻一个目标的状态值。
步骤S1中,以k-1表示前一时刻,以k表示当前时刻;
前一时刻的狄拉克项为
Figure PCTCN2015078026-appb-000088
i=1,2,…,Jk-1,其中,
Figure PCTCN2015078026-appb-000089
表示前一时刻第i个狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000090
表示前一时刻第i个狄拉克项的目标状态值,Jk-1表示前一时刻狄拉克项的数目,i为索引号,取值为1至Jk-1
Figure PCTCN2015078026-appb-000091
表示第i个狄拉克项对应的新生狄拉克项产生的时间;
前一时刻的目标矩为
Figure PCTCN2015078026-appb-000092
其中,δ表示狄拉克分布,x表示目标的状态;
所述步骤S1具体包括下述步骤:
根据前一时刻目标矩pk-1(x)及狄拉克项
Figure PCTCN2015078026-appb-000093
预测前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000094
i=1,2,…,Jk-1;其中,
Figure PCTCN2015078026-appb-000095
表示第i个预测的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000096
表示第i个预测的狄拉克项的目标状 态值,
Figure PCTCN2015078026-appb-000097
表示第i个预测的狄拉克项对应的新生狄拉克项产生的时间,并且
Figure PCTCN2015078026-appb-000098
其中,Fk-1为状态转移矩阵,psk为目标幸存的概率;预测的目标矩pk|k-1(x)为
Figure PCTCN2015078026-appb-000099
指定当前时刻新生目标的狄拉克项为
Figure PCTCN2015078026-appb-000100
i=1,2,…,Jγk,新生目标的目标矩为
Figure PCTCN2015078026-appb-000101
其中,
Figure PCTCN2015078026-appb-000102
表示新生的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000103
表示新生的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000104
表示新生狄拉克项产生的时间,Jγk为新生狄拉克项的个数,i为索引号,并且
Figure PCTCN2015078026-appb-000105
t表示当前时刻的时间。
步骤S2具体为:
根据预测的前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000106
及目标矩pk|k-1(x),以及指定的当前时刻新生目标的狄拉克项
Figure PCTCN2015078026-appb-000107
及目标矩pγk(x),以及当前时刻的位置测量,利用变系数α-β滤波器确定出当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000108
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1,其中,为更新的狄拉克项的权重系数、
Figure PCTCN2015078026-appb-000110
为更新的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000111
为更新的狄拉克项产生的时间,nk为当前时刻位置测量的个数,所述当前时刻的位置测量包含由当前时刻目标产生的位置测量和由当前时刻杂波产生的位置测量,i和j是索引号,并且当j≤nk时,
Figure PCTCN2015078026-appb-000112
Figure PCTCN2015078026-appb-000113
其中,t表示当前时刻的时间,Ki为增益矩阵,zj表示当前时刻nk个位置测量中的第j个位置测量,Hk为观测矩阵,Rj为观测噪声的方差矩阵,c为放大因子,R0是为探测新目标而设置的协方差,pDk为目标的检测概率,λC为当前时刻观测空间中杂波的密度,当j=nk+1时,
Figure PCTCN2015078026-appb-000114
Figure PCTCN2015078026-appb-000115
Ki取变系数α-β滤波器的增益矩阵,并且
Figure PCTCN2015078026-appb-000116
其中,T为当前时刻与前一时刻的时间差,αk和βk是两个时变的系数,并且
Figure PCTCN2015078026-appb-000117
步骤S3具体为:
将当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000118
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1表述为
Figure PCTCN2015078026-appb-000119
q=1,…,Jk|k;其中,q为索引号,取值为1至Jk|k,Jk|k为更新的狄拉克项的个数,
Figure PCTCN2015078026-appb-000120
为第q个狄拉克项的权重,
Figure PCTCN2015078026-appb-000121
为第q个狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000122
为第q个狄拉克项对应的新生狄拉克项产生的时间,并且Jk|k=(nk+1)(Jk-1+Jγk),q=(Jk-1+Jγk)(j-1)+i,
Figure PCTCN2015078026-appb-000123
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1;
从当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000124
q=1,…,Jk|k中裁减掉权重系数
Figure PCTCN2015078026-appb-000125
小于第二阈值的狄拉克项;
从裁减后余下的狄拉克项中将距离dij小于第三阈值的狄拉克项合并成一个;其中,合并距离dij
Figure PCTCN2015078026-appb-000126
多个狄拉克项的合并方法为:
Figure PCTCN2015078026-appb-000127
其中,L为合并狄拉克项上标形成的集合,上标T和b分别表示矩阵的转置和合并后狄拉克项的索引号;合并狄拉克项的时间
Figure PCTCN2015078026-appb-000128
取合并前狄拉克项中权重系数最大狄拉克项的时间;
将裁减与合并后的狄拉克项
Figure PCTCN2015078026-appb-000129
i=1,…,Jk作为当前时刻的狄拉克项,并将裁减与合并后的狄拉克项的加权和作为当前时刻的目标矩
Figure PCTCN2015078026-appb-000130
其中,Jk为当前时刻狄拉克项的个数,i为索引号,取值为1至Jk。τ即为设定阈值,可取0.5。
如图2所示,本发明还提供一种基于变系数α-β滤波器的目标跟踪系统,包括:
预测模块1,用于根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项;
更新模块2,用于根据预测的前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项、当前时刻新生目标的目标矩及狄拉克项以及当前时刻的位置测量确定更新的狄拉克项;
裁减与合并模块3,用于对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项;并将当前时刻的目标矩及狄拉克项作为下一时刻递归的输入;
目标状态提取模块4,用于根据当前时刻的目标矩及狄拉克项,提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,并将所输出狄拉克项中的 目标状态值为当前时刻一个目标的状态值。
其中,在预测模块1中,以k-1表示前一时刻,以k表示当前时刻;
前一时刻的狄拉克项为
Figure PCTCN2015078026-appb-000131
i=1,2,…,Jk-1,其中,
Figure PCTCN2015078026-appb-000132
表示前一时刻第i个狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000133
表示前一时刻第i个狄拉克项的目标状态值,Jk-1表示前一时刻狄拉克项的数目,i为索引号,取值为1至Jk-1
Figure PCTCN2015078026-appb-000134
表示第i个狄拉克项对应的新生狄拉克项产生的时间;
前一时刻的目标矩为
Figure PCTCN2015078026-appb-000135
其中,δ表示狄拉克分布,x表示目标的状态;所述预测模块用于根据前一时刻目标矩pk-1(x)及狄拉克项
Figure PCTCN2015078026-appb-000136
预测前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000137
i=1,2,…,Jk-1;其中,
Figure PCTCN2015078026-appb-000138
表示第i个预测的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000139
表示第i个预测的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000140
表示第i个预测的狄拉克项对应的新生狄拉克项产生的时间,并且
Figure PCTCN2015078026-appb-000141
Figure PCTCN2015078026-appb-000142
其中,Fk-1为状态转移矩阵,psk为目标幸存的概率;预测的目标矩pk|k-1(x)为
Figure PCTCN2015078026-appb-000143
指定当前时刻新生目标的狄拉克项为
Figure PCTCN2015078026-appb-000144
i=1,2,…,Jγk,新生目标的目标矩为
Figure PCTCN2015078026-appb-000145
其中,
Figure PCTCN2015078026-appb-000146
表示新生的狄拉克项的权重系数,
Figure PCTCN2015078026-appb-000147
表示新生的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000148
表示新生狄拉克项产生的时间,Jγk为新生狄拉克项的个数,i为索引号,并且
Figure PCTCN2015078026-appb-000149
t表示当前时刻的时间。
更新模块2用于根据预测的前一时刻已经存在的目标在当前时刻的狄拉克项
Figure PCTCN2015078026-appb-000150
及目标矩pk|k-1(x),以及指定的当前时刻新生目标的狄拉克项
Figure PCTCN2015078026-appb-000151
及目标矩pγk(x),以及当前时刻的位置测量,利用变系数α-β滤波器确定出当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000152
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1,其中,
Figure PCTCN2015078026-appb-000153
为更新的狄拉克项的权重系数、
Figure PCTCN2015078026-appb-000154
为更新的狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000155
为更新的狄拉克项产生的时间,nk为当前时刻位置测量的个数,所述当前时刻的位置测量包含由当前时刻目标产生的位置测量和由当前时刻杂波产生的位置测量,i和j是索引号,并且当j≤nk时,
Figure PCTCN2015078026-appb-000156
Figure PCTCN2015078026-appb-000157
其中,t表示当前时刻的时间,Ki为增益矩阵,zj表示当前时刻nk个位置测量中的第j个位置测量,Hk为观测矩阵,Rj为观测噪声的方差矩阵,c为放大因子,R0是为探测新目标而设置的协方差,pDk为目标的检测概率,λC为当前时刻观测空间中杂波的密度,当j=nk+1时,
Figure PCTCN2015078026-appb-000158
Figure PCTCN2015078026-appb-000159
Ki取变系数α-β滤波器的增益矩阵,并且
Figure PCTCN2015078026-appb-000160
其中,T为当前时刻与前一时刻的时间差,αk和βk是两个时变的系数,并且
Figure PCTCN2015078026-appb-000161
裁减与合并模块3用于将当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000162
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1表述为
Figure PCTCN2015078026-appb-000163
q=1,…,Jk|k;其中,q为索引号,取值为1至Jk|k,Jk|k为更新的狄拉克项的个数,
Figure PCTCN2015078026-appb-000164
为第q个狄拉克项的权重,
Figure PCTCN2015078026-appb-000165
为第q个狄拉克项的目标状态值,
Figure PCTCN2015078026-appb-000166
为第q个狄拉克项对应的新生狄拉克项产生的时间,并且Jk|k=(nk+1)(Jk-1+Jγk),q=(Jk-1+Jγk)(j-1)+i,
Figure PCTCN2015078026-appb-000167
Figure PCTCN2015078026-appb-000168
i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1;
从当前时刻更新的狄拉克项
Figure PCTCN2015078026-appb-000169
q=1,…,Jk|k中裁减掉权重系数
Figure PCTCN2015078026-appb-000170
小于第二阈值的狄拉克项;
从裁减后余下的狄拉克项中将距离dij小于第三阈值的狄拉克项合并成一个;其中,合并距离dij
Figure PCTCN2015078026-appb-000171
多个狄拉克项的合并方法为:
Figure PCTCN2015078026-appb-000172
其中,L为合并狄拉克项上标形成的集合,上标T和b分别表示矩阵的转置和合并后狄拉克项的索引号;合并狄拉克项的时间
Figure PCTCN2015078026-appb-000173
取合并前狄拉克项中权重系数最大狄拉克项的时间;
将裁减与合并后的狄拉克项
Figure PCTCN2015078026-appb-000174
i=1,…,Jk作为当前时刻的狄拉克项,并将裁减与合并后的狄拉克项的加权和作为当前时刻的目标矩
Figure PCTCN2015078026-appb-000175
其中,Jk为当前时刻狄拉克项的个数,i为索引号,取值为1至Jk
本发明所述的一种基于变系数α-β滤波器的目标跟踪方法利用时变的滤波增益提高了本发明对新目标的探测能力及多目标跟踪精度。作为本发明的一 个实施例,取杂波密度λc=5×10-6m-2,目标状态
Figure PCTCN2015078026-appb-000176
其中,x和y分别表示位置分量,
Figure PCTCN2015078026-appb-000177
Figure PCTCN2015078026-appb-000178
表示速度分量,上标T表示转置,状态转移矩阵
Figure PCTCN2015078026-appb-000179
其中,T=1秒表示传感器的采样时间间隔,传感器对目标观测获得的测量为目标当前的位置,其观测矩阵
Figure PCTCN2015078026-appb-000180
观测噪声的方差矩阵
Figure PCTCN2015078026-appb-000181
幸存概率psk=1.0,目标检测概率pDk=1.0,R0=(diag([45(m) 45(m)]))2,放大因子c=3,第一阈值取为0.5,第二阈值取为10-3,第三阈值取为2m;实验中,当前时刻新生目标的狄拉克项有4个,其目标状态值分别为[-900(m),0(ms-1),-900(m),0(ms-1)]T、[-900(m),0(ms-1),900(m),0(ms-1)]T、[-900(m),0(ms-1),200(m),0(ms-1)]T和[-900(m),0(ms-1),-400(m),0(ms-1)]T,4个新生目标的狄拉克项的权重系数均为0.1;在既存在新目标出现又存在已有目标消失、同时目标的过程噪声协方差未知的情况下,本发明与现有的基于常系数α-β滤波器的目标跟踪方法对图3所示的仿真数据(仿真实验数据有4批目标)处理时100次Monte Carlo实验得到的平均OSPA(Optimal Subpattern Assignment,最优亚模式分配)距离图4所示。从图4中可看出,与现有的基于常系数的α-β滤波器的目标跟踪方法相比,本发明的多目标跟踪精度好于现有方法,其OSPA距离比现有方法得到的OSPA距离要小。

Claims (8)

  1. 一种基于变系数α-β滤波器的目标跟踪方法,其特征在于,包括以下步骤:
    步骤S1:根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项;
    步骤S2:根据预测的前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项、当前时刻新生目标的目标矩及狄拉克项以及当前时刻的位置测量确定更新的狄拉克项;
    步骤S3:对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项;并将当前时刻的目标矩及狄拉克项作为下一时刻递归的输入;
    步骤S4:根据当前时刻的目标矩及狄拉克项,提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,并将所输出狄拉克项中的目标状态值为当前时刻一个目标的状态值。
  2. 根据权利要求1所述的一种基于变系数α-β滤波器的目标跟踪方法,其特征在于,所述步骤S1中,以k-1表示前一时刻,以k表示当前时刻;
    前一时刻的狄拉克项为
    Figure PCTCN2015078026-appb-100001
    i=1,2,…,Jk-1,其中,
    Figure PCTCN2015078026-appb-100002
    表示前一时刻第i个狄拉克项的权重系数,
    Figure PCTCN2015078026-appb-100003
    表示前一时刻第i个狄拉克项的目标状态值,Jk-1表示前一时刻狄拉克项的数目,i为索引号,取值为1至Jk-1
    Figure PCTCN2015078026-appb-100004
    表示第i个狄拉克项对应的新生狄拉克项产生的时间;
    前一时刻的目标矩为
    Figure PCTCN2015078026-appb-100005
    其中,δ表示狄拉克分布,x表示目标的状态;
    所述步骤S1具体包括下述步骤:
    根据前一时刻目标矩pk-1(x)及狄拉克项
    Figure PCTCN2015078026-appb-100006
    预测前一时刻已经存在的目标在当前时刻的狄拉克项
    Figure PCTCN2015078026-appb-100007
    i=1,2,…,Jk-1;其中,
    Figure PCTCN2015078026-appb-100008
    表示第i个预测的狄拉克项的权重系数,
    Figure PCTCN2015078026-appb-100009
    表示第i个预测的狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100010
    表示第i个预测的狄拉克项对应的新生狄拉克项产生的时间,并且
    Figure PCTCN2015078026-appb-100011
    其中,Fk-1为状态转移矩阵,psk为目标幸存的概率;预测的目标矩pk|k-1(x)为
    Figure PCTCN2015078026-appb-100012
    指定当前时刻新生目标的狄拉克项为
    Figure PCTCN2015078026-appb-100013
    i=1,2,…,Jγk,新 生目标的目标矩为
    Figure PCTCN2015078026-appb-100014
    其中,
    Figure PCTCN2015078026-appb-100015
    表示新生的狄拉克项的权重系数,
    Figure PCTCN2015078026-appb-100016
    表示新生的狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100017
    表示新生狄拉克项产生的时间,Jγk为新生狄拉克项的个数,i为索引号,并且
    Figure PCTCN2015078026-appb-100018
    t表示当前时刻的时间。
  3. 根据权利要求2所述的一种基于变系数α-β滤波器的目标跟踪方法,其特征在于,所述步骤S2具体为:
    根据预测的前一时刻已经存在的目标在当前时刻的狄拉克项
    Figure PCTCN2015078026-appb-100019
    及目标矩pk|k-1(x),以及指定的当前时刻新生目标的狄拉克项
    Figure PCTCN2015078026-appb-100020
    及目标矩pγk(x),以及当前时刻的位置测量,利用变系数α-β滤波器确定出当前时刻更新的狄拉克项
    Figure PCTCN2015078026-appb-100021
    i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1,其中,
    Figure PCTCN2015078026-appb-100022
    为更新的狄拉克项的权重系数、
    Figure PCTCN2015078026-appb-100023
    为更新的狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100024
    为更新的狄拉克项产生的时间,nk为当前时刻位置测量的个数,所述当前时刻的位置测量包含由当前时刻目标产生的位置测量和由当前时刻杂波产生的位置测量,i和j是索引号,并且当j≤nk时,
    Figure PCTCN2015078026-appb-100025
    Figure PCTCN2015078026-appb-100026
    Figure PCTCN2015078026-appb-100027
    其中,t表示当前时刻的时间,Ki为增益矩阵,zj表示当前时刻nk个位置测量中的第j个位置测量,Hk为观测矩阵,Rj为观测噪声的方差矩阵,c为放大因子,R0是为探测新目标而设置的协方差,pDk为目标的检测概率,λC为当前时刻观测空间中杂波的密度,当j=nk+1时,
    Figure PCTCN2015078026-appb-100028
    Figure PCTCN2015078026-appb-100029
    Ki取变系数α-β滤波器的增益矩阵,并且
    Figure PCTCN2015078026-appb-100030
    其中,T为当前时刻与前一时刻的时间差,αk和βk是两个时变的系数,并且
    Figure PCTCN2015078026-appb-100031
  4. 根据权利要求3所述的一种基于变系数α-β滤波器的目标跟踪方法,其特征在于,所述步骤S3具体为:
    将当前时刻更新的狄拉克项
    Figure PCTCN2015078026-appb-100032
    i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1表述为
    Figure PCTCN2015078026-appb-100033
    q=1,…,Jk|k;其中,q为索引号,取值为1至Jk|k,Jk|k为更新的狄拉克项的个数,
    Figure PCTCN2015078026-appb-100034
    为第q个狄拉克项的权重,
    Figure PCTCN2015078026-appb-100035
    为第q个狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100036
    为第q个狄拉克项对应的新生狄拉克项产生的时间,并且Jk|k=(nk+1)(Jk-1+Jγk),q=(Jk-1+Jγk)(j-1)+i,
    Figure PCTCN2015078026-appb-100037
    i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1;
    从当前时刻更新的狄拉克项
    Figure PCTCN2015078026-appb-100038
    q=1,…,Jk|k中裁减掉权重系数
    Figure PCTCN2015078026-appb-100039
    小于第二阈值的狄拉克项;
    从裁减后余下的狄拉克项中将距离dij小于第三阈值的狄拉克项合并成一个;其中,合并距离dij
    Figure PCTCN2015078026-appb-100040
    多个狄拉克项的合并方法为:
    Figure PCTCN2015078026-appb-100041
    其中,L为合并狄拉克项上标形成的集合,上标T和b分别表示矩阵的转置和合并后狄拉克项的索引号;合并狄拉克项的时间
    Figure PCTCN2015078026-appb-100042
    取合并前狄拉克项中权重系数最大狄拉克项的时间;
    将裁减与合并后的狄拉克项
    Figure PCTCN2015078026-appb-100043
    i=1,…,Jk作为当前时刻的狄拉克项,并将裁减与合并后的狄拉克项的加权和作为当前时刻的目标矩
    Figure PCTCN2015078026-appb-100044
    其中,Jk为当前时刻狄拉克项的个数,i为索引号,取值为1至Jk
  5. 一种基于变系数α-β滤波器的目标跟踪系统,其特征在于,包括:
    预测模块,用于根据前一时刻的目标矩及狄拉克项,预测前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项,并为当前时刻新生的目标指定相应的目标矩及狄拉克项;
    更新模块,用于根据预测的前一时刻已经存在的目标在当前时刻的目标矩及狄拉克项、当前时刻新生目标的目标矩及狄拉克项以及当前时刻的位置测量确定更新的狄拉克项;
    裁减与合并模块,用于对更新的狄拉克项进行裁减与合并,得到当前时刻的目标矩及狄拉克项;并将当前时刻的目标矩及狄拉克项作为下一时刻递归的输入;
    目标状态提取模块,用于根据当前时刻的目标矩及狄拉克项,提取权重系数大于第一阈值的狄拉克项作为当前时刻的输出,并将所输出狄拉克项中的目标状态值为当前时刻一个目标的状态值。
  6. 根据权利要求5所述的一种基于变系数α-β滤波器的目标跟踪系统,其特征在于,在所述预测模块中,以k-1表示前一时刻,以k表示当前时刻;
    前一时刻的狄拉克项为
    Figure PCTCN2015078026-appb-100045
    i=1,2,…,Jk-1,其中,
    Figure PCTCN2015078026-appb-100046
    表示前一时刻第i个狄拉克项的权重系数,
    Figure PCTCN2015078026-appb-100047
    表示前一时刻第i个狄拉克项的目标状态值,Jk-1表示前一时刻狄拉克项的数目,i为索引号,取值为1至Jk-1
    Figure PCTCN2015078026-appb-100048
    表示第i个狄拉克项对应的新生狄拉克项产生的时间;
    前一时刻的目标矩为
    Figure PCTCN2015078026-appb-100049
    其中,δ表示狄拉克分布,x表示目标的状态;所述预测模块用于根据前一时刻目标矩pk-1(x)及狄拉克项
    Figure PCTCN2015078026-appb-100050
    预测前一时刻已经存在的目标在当前时刻的狄拉克项
    Figure PCTCN2015078026-appb-100051
    i=1,2,…,Jk-1;其中,
    Figure PCTCN2015078026-appb-100052
    表示第i个预测的狄拉克项的权重系数,
    Figure PCTCN2015078026-appb-100053
    表示第i个预测的狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100054
    表示第i个预测的狄拉克项对应的新生狄拉克项产生的时间,并且
    Figure PCTCN2015078026-appb-100055
    Figure PCTCN2015078026-appb-100056
    其中,Fk-1为状态转移矩阵,psk为目标幸存的概率;预测的目标矩pk|k-1(x)为
    Figure PCTCN2015078026-appb-100057
    指定当前时刻新生目标的狄拉克项为
    Figure PCTCN2015078026-appb-100058
    i=1,2,…,Jγk,新生目标的目标矩为
    Figure PCTCN2015078026-appb-100059
    其中,
    Figure PCTCN2015078026-appb-100060
    表示新生的狄拉克项的权重系数,
    Figure PCTCN2015078026-appb-100061
    表示新生的狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100062
    表示新生狄拉克项产生的时间,Jγk为新生狄拉克项的个数,i为索引号,并且
    Figure PCTCN2015078026-appb-100063
    t表示当前时刻的时间。
  7. 根据权利要求6所述的一种基于变系数α-β滤波器的目标跟踪系统,其特征在于,所述更新模块用于根据预测的前一时刻已经存在的目标在当前时刻的狄拉克项
    Figure PCTCN2015078026-appb-100064
    及目标矩pk|k-1(x),以及指定的当前时刻新生目标的狄拉克项
    Figure PCTCN2015078026-appb-100065
    及目标矩pγk(x),以及当前时刻的位置测量,利用变系数α-β滤波器确定出当前时刻更新的狄拉克项
    Figure PCTCN2015078026-appb-100066
    i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1,其中,
    Figure PCTCN2015078026-appb-100067
    为更新的狄拉克项的权重系数、
    Figure PCTCN2015078026-appb-100068
    为更新的狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100069
    为更新的狄拉克项产生的时间,nk为当前时刻位置测量的个数,所述当前时刻的位置测量包含由当前时刻目标产生的位置测量和由当前时刻杂波产生的位置测量,i和j是索引号,并且当j≤nk时,
    Figure PCTCN2015078026-appb-100070
    Figure PCTCN2015078026-appb-100071
    Figure PCTCN2015078026-appb-100072
    其中,t表示当前时刻的时间,Ki为增益矩阵,zj表示当前时刻nk个位置测量中的第j个位置测量,Hk为观测矩阵,Rj为观测噪声的方差矩阵,c为放大因子,R0是为探测新目标而设置的协方差,pDk为目标的检测概率,λC为当前时刻观测空间中杂波的密度,当j=nk+1时,
    Figure PCTCN2015078026-appb-100073
    Figure PCTCN2015078026-appb-100074
    Ki取变系数α-β滤波器的增益矩阵,并且
    Figure PCTCN2015078026-appb-100075
    其中,T为当前时刻与前一时刻的时间差,αk和βk是两个时变的系数,并且
    Figure PCTCN2015078026-appb-100076
  8. 根据权利要求7所述的一种基于变系数α-β滤波器的目标跟踪系统,其特征在于,所述裁减与合并模块用于将当前时刻更新的狄拉克项
    Figure PCTCN2015078026-appb-100077
    i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1表述为
    Figure PCTCN2015078026-appb-100078
    q=1,…,Jk|k;其中,q为索引号,取值为1至Jk|k,Jk|k为更新的狄拉克项的个数,
    Figure PCTCN2015078026-appb-100079
    为第q个狄拉克项的权重,
    Figure PCTCN2015078026-appb-100080
    为第q个狄拉克项的目标状态值,
    Figure PCTCN2015078026-appb-100081
    为第q个狄拉克项对应的新生狄拉克项产生的时间,并且Jk|k=(nk+1)(Jk-1+Jγk),q=(Jk-1+Jγk)(j-1)+i,
    Figure PCTCN2015078026-appb-100082
    i=1,2,…,Jk-1+Jγk,j=1,2,…,nk+1;
    从当前时刻更新的狄拉克项
    Figure PCTCN2015078026-appb-100083
    q=1,…,Jk|k中裁减掉权重系数
    Figure PCTCN2015078026-appb-100084
    小于第二阈值的狄拉克项;
    从裁减后余下的狄拉克项中将距离dij小于第三阈值的狄拉克项合并成一个;其中,合并距离dij
    Figure PCTCN2015078026-appb-100085
    多个狄拉克项的合并方法为:
    Figure PCTCN2015078026-appb-100086
    其中,L为合并狄拉克项上标形成的集合,上标T和b分别表示矩阵的转置和合并后狄拉克项的索引号;合并狄拉克项的时间
    Figure PCTCN2015078026-appb-100087
    取合并前狄拉克项中权重系数最大狄拉克项的时间;
    将裁减与合并后的狄拉克项
    Figure PCTCN2015078026-appb-100088
    i=1,…,Jk作为当前时刻的狄拉克项,并将裁减与合并后的狄拉克项的加权和作为当前时刻的目标矩
    Figure PCTCN2015078026-appb-100089
    其中,Jk为当前时刻狄拉克项的个数,i为索引号,取值为1至Jk
PCT/CN2015/078026 2014-07-03 2015-04-30 一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统 Ceased WO2016000487A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201410315478.3A CN104063615B (zh) 2014-07-03 2014-07-03 一种基于变系数α‑β滤波器的目标跟踪方法与跟踪系统
CN201410315478.3 2014-07-03

Publications (1)

Publication Number Publication Date
WO2016000487A1 true WO2016000487A1 (zh) 2016-01-07

Family

ID=51551325

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/078026 Ceased WO2016000487A1 (zh) 2014-07-03 2015-04-30 一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统

Country Status (2)

Country Link
CN (1) CN104063615B (zh)
WO (1) WO2016000487A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111523090A (zh) * 2020-04-24 2020-08-11 商丘师范学院 基于高斯混合概率假设密度的数目时变多目标跟踪方法

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104063615B (zh) * 2014-07-03 2017-02-15 深圳大学 一种基于变系数α‑β滤波器的目标跟踪方法与跟踪系统
CN112731374A (zh) * 2020-12-15 2021-04-30 四川九洲空管科技有限责任公司 一种二次雷达自适应航迹滤波方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103324835A (zh) * 2013-05-30 2013-09-25 深圳大学 概率假设密度滤波器目标信息的保持方法及信息保持系统
CN103390107A (zh) * 2013-07-24 2013-11-13 深圳大学 一种基于狄拉克加权和的目标跟踪方法与目标跟踪系统
CN103679753A (zh) * 2013-12-16 2014-03-26 深圳大学 一种概率假设密度滤波器的轨迹标识方法及轨迹标识系统
CN104063615A (zh) * 2014-07-03 2014-09-24 深圳大学 一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060245500A1 (en) * 2004-12-15 2006-11-02 David Yonovitz Tunable wavelet target extraction preprocessor system
CN102129687B (zh) * 2010-01-19 2014-03-19 中国科学院自动化研究所 动态场景下基于局部背景剪除的自适应目标跟踪方法
CN102685772B (zh) * 2012-04-17 2014-12-24 中国科学院上海微系统与信息技术研究所 一种基于无线全向传感器网络的跟踪节点选择方法
CN103176164B (zh) * 2013-04-11 2016-01-13 北京空间飞行器总体设计部 基于无线传感器网络的多目标无源跟踪方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103324835A (zh) * 2013-05-30 2013-09-25 深圳大学 概率假设密度滤波器目标信息的保持方法及信息保持系统
CN103390107A (zh) * 2013-07-24 2013-11-13 深圳大学 一种基于狄拉克加权和的目标跟踪方法与目标跟踪系统
CN103679753A (zh) * 2013-12-16 2014-03-26 深圳大学 一种概率假设密度滤波器的轨迹标识方法及轨迹标识系统
CN104063615A (zh) * 2014-07-03 2014-09-24 深圳大学 一种基于变系数α-β滤波器的目标跟踪方法与跟踪系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111523090A (zh) * 2020-04-24 2020-08-11 商丘师范学院 基于高斯混合概率假设密度的数目时变多目标跟踪方法
CN111523090B (zh) * 2020-04-24 2023-03-31 商丘师范学院 基于高斯混合概率假设密度的数目时变多目标跟踪方法

Also Published As

Publication number Publication date
CN104063615A (zh) 2014-09-24
CN104063615B (zh) 2017-02-15

Similar Documents

Publication Publication Date Title
CN110503071B (zh) 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
CN105719312B (zh) 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN105182291B (zh) 自适应目标新生强度的phd平滑器的多目标跟踪方法
CN103902829B (zh) 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
CN101975575B (zh) 基于粒子滤波的被动传感器多目标跟踪方法
CN106772353B (zh) 一种适用于闪烁噪声的多目标跟踪方法及系统
CN103324835B (zh) 概率假设密度滤波器目标信息的保持方法及信息保持系统
WO2019047455A1 (zh) 一种适用于闪烁噪声的多机动目标跟踪方法及系统
Li et al. Multi-EAP: Extended EAP for multi-estimate extraction for SMC-PHD filter
CN111127523B (zh) 基于量测迭代更新的多传感器gmphd自适应融合方法
CN104019816A (zh) 基于全局时空相关的概率假设密度滤波器航迹提取方法
CN104318059A (zh) 用于非线性高斯系统的目标跟踪方法和跟踪系统
CN104867163A (zh) 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统
CN108344981B (zh) 面向杂波的多传感器异步检测tsbf多目标跟踪方法
CN106291534B (zh) 一种改进的航迹确认方法
CN106054167B (zh) 基于强度滤波器的多扩展目标跟踪方法
CN104181524B (zh) 一种自适应粒子数的多目标粒子滤波检测前跟踪方法
CN110889862A (zh) 一种网络传输攻击环境中多目标跟踪的组合测量方法
CN115204212A (zh) 一种基于stm-pmbm滤波算法的多目标跟踪方法
CN104680002B (zh) 一种基于随机集理论的分布式融合方法
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
CN103424742B (zh) 一种序贯处理测量数据的目标跟踪方法与目标跟踪系统
CN103390107B (zh) 一种基于狄拉克加权和的目标跟踪方法与目标跟踪系统
CN106168943A (zh) 一种用于跟踪转弯机动目标的方法及其系统
CN113409363B (zh) 一种基于bp-pmbm滤波算法的多目标跟踪方法

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: 15814282

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 15814282

Country of ref document: EP

Kind code of ref document: A1

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 1205 DATED 14/03/2017)

122 Ep: pct application non-entry in european phase

Ref document number: 15814282

Country of ref document: EP

Kind code of ref document: A1