WO2016187870A1 - 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统 - Google Patents

一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统 Download PDF

Info

Publication number
WO2016187870A1
WO2016187870A1 PCT/CN2015/080051 CN2015080051W WO2016187870A1 WO 2016187870 A1 WO2016187870 A1 WO 2016187870A1 CN 2015080051 W CN2015080051 W CN 2015080051W WO 2016187870 A1 WO2016187870 A1 WO 2016187870A1
Authority
WO
WIPO (PCT)
Prior art keywords
edge distribution
current time
existence probability
probability
measurement data
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/CN2015/080051
Other languages
English (en)
French (fr)
Inventor
刘宗香
李丽娟
谢维信
李良群
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
Priority to PCT/CN2015/080051 priority Critical patent/WO2016187870A1/zh
Publication of WO2016187870A1 publication Critical patent/WO2016187870A1/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 measurement driving target tracking method and a tracking system for transmitting edge distribution.
  • the multi-target Bayesian filtering method is an effective method for solving target detection and tracking.
  • the multi-target Bayesian filtering method has the following two problems: First, it is necessary to wait until all the measurement data of one cycle is received before processing the measurement data, so that the sensor can be started. The measurement period is too long, and the measurement data is received at different times in one cycle. The received measurement data is not processed in time during the waiting period, which may cause serious information delay. Second, the recursion of the filter needs to know the initial position of the target. When the target initial position information cannot be obtained, the filter is difficult to use. The delay problem of information processing and the multi-target tracking problem under the initial position of unknown target are the key technical problems that multi-objective Bayesian filtering method needs to explore and solve.
  • the technical problem to be solved by the present invention is to provide a measurement driving target tracking method and a tracking system for transmitting edge distribution, aiming at solving the information delay problem caused by the newly received measurement data not being processed in time and the initial position of the unknown target.
  • Multi-target tracking problem is implemented as follows:
  • a measurement driven target tracking method for transmitting edge distribution includes the following steps:
  • Step 1 When receiving new measurement data, calculate the time difference between the current time and the previous time, and obtain the edge distribution of the current time prediction and the existence probability according to the time difference and the edge distribution of the previous time and the existence probability thereof;
  • Step 2 According to the edge distribution predicted by the current time and its existence probability, the Bayesian rule is used to sequentially process the measurement data received at the current time to obtain the current time. New edge distribution and its probability of existence;
  • Step 3 Use the current time measurement data to generate the edge distribution of the new target at the current time, and assign the existence probability to it, and at the same time, the edge distribution of the new target at the current time and its existence probability are respectively updated with the current time distribution and its existence. Probabilistic merging, generating the edge distribution of the current moment and its existence probability;
  • Step 4 The edge distribution with the probability of existence less than the first threshold is cut off from the edge distribution of the current time generated after the combination, and the reduced edge distribution and its existence probability are used as the recursive input of the next time filter.
  • the edge distribution with the existence probability greater than the second threshold is extracted from the edge distribution after the clipping as the output of the current time, and the mean and variance of each output edge distribution are respectively used as the state estimation and the error estimation of the survival target.
  • the previous time is represented by k-1, k represents the current time, t k-1 represents the time of the previous time, and t k represents the time of the current time;
  • N is a Gaussian distribution
  • x i,k-1 is the state of the i-th edge distribution at the previous moment.
  • m i,k-1 is the mean of the i-th edge distribution at the previous moment
  • P i,k-1 is the variance of the i-th edge distribution at the previous moment
  • n k-1 is the total number of the previous moment target, i Is the index number;
  • the predicted edge distribution of each target at the current moment is N(x i,k ;m i,k
  • k-1 ), i 1,2,...,n k-1 ;
  • the existence probability of each target prediction edge distribution at the current time is ⁇ i,k
  • k-1 F k-1 m i,k-1 Is the mean of the i-th edge distribution at the current time;
  • the surviving probability of the target; ⁇ t t k -t
  • the steps of the Zeiss rule to sequentially process the measurement data received at the current time include:
  • step 3 includes:
  • Step 4 includes: trimming the edge distribution whose existence probability is less than the first threshold from the edge distribution of the current time generated after the combination, the reduced edge The distribution and its existence probability are used as the input of the recursive filter at the next moment.
  • the edge distribution with the existence probability greater than the second threshold is extracted from the reduced edge distribution as the output of the current time.
  • a measurement driven target tracking system that delivers edge distribution including:
  • the prediction module when receiving the new measurement data, calculates the time difference between the current time and the previous time, and obtains the edge distribution of the current time prediction and the existence probability according to the time difference and the edge distribution of the previous time and the existence probability thereof;
  • the update module performs sequential processing on the measurement data received at the current time by using the Bayes rule according to the edge distribution predicted by the current time and its existence probability, and obtains the edge distribution of the current time update and its existence probability;
  • the current time edge distribution generation module generates the edge distribution of the new target at the current time by using the current time measurement data, and assigns the existence probability to the current time, and simultaneously increases the edge distribution of the new target at the current time and the existence probability thereof with the edge of the current time update.
  • the distribution and its existence probability are combined to generate the edge distribution of the current moment and its existence probability;
  • the edge distribution extraction module reduces the edge distribution having a probability less than the first threshold from the edge distribution of the current time generated after the merge, and distributes the edge after the reduction
  • the existence probability is used as the recursive input of the next time filter.
  • the edge distribution with the existence probability greater than the second threshold is extracted from the reduced edge distribution as the output of the current time, and the mean and variance of each output edge distribution are respectively taken as State estimation and error estimates for surviving targets.
  • k-1 represents a previous time
  • k represents a current time
  • t k-1 represents a time of a previous time
  • t k represents a time of the current time
  • N is a Gaussian distribution
  • x i,k-1 is the state of the i-th edge distribution at the previous moment.
  • m i,k-1 is the mean of the i-th edge distribution at the previous moment
  • P i,k-1 is the variance of the i-th edge distribution at the previous moment
  • n k-1 is the total number of the previous moment target, i Is the index number;
  • the predicted edge distribution of each target at the current moment is N(x i,k ;m i,k
  • k-1 ), i 1,2,...,n k-1 ;
  • the existence probability of each target prediction edge distribution at the current time is ⁇ i,k
  • k-1 F k-1 m i,k-1 Is the mean of the i-th edge distribution at the current time;
  • the surviving probability of the target; ⁇ t t k -t
  • the update module is specifically used to:
  • the current time edge distribution generation module is specifically configured to:
  • the edge distribution extraction module is specifically configured to: cut off an edge distribution whose existence probability is less than a first threshold from an edge distribution of a current time generated after the merge, and the edge after the clipping The distribution and its existence probability are used as the input of the recursive filter at the next moment.
  • the edge distribution with the existence probability greater than the second threshold is extracted from the reduced edge distribution as the output of the current time.
  • the present invention has the beneficial effects of: sequentially processing the measurement data at the current time, so that the received data can be processed in time, thereby avoiding the delay of information processing and improving the real-time performance of the target tracking;
  • the measurement data of the moment generates the edge distribution of the new target at the current time, and the applicability of the present invention is expanded without grasping the target initial position information.
  • FIG. 1 is a flowchart of a measurement driving target tracking method for transmitting edge distribution according to an embodiment of the present invention
  • FIG. 2 is a connection block diagram of a measurement driving target tracking system for transmitting edge distribution according to an embodiment of the present invention
  • 3 is measurement data of 50 scan cycles of a sensor according to an embodiment of the present invention.
  • Figure 5 is an average OFAC distance from existing GM-PHD and GM-CPHD filtering methods in accordance with the present invention.
  • the invention obtains the edge distribution and the existence probability of the current time prediction by using the edge distribution obtained by the measurement data of the previous moment measurement data and the existence probability thereof, and sequentially processes the current time reception according to the predicted edge distribution and the existence probability by using the Bayesian rule.
  • the obtained measurement data obtains the edge distribution of the current time update and its existence probability, and uses the measurement data of the current time to generate the edge distribution of the new target at the current time, so that the present invention can receive the current time without grasping the initial position of the target.
  • the measured data is processed in time.
  • a measurement driving target tracking method for transmitting edge distribution includes the following steps:
  • Step 1 When receiving new measurement data, calculate the time difference between the current time and the previous time, and predict the edge distribution and the existence probability of the current time according to the time difference and the edge distribution of the previous time and its existence probability.
  • step 1 k-1 represents the previous time, k represents the current time, t k-1 represents the time of the previous time, and t k represents the time of the current time;
  • N is a Gaussian distribution
  • x i,k-1 is the state of the i-th edge distribution at the previous moment.
  • m i,k-1 is the mean of the i-th edge distribution at the previous moment
  • P i,k-1 is the variance of the i-th edge distribution at the previous moment
  • n k-1 is the total number of the previous moment target, i Is the index number;
  • the existence probability of each edge distribution at the previous moment is N(x i,k ;m i,k
  • the existence probability of each edge distribution at the current time is ⁇ i,k
  • k-1 F k-1 m i,k-1 , is the mean of the i-th edge distribution at the current time;
  • k-1 Q k-1 + F k-1 P i,k-1 F k-1 T , which is the variance of the i-th edge distribution at the current time;
  • Step 2 According to the predicted edge distribution of the current time and its existence probability, the Bayesian rule is used to sequentially process the measurement data received at the current time to obtain the edge distribution and the existence probability of the current time update.
  • the steps of sequentially receiving the measured data for sequential processing include:
  • Step 3 Use the current time measurement data to generate the edge distribution of the new target at the current time, and assign the existence probability to it, and at the same time, the edge distribution of the new target at the current time and its existence probability are respectively updated with the current time distribution and its existence. The probabilities are combined to generate the edge distribution of the current moment and its existence probability.
  • Step 3 specifically includes:
  • Step 4 The edge distribution with the probability of existence less than the first threshold is cut off from the edge distribution of the current time generated after the combination, and the reduced edge distribution and its existence probability are used as the recursive input of the next time filter.
  • the edge distribution with the existence probability greater than the second threshold is extracted from the edge distribution after the clipping as the output of the current time, and the mean and variance of each output edge distribution are respectively used as the state estimation and the error estimation of the survival target.
  • Step 4 includes: trimming the edge distribution whose existence probability is less than the first threshold from the edge distribution of the current time generated after the combination, the edge distribution after clipping and its existence probability As the input of the recursive filter of the next time, at the same time, the edge distribution whose existence probability is greater than the second threshold is extracted from the reduced edge distribution as the output of the current time.
  • the present invention further provides a measurement driving target tracking system for transmitting edge distribution, comprising: a prediction module 201, when receiving new measurement data, calculating a time difference between a current time and a previous time, and according to the time difference The edge distribution and the existence probability of the previous moment are used to predict the edge distribution of the current moment and its existence probability; the updating module 202 uses the Bayesian rule to receive the current time according to the predicted edge distribution of the current moment and its existence probability.
  • the measurement data is sequentially processed to obtain an edge distribution updated at the current time and its existence probability;
  • the current time edge distribution generation module 203 generates the edge distribution of the new target at the current time by using the measurement data of the current time, and specifies the existence probability thereof,
  • the edge distribution of the new target at the current time and its existence probability are respectively merged with the edge distribution updated at the current time and the existence probability thereof to generate an edge distribution of the current time and its existence probability;
  • the edge distribution extraction module 204 generates the generated from the combination.
  • the edge distribution of the current moment will have a probability less than the first
  • the edge distribution of a threshold is cut off, and the reduced edge distribution and its existence probability are used as the recursive input of the next time filter, and the edge distribution with the existence probability greater than the second threshold is extracted from the reduced edge distribution as the current
  • the output of the moment, and the mean and variance of each output edge distribution are respectively used as the state estimation and error estimation of the surviving target.
  • k-1 represents the previous time
  • k represents the current time
  • t k-1 represents the time of the previous time
  • t k represents the time of the current time
  • N is a Gaussian distribution
  • x i,k-1 is the state of the i-th edge distribution at the previous moment.
  • m i,k-1 is the mean of the i-th edge distribution at the previous moment
  • P i,k-1 is the variance of the i-th edge distribution at the previous moment
  • n k-1 is the total number of the previous moment target, i Is the index number;
  • the existence probability of each edge distribution at the previous moment is N(x i,k ;m i,k
  • the existence probability of each edge distribution at the current time is ⁇ i,k
  • k-1 F k-1 m i,k-1 , is the mean of the i-th edge distribution at the current time;
  • k-1 Q k-1 + F k-1 P i,k-1 F k-1 T , which is the variance of the i-th edge distribution at the current time;
  • the surviving probability of the target; ⁇ t t k -t k-1 , which is the
  • the predicted edge distribution at time k is N(x i,k ;m i,k
  • k-1 ), i 1,2,...,n k- 1.
  • the existence probability of each predicted edge distribution is ⁇ i,k
  • k-1 , i 1,2,...,n k-1 ;
  • the update module 202 sequentially processes each measurement data received at the current time by using a Bayes rule, and the processing method is as follows:
  • the edge distribution extraction module 204 is specifically configured to: cut off the edge distribution whose existence probability is less than the first threshold from the edge distribution of the current time generated after the combination, and reduce the edge distribution and the existence probability as the next time filter recursively Input, at the same time, extract the edge distribution whose existence probability is greater than the second threshold from the reduced edge distribution as the output of the current time.
  • the measurement driving target tracking method for transmitting edge distribution of the present invention sequentially processes the measurement data of the current time to obtain the edge distribution of the current time update and its existence probability, and generates the new target of the current time by using the measurement data of the current time.
  • the edge distribution makes it possible to timely process the measurement data received at the current time without grasping the initial position of the target, thereby avoiding the delay of information processing, improving the real-time performance of the target tracking, and expanding the applicability of the present invention.
  • a target of uniform motion in a two-dimensional space [-1000 m, 1000 m] ⁇ [-1000 m, 1000 m] is considered.
  • the target state consists of position and velocity, expressed as Where x and y represent positional components, respectively.
  • the process noise variance matrix is
  • GM-PHD filter Gaussian Mixture probability hypothesis density filter
  • Gaussian Mixture cardinalized probability hypothesis density filter Gaussian Mixture cardinalized probability hypothesis density filter
  • the first threshold is 10 -3 and the second threshold is 0.5.
  • 4 is a result of processing the simulation data of FIG. 3 using the method of the present invention.
  • the present invention is processed with the existing GM-PHD filter and the GM-CPHD filter for the simulation data shown in FIG. 3, and the average OFAP (Optimal Subpattern Assignment) distance obtained by 100 Monte Carlo experiments is as follows.
  • Figure 5 shows. As can be seen from FIG.
  • the multi-target tracking method of the present invention can obtain more in the presence of correlation uncertainty, detection uncertainty, and clutter.
  • its OFAC distance is smaller than the existing SPAC distance obtained by the two methods.

Landscapes

  • Radar Systems Or Details Thereof (AREA)

Abstract

本发明适用于多传感器信息融合领域,提供了一种传递边缘分布的测量驱动目标跟踪方法,包括:根据前一时刻的边缘分布及其存在概率得到当前时刻预测的边缘分布及其存在概率;根据预测的边缘分布及其存在概率,利用贝叶斯规则对当前时刻的测量数据序贯处理得到更新的边缘分布及其存在概率;利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,为其指定存在概率,并将它们分别与更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率;将生成的当前时刻的边缘分布中存在概率小于第一阈值的边缘分布裁减掉,裁减后的边缘分布及其存在概率作为下一时刻递归的输入,同时,提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。

Description

一种传递边缘分布的测量驱动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,尤其涉及一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统。
背景技术
多目标贝叶斯滤波方法是解决目标检测和跟踪的有效方法。然而,在实际应用过程中,我们发现多目标贝叶斯滤波方法存在以下两个问题:一是对测量数据处理时需要等到一个周期的所有测量数据全部收到后才能开始处理,这样,如果传感器的测量周期过长,测量数据在一个周期内不同的时刻接收到,收到的测量数据在等待周期内得不到及时处理会造成严重的信息延迟。二是滤波器的递归需要知道目标的初始位置,当目标初始位置信息无法获取时,滤波器难以使用。信息处理的延迟问题、未知目标初始位置情况下的多目标跟踪问题是多目标贝叶斯滤波方法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传递边缘分布的测量驱动目标跟踪方法与跟踪系统,旨在解决新收到的测量数据不能被及时处理而产生的信息延迟问题以及未知目标初始位置情况下的多目标跟踪问题。本发明是这样实现的:
一种传递边缘分布的测量驱动目标跟踪方法,包括以下步骤:
步骤1:当接收到新测量数据时,计算当前时刻与前一时刻的时间差,并根据该时间差与前一时刻的边缘分布及其存在概率得到当前时刻预测的边缘分布及其存在概率;
步骤2:根据当前时刻预测的边缘分布及其存在概率,利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理,得到当前时刻更 新的边缘分布及其存在概率;
步骤3:利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,并为其指定存在概率,同时,将当前时刻新生目标的边缘分布及其存在概率分别与当前时刻更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率;
步骤4:从合并后所生成的当前时刻的边缘分布中将存在概率小于第一阈值的边缘分布裁减掉,并将裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出,并将各个输出边缘分布的均值与方差分别作为存活目标的状态估计与误差估计。
进一步地,所述步骤1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;
已知前一时刻的边缘分布为N(xi,k-1;mi,k-1,Pi,k-1),i=1,2,…,nk-1;前一时刻各边缘分布的存在概率为ρi,k-1,i=1,2,…,nk-1;其中,N为高斯分布,xi,k-1为前一时刻第i个边缘分布的状态,mi,k-1为前一时刻第i个边缘分布的均值,Pi,k-1为前一时刻第i个边缘分布的方差,nk-1为前一时刻目标的总数,i为索引号;
由前一时刻的边缘分布,前一时刻各边缘分布的存在概率,以及当前时刻与前一时刻的时间差得到当前时刻各目标的预测边缘分布为N(xi,k;mi,k|k-1,Pi,k|k-1),i=1,2,…,nk-1;当前时刻各目标预测边缘分布的存在概率为ρi,k|k-1=pS,k(tk,tk-1i,k-1,i=1,2,…,nk-1;其中,mi,k|k-1=Fk-1mi,k-1,为当前时刻第i个边缘分布的均值;Pi,k|k-1=Qk-1+Fk-1Pi,k-1Fk-1 T,为当前时刻第i个边缘分布的方差;
Figure PCTCN2015080051-appb-000001
为目标的幸存概率;Δt=tk-tk-1,为当前时刻与前一时刻的时间差;δ为已知常数;T为采样周期;
Figure PCTCN2015080051-appb-000002
为前一时刻的状态转移矩阵;Qk-1为 前一时刻的过程噪声方差矩阵;上标T表示矩阵的转置。
进一步地,所述步骤2中,设当前时刻接收到的测量数据为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到的测量数据总数;利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理的步骤包括:
步骤A:取边缘分布
Figure PCTCN2015080051-appb-000003
i=1,2,…,nk-1,和存在概率
Figure PCTCN2015080051-appb-000004
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000005
步骤B:利用贝叶斯规则依次对第1至M个测量数据进行处理:设第j个测量数据在处理前的边缘分布为
Figure PCTCN2015080051-appb-000006
i=1,2,…,nk-1,第j个测量数据在处理前的各边缘分布的存在概率为
Figure PCTCN2015080051-appb-000007
i=1,2,…,nk-1,其中,1≤j≤M;由
Figure PCTCN2015080051-appb-000008
Figure PCTCN2015080051-appb-000009
求得第j个测量数据更新的存在概率
Figure PCTCN2015080051-appb-000010
滤波器增益
Figure PCTCN2015080051-appb-000011
均值向量
Figure PCTCN2015080051-appb-000012
协方差矩阵
Figure PCTCN2015080051-appb-000013
其中,Hk为观测矩阵,Rk为观测噪声的方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度,yj,k为当前时刻接收到的第j个测量数据,I表示单位矩阵,上标T表示矩阵或向量的转置;
如果
Figure PCTCN2015080051-appb-000014
则第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000015
其存在概率为
Figure PCTCN2015080051-appb-000016
其中
Figure PCTCN2015080051-appb-000017
Figure PCTCN2015080051-appb-000018
如果
Figure PCTCN2015080051-appb-000019
第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000020
其存在概率为
Figure PCTCN2015080051-appb-000021
其中
Figure PCTCN2015080051-appb-000022
Figure PCTCN2015080051-appb-000023
步骤C:第M个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000024
i=1,2,…,nk-1,其存在概率为
Figure PCTCN2015080051-appb-000025
i=1,2,…,nk-1
得到的所述当前时刻更新的边缘分布为
Figure PCTCN2015080051-appb-000026
i=1,2,…,nk-1,各更新的边缘分布的存在概率为
Figure PCTCN2015080051-appb-000027
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000028
进一步地,所述步骤3包括:
利用当前时刻的M个测量数据生成当前时刻新生目标的边缘分布
Figure PCTCN2015080051-appb-000029
j=1,2,…,M,同时,为各当前时刻新生目标的边缘分布指定存在概率
Figure PCTCN2015080051-appb-000030
j=1,2,…,M;
将当前时刻更新的边缘分布与当前时刻新生目标的边缘分布合并,生成当前时刻的边缘分布
Figure PCTCN2015080051-appb-000031
将当前时刻更新的边缘分布的存在概率与当前时刻新生目标的边缘分布的存在概率合并,生成当前时刻边缘分布的存在概率
Figure PCTCN2015080051-appb-000032
其中nk=nk-1+M。
进一步地,所述步骤4中,当前时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk,各边缘分布的存在概率为ρi,k,i=1,2,…,nk;步骤4包括:从合并后所生成的当前时刻的边缘分布中裁减掉存在概率小于第一阈值的边缘分布,裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。
一种传递边缘分布的测量驱动目标跟踪系统,包括:
预测模块,当接收到新测量数据时,计算当前时刻与前一时刻的时间差,并根据该时间差与前一时刻的边缘分布及其存在概率得到当前时刻预测的边缘分布及其存在概率;
更新模块,根据当前时刻预测的边缘分布及其存在概率,利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理,得到当前时刻更新的边缘分布及其存在概率;
当前时刻边缘分布生成模块,利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,并为其指定存在概率,同时,将当前时刻新生目标的边缘分布及其存在概率分别与当前时刻更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率;
边缘分布提取模块,从合并后所生成的当前时刻的边缘分布中将存在概率小于第一阈值的边缘分布裁减掉,并将裁减后的边缘分布及 其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出,并将各个输出边缘分布的均值与方差分别作为存活目标的状态估计与误差估计。
进一步地,在所述预测模块中,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;
已知前一时刻的边缘分布为N(xi,k-1;mi,k-1,Pi,k-1),i=1,2,…,nk-1;前一时刻各边缘分布的存在概率为ρi,k-1,i=1,2,…,nk-1;其中,N为高斯分布,xi,k-1为前一时刻第i个边缘分布的状态,mi,k-1为前一时刻第i个边缘分布的均值,Pi,k-1为前一时刻第i个边缘分布的方差,nk-1为前一时刻目标的总数,i为索引号;
由前一时刻的边缘分布,前一时刻各边缘分布的存在概率,以及当前时刻与前一时刻的时间差得到当前时刻各目标的预测边缘分布为N(xi,k;mi,k|k-1,Pi,k|k-1),i=1,2,…,nk-1;当前时刻各目标预测边缘分布的存在概率为ρi,k|k-1=pS,k(tk,tk-1i,k-1,i=1,2,…,nk-1;其中,mi,k|k-1=Fk-1mi,k-1,为当前时刻第i个边缘分布的均值;Pi,k|k-1=Qk-1+Fk-1Pi,k-1Fk-1 T,为当前时刻第i个边缘分布的方差;
Figure PCTCN2015080051-appb-000033
为目标的幸存概率;Δt=tk-tk-1,为当前时刻与前一时刻的时间差;δ为已知常数;T为采样周期;
Figure PCTCN2015080051-appb-000034
为前一时刻的状态转移矩阵;Qk-1为前一时刻的过程噪声方差矩阵;上标T表示矩阵的转置。
进一步地,所述更新模块中,设当前时刻接收到的测量数据为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到的测量数据总数;所述更新模块具体用于:
取边缘分布
Figure PCTCN2015080051-appb-000035
i=1,2,…,nk-1,和存在概率
Figure PCTCN2015080051-appb-000036
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000037
以及
利用贝叶斯规则依次对第1至M个测量数据进行处理:设第j个测量数据在处理前的边缘分布为
Figure PCTCN2015080051-appb-000038
i=1,2,…,nk-1,第j个测量数据在处理前的各边缘分布的存在概率为
Figure PCTCN2015080051-appb-000039
i=1,2,…,nk-1,其中,1≤j≤M;由
Figure PCTCN2015080051-appb-000040
Figure PCTCN2015080051-appb-000041
求得第j个测量数据更新的存在概率
Figure PCTCN2015080051-appb-000042
滤波器增益
Figure PCTCN2015080051-appb-000043
均值向量
Figure PCTCN2015080051-appb-000044
协方差矩阵
Figure PCTCN2015080051-appb-000045
其中,Hk为观测矩阵,Rk为观测噪声的方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度,yj,k为当前时刻接收到的第j个测量数据,I表示单位矩阵,上标T表示矩阵或向量的转置;
如果
Figure PCTCN2015080051-appb-000046
则第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000047
其存在概率为
Figure PCTCN2015080051-appb-000048
其中
Figure PCTCN2015080051-appb-000049
Figure PCTCN2015080051-appb-000050
如果
Figure PCTCN2015080051-appb-000051
第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000052
其存在概率为
Figure PCTCN2015080051-appb-000053
其中
Figure PCTCN2015080051-appb-000054
Figure PCTCN2015080051-appb-000055
第M个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000056
i=1,2,…,nk-1,其存在概率为
Figure PCTCN2015080051-appb-000057
i=1,2,…,nk-1
得到的所述当前时刻更新的边缘分布为
Figure PCTCN2015080051-appb-000058
i=1,2,…,nk-1,各更新的边缘分布的存在概率为
Figure PCTCN2015080051-appb-000059
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000060
进一步地,所述当前时刻边缘分布生成模块具体用于:
利用当前时刻的M个测量数据生成当前时刻新生目标的边缘分布
Figure PCTCN2015080051-appb-000061
j=1,2,…,M,同时,为各当前时刻新生目标的边缘分布指定存在概率
Figure PCTCN2015080051-appb-000062
j=1,2,…,M;以及
将当前时刻更新的边缘分布与当前时刻新生目标的边缘分布合并,生成当前时刻的边缘分布
Figure PCTCN2015080051-appb-000063
将当前时刻更新的边缘分布的存在概率与当前时刻新生目标的边缘分布的存在概率合并,生成当前时刻边缘分布的存在概率
Figure PCTCN2015080051-appb-000064
其中nk=nk-1+M。
进一步地,当前时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk,各边缘分布的存在概率为ρi,k,i=1,2,…,nk;所述边缘分布提取模块具体用于:从合并后所生成的当前时刻的边缘分布中裁减掉存在概率小于第一阈值的边缘分布,裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。
本发明与现有技术相比,有益效果在于:序贯处理当前时刻的测量数据,使得收到的数据能得到及时处理,从而避免了信息处理的延迟,提高了目标跟踪的实时性;利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,无需掌握目标初始位置信息,扩大了本发明的适用性。
附图说明
图1是本发明实施例提供的传递边缘分布的测量驱动目标跟踪方法的流程图;
图2是本发明实施例提供的传递边缘分布的测量驱动目标跟踪系统的连接框图;
图3是本发明实施例提供的传感器50个扫描周期的测量数据;
图4是按照本发明提供的目标跟踪方法处理得到的结果。
图5是按照本发明与现有GM-PHD与GM-CPHD滤波方法的平均OSPA距离。
具体实施方式
为了使本发明的目的、技术方案及优点更加明白,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本发明,并不用于限定本发明。
本发明利用前一时刻测量数据处理后得到的边缘分布及其存在概率得到当前时刻预测的边缘分布及其存在概率,根据预测的边缘分布及其存在概率利用贝叶斯规则序贯处理当前时刻接收到的测量数据得到当前时刻更新的边缘分布及其存在概率,并利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,这样使得本发明在未掌握目标初始位置的情况下能对当前时刻接收到的测量数据进行及时处理。
如图1所示,一种传递边缘分布的测量驱动目标跟踪方法,包括以下步骤:
步骤1:当接收到新测量数据时,计算当前时刻与前一时刻的时间差,并根据该时间差与前一时刻的边缘分布及其存在概率预测当前时刻的边缘分布及其存在概率。
步骤1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;
已知前一时刻的边缘分布为N(xi,k-1;mi,k-1,Pi,k-1),i=1,2,…,nk-1;前一时刻各边缘分布的存在概率为ρi,k-1,i=1,2,…,nk-1;其中,N为高斯分布,xi,k-1为前一时刻第i个边缘分布的状态,mi,k-1为前一时刻第i个边缘分布的均值,Pi,k-1为前一时刻第i个边缘分布的方差,nk-1为前一时刻目标的总数,i为索引号;
由前一时刻的边缘分布,前一时刻各边缘分布的存在概率,以及当前时刻与前一时刻的时间差得到当前时刻各目标的预测边缘分布为N(xi,k;mi,k|k-1,Pi,k|k-1),;当前时刻各边缘分布的存在概率为ρi,k|k-1=pS,k(tk,tk-1i,k-1,i=1,2,…,nk-1;其中,mi,k|k-1=Fk-1mi,k-1,为当前时刻第i个边缘分布的均值;Pi,k|k-1=Qk-1+Fk-1Pi,k-1Fk-1 T,为当前时刻第i个边缘分布的方差;
Figure PCTCN2015080051-appb-000065
为目标的幸存概率;Δt=tk-tk-1,为当前时刻与前一时刻的时间差;δ为已知常数;T为采 样周期;
Figure PCTCN2015080051-appb-000066
为前一时刻的状态转移矩阵;Qk-1为前一时刻的过程噪声方差矩阵;上标T表示状态转移矩阵的转置。
步骤2:根据预测的当前时刻的边缘分布及其存在概率,利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理,得到当前时刻更新的边缘分布及其存在概率。
步骤2中,设当前时刻接收到的测量数据为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到的测量数据总数;利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理的步骤包括:
步骤A:在测量数据处理前,取边缘分布
Figure PCTCN2015080051-appb-000067
i=1,2,…,nk-1,和存在概率
Figure PCTCN2015080051-appb-000068
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000069
步骤B:利用贝叶斯规则依次对第1至M个测量数据进行处理:设第j个测量数据在处理前(或第j-1个测量数据处理后)的边缘分布为
Figure PCTCN2015080051-appb-000070
i=1,2,…,nk-1,第j个测量数据在处理前的各边缘分布的存在概率为
Figure PCTCN2015080051-appb-000071
i=1,2,…,nk-1,其中,1≤j≤M;由
Figure PCTCN2015080051-appb-000072
Figure PCTCN2015080051-appb-000073
求得第j个测量数据更新的存在概率
Figure PCTCN2015080051-appb-000074
滤波器增益
Figure PCTCN2015080051-appb-000075
均值向量
Figure PCTCN2015080051-appb-000076
协方差矩阵
Figure PCTCN2015080051-appb-000077
其中,Hk为观测矩阵,Rk为观测噪声的方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度,yj,k为当前时刻接收到的第j个测量数据,I表示单位矩阵,上标T表示矩阵或向量的转置;
如果
Figure PCTCN2015080051-appb-000078
则第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000079
其存在概率为
Figure PCTCN2015080051-appb-000080
其中
Figure PCTCN2015080051-appb-000081
Figure PCTCN2015080051-appb-000082
如果
Figure PCTCN2015080051-appb-000083
第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000084
其存在概率为
Figure PCTCN2015080051-appb-000085
其中
Figure PCTCN2015080051-appb-000086
Figure PCTCN2015080051-appb-000087
第M个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000088
i=1,2,…,nk-1,其存在概率为
Figure PCTCN2015080051-appb-000089
i=1,2,…,nk-1
得到的当前时刻更新的边缘分布为第M个测量数据处理后的边缘分布,即
Figure PCTCN2015080051-appb-000090
i=1,2,…,nk-1,各更新的边缘分布的存在概率为
Figure PCTCN2015080051-appb-000091
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000092
步骤3:利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,并为其指定存在概率,同时,将当前时刻新生目标的边缘分布及其存在概率分别与当前时刻更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率。
步骤3具体包括:
利用当前时刻的M个测量数据生成当前时刻新生目标的边缘分布
Figure PCTCN2015080051-appb-000093
j=1,2,…,M,同时,为各当前时刻新生目标的边缘分布指定存在概率
Figure PCTCN2015080051-appb-000094
j=1,2,…,M;将当前时刻更新的边缘分布与当前时刻新生目标的边缘分布合并,生成当前时刻的边缘分布
Figure PCTCN2015080051-appb-000095
将当前时刻更新的边缘分布的存在概率与当前时刻新生目标的边缘分布的存在概率合并,生成当前时刻边缘分布的存在概率
Figure PCTCN2015080051-appb-000096
其中nk=nk-1+M。
步骤4:从合并后所生成的当前时刻的边缘分布中将存在概率小于第一阈值的边缘分布裁减掉,并将裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出,并将各个输出边缘分布的均值与方差分别作为存活目标的状态估计与误差估计。步骤4中,当前时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk,各边缘分布的存在概率为ρi,k,i=1,2,…,nk;步骤4包括:从合并后所 生成的当前时刻的边缘分布中裁减掉存在概率小于第一阈值的边缘分布,裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。
如图2所示,本发明还提供一种传递边缘分布的测量驱动目标跟踪系统,包括:预测模块201,当接收到新测量数据时,计算当前时刻与前一时刻的时间差,并根据该时间差与前一时刻的边缘分布及其存在概率预测当前时刻的边缘分布及其存在概率;更新模块202,根据预测的当前时刻的边缘分布及其存在概率,利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理,得到当前时刻更新的边缘分布及其存在概率;当前时刻边缘分布生成模块203,利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,并为其指定存在概率,同时,将当前时刻新生目标的边缘分布及其存在概率分别与当前时刻更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率;边缘分布提取模块204,从合并后所生成的当前时刻的边缘分布中将存在概率小于第一阈值的边缘分布裁减掉,并将裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出,并将各个输出边缘分布的均值与方差分别作为存活目标的状态估计与误差估计。
在预测模块201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;
已知前一时刻的边缘分布为N(xi,k-1;mi,k-1,Pi,k-1),i=1,2,…,nk-1;前一时刻各边缘分布的存在概率为ρi,k-1,i=1,2,…,nk-1;其中,N为高斯分布,xi,k-1为前一时刻第i个边缘分布的状态,mi,k-1为前一时刻第i个边缘分布的均值,Pi,k-1为前一时刻第i个边缘分布的方差,nk-1为前一时刻目标的总数,i为索引号;
由前一时刻的边缘分布,前一时刻各边缘分布的存在概率,以及当前时刻与前一时刻的时间差得到当前时刻各目标的预测边缘分布 为N(xi,k;mi,k|k-1,Pi,k|k-1),;当前时刻各边缘分布的存在概率为ρi,k|k-1=pS,k(tk,tk-1i,k-1,i=1,2,…,nk-1;其中,mi,k|k-1=Fk-1mi,k-1,为当前时刻第i个边缘分布的均值;Pi,k|k-1=Qk-1+Fk-1Pi,k-1Fk-1 T,为当前时刻第i个边缘分布的方差;
Figure PCTCN2015080051-appb-000097
为目标的幸存概率;Δt=tk-tk-1,为当前时刻与前一时刻的时间差;δ为已知常数;T为采样周期;
Figure PCTCN2015080051-appb-000098
为前一时刻的状态转移矩阵;Qk-1为前一时刻的过程噪声方差矩阵;上标T表示状态转移矩阵的转置。
在更新模块202中,k时刻的预测边缘分布为N(xi,k;mi,k|k-1,Pi,k|k-1),i=1,2,…,nk-1,各预测边缘分布的存在概率为ρi,k|k-1,i=1,2,…,nk-1;设当前时刻接收到的测量数据为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到的测量数据总数;更新模块202利用贝叶斯规则序贯地处理当前时刻接收到的各个测量数据,处理方法如下:
(1)在测量数据处理前,取边缘分布
Figure PCTCN2015080051-appb-000099
i=1,2,…,nk-1,和存在概率
Figure PCTCN2015080051-appb-000100
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000101
Figure PCTCN2015080051-appb-000102
(2)利用贝叶斯规则依次对第1至M个测量数据进行处理:设第j个测量数据在处理前(或第j-1个测量数据处理后)的边缘分布为
Figure PCTCN2015080051-appb-000103
i=1,2,…,nk-1,第j个测量数据在处理前的各边缘分布的存在概率为
Figure PCTCN2015080051-appb-000104
i=1,2,…,nk-1,其中,1≤j≤M;由
Figure PCTCN2015080051-appb-000105
Figure PCTCN2015080051-appb-000106
求得第j个测量数据更新的存在概率
Figure PCTCN2015080051-appb-000107
滤波器增益
Figure PCTCN2015080051-appb-000108
均值向量
Figure PCTCN2015080051-appb-000109
协方差矩阵
Figure PCTCN2015080051-appb-000110
其中,Hk为观测矩阵,Rk为观测噪声的 方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度,yj,k为当前时刻接收到的第j个测量数据,I表示单位矩阵,上标T表示矩阵或向量的转置;
如果
Figure PCTCN2015080051-appb-000111
则第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000112
其存在概率为
Figure PCTCN2015080051-appb-000113
其中
Figure PCTCN2015080051-appb-000114
Figure PCTCN2015080051-appb-000115
如果
Figure PCTCN2015080051-appb-000116
第j个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000117
其存在概率为
Figure PCTCN2015080051-appb-000118
其中
Figure PCTCN2015080051-appb-000119
Figure PCTCN2015080051-appb-000120
第M个测量数据处理后的第i个边缘分布为
Figure PCTCN2015080051-appb-000121
i=1,2,…,nk-1,其存在概率为
Figure PCTCN2015080051-appb-000122
i=1,2,…,nk-1
得到的当前时刻更新的边缘分布为第M个测量数据处理后的边缘分布,即
Figure PCTCN2015080051-appb-000123
i=1,2,…,nk-1,各更新的边缘分布的存在概率为
Figure PCTCN2015080051-appb-000124
i=1,2,…,nk-1;其中
Figure PCTCN2015080051-appb-000125
在当前时刻边缘分布生成模块203中,当前时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk-1,各更新的边缘分布的存在概率为ρi,k,i=1,2,…,nk-1。当前时刻边缘分布生成模块203利用当前时刻的M个测量数据生成当前时刻新生目标的边缘分布
Figure PCTCN2015080051-appb-000126
j=1,2,…,M,同时,为各当前时刻新生目标的边缘分布指定存在概率
Figure PCTCN2015080051-appb-000127
j=1,2,…,M;以及
将当前时刻更新的边缘分布与当前时刻新生目标的边缘分布合并,生成当前时刻的边缘分布
Figure PCTCN2015080051-appb-000128
将当前时刻更新的边缘分布的存在概率与当前时刻新生目标的边缘分布的存在概率合并,生成当前时刻边缘分布的存在概率
Figure PCTCN2015080051-appb-000129
其中nk=nk-1+M。
在边缘分布提取模块204中,k时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk,各边缘分布的存在概率为ρi,k,i=1,2,…,nk。 边缘分布提取模块204具体用于:从合并后所生成的当前时刻的边缘分布中裁减掉存在概率小于第一阈值的边缘分布,裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。
本发明的一种传递边缘分布的测量驱动目标跟踪方法通过对当前时刻的测量数据序贯处理以获取当前时刻更新的边缘分布及其存在概率,并利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,使得在未掌握目标初始位置的情况下能对当前时刻接收到的测量数据进行及时处理,从而避免了信息处理的延迟,提高了对目标跟踪的实时性,扩大了本发明的适用性。作为本发明的一个实施例,考虑在二维空间[-1000m,1000m]×[-1000m,1000m]中匀速运动的目标。目标状态由位置和速度构成,表示为
Figure PCTCN2015080051-appb-000130
其中x和y分别表示位置分量,
Figure PCTCN2015080051-appb-000131
Figure PCTCN2015080051-appb-000132
分别表示速度分量,上标T表示向量的转置。状态转移矩阵为
Figure PCTCN2015080051-appb-000133
过程噪声方差矩阵为
Figure PCTCN2015080051-appb-000134
其中,Δtk=tk-tk-1表示当前时刻与前一时刻之间的时间差,σv=1ms-2为过程噪声标准差;观测矩阵
Figure PCTCN2015080051-appb-000135
观测噪声方差矩阵
Figure PCTCN2015080051-appb-000136
σw=2m为观测噪声标准差。
为了产生仿真数据,取幸存概率pS,k=1.0、检测概率pD,k=0.8、杂波密度λc,k=5×10-6m-2、过程噪声标准差σv=1m/s2和观测噪声标准差σw=2m。一次实验的仿真观测数据如图3所示。为了处理仿真数据,我们将本发明和高斯混合概率假设密度滤波器(Gaussian Mixture probability hypothesis density filter,GM-PHD滤波器)以及高斯混合势 分布概率假设密度滤波器(Gaussian Mixture cardinalized probability hypothesis density filter,GM-CPHD滤波器)的相关参数设置为pS,k=1.0、pD,k=0.8、λc,k=5×10-6m-2、σv=1ms-2、σw=2m、第一阈值为10-3、第二阈值为0.5。图4为利用本发明的方法对图3中的仿真数据处理后的结果。将本发明与现有GM-PHD滤波器以及GM-CPHD滤波器对图3所示的仿真数据进行处理,100次Monte Carlo实验得到的平均OSPA(Optimal Subpattern Assignment,最优亚模式分配)距离如图5所示。从图5中可看出,与现有的GM-PHD滤波以及GM-CPHD滤波方法相比,本发明的多目标跟踪方法在存在关联不确定、检测不确定和杂波的情况下能获得更为精确和可靠的目标状态估计,绝大多数情况下,其OSPA距离比现有的两种方法得到的OSPA距离要小。
以上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做的任何修改、等同替换和改进等,均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种传递边缘分布的测量驱动目标跟踪方法,其特征在于,包括以下步骤:
    步骤1:当接收到新测量数据时,计算当前时刻与前一时刻的时间差,并根据该时间差与前一时刻的边缘分布及其存在概率得到当前时刻预测的边缘分布及其存在概率;
    步骤2:根据当前时刻预测的边缘分布及其存在概率,利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理,得到当前时刻更新的边缘分布及其存在概率;
    步骤3:利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,并为其指定存在概率,同时,将当前时刻新生目标的边缘分布及其存在概率分别与当前时刻更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率;
    步骤4:从合并后所生成的当前时刻的边缘分布中将存在概率小于第一阈值的边缘分布裁减掉,并将裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出,并将各个输出边缘分布的均值与方差分别作为存活目标的状态估计与误差估计。
  2. 根据权利要求1所述的目标跟踪方法,其特征在于,所述步骤1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;
    已知前一时刻的边缘分布为N(xi,k-1;mi,k-1,Pi,k-1),i=1,2,…,nk-1;前一时刻各边缘分布的存在概率为ρi,k-1,i=1,2,…,nk-1;其中,N为高斯分布,xi,k-1为前一时刻第i个边缘分布的状态,mi,k-1为前一时刻第i个边缘分布的均值,Pi,k-1为前一时刻第i个边缘分布的方差,nk-1为前一时刻目标的总数,i为索引号;
    由前一时刻的边缘分布,前一时刻各边缘分布的存在概率,以及当前时刻与前一时刻的时间差得到当前时刻各目标的预测边缘分布为N(xi,k;mi,k|k-1,Pi,k|k-1),i=1,2,…,nk-1;当前时刻各目标预测边缘分布的 存在概率为ρi,k|k-1=pS,k(tk,tk-1i,k-1,i=1,2,…,nk-1;其中,mi,k|k-1=Fk-1mi,k-1,为当前时刻第i个边缘分布的均值;Pi,k|k-1=Qk-1+Fk-1Pi,k-1Fk-1 T,为当前时刻第i个边缘分布的方差;
    Figure PCTCN2015080051-appb-100001
    为目标的幸存概率;Δt=tk-tk-1,为当前时刻与前一时刻的时间差;δ为已知常数;T为采样周期;
    Figure PCTCN2015080051-appb-100002
    为前一时刻的状态转移矩阵;Qk-1为前一时刻的过程噪声方差矩阵;上标T表示矩阵的转置。
  3. 根据权利要求2所述的目标跟踪方法,其特征在于,所述步骤2中,设当前时刻接收到的测量数据为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到的测量数据总数;利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理的步骤包括:
    步骤A:取边缘分布
    Figure PCTCN2015080051-appb-100003
    i=1,2,…,nk-1,和存在概率
    Figure PCTCN2015080051-appb-100004
    i=1,2,…,nk-1;其中
    Figure PCTCN2015080051-appb-100005
    Figure PCTCN2015080051-appb-100006
    步骤B:利用贝叶斯规则依次对第1至M个测量数据进行处理:设第j个测量数据在处理前的边缘分布为
    Figure PCTCN2015080051-appb-100007
    i=1,2,…,nk-1,第j个测量数据在处理前的各边缘分布的存在概率为
    Figure PCTCN2015080051-appb-100008
    i=1,2,…,nk-1,其中,1≤j≤M;由
    Figure PCTCN2015080051-appb-100009
    Figure PCTCN2015080051-appb-100010
    求得用第j个测量数据更新时的存在概率
    Figure PCTCN2015080051-appb-100011
    滤波器增益
    Figure PCTCN2015080051-appb-100012
    均值向量
    Figure PCTCN2015080051-appb-100013
    协方差矩阵
    Figure PCTCN2015080051-appb-100014
    其中,Hk为观测矩阵,Rk为观测噪声的方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度,yj,k为当前时刻接收到的第j个测量数据,I表示单位矩阵,上标T表示矩阵或向量的转置;
    如果
    Figure PCTCN2015080051-appb-100015
    则第j个测量数据处理后的第i个边缘分布为
    Figure PCTCN2015080051-appb-100016
    其存在概率为
    Figure PCTCN2015080051-appb-100017
    其中
    Figure PCTCN2015080051-appb-100018
    Figure PCTCN2015080051-appb-100019
    如果
    Figure PCTCN2015080051-appb-100020
    第j个测量数据处理后的第i个边缘分布为
    Figure PCTCN2015080051-appb-100021
    其存在概率为
    Figure PCTCN2015080051-appb-100022
    其中
    Figure PCTCN2015080051-appb-100023
    Figure PCTCN2015080051-appb-100024
    步骤C:第M个测量数据处理后的第i个边缘分布为
    Figure PCTCN2015080051-appb-100025
    i=1,2,…,nk-1,其存在概率为
    Figure PCTCN2015080051-appb-100026
    i=1,2,…,nk-1
    由此得到的所述当前时刻更新的边缘分布为
    Figure PCTCN2015080051-appb-100027
    i=1,2,…,nk-1,各更新的边缘分布的存在概率为
    Figure PCTCN2015080051-appb-100028
    i=1,2,…,nk-1;其中
    Figure PCTCN2015080051-appb-100029
    Figure PCTCN2015080051-appb-100030
  4. 根据权利要求3所述的目标跟踪方法,其特征在于,所述步骤3包括:
    利用当前时刻的M个测量数据生成当前时刻新生目标的边缘分布
    Figure PCTCN2015080051-appb-100031
    j=1,2,…,M,同时,为各当前时刻新生目标的边缘分布指定存在概率
    Figure PCTCN2015080051-appb-100032
    j=1,2,…,M;其中,
    Figure PCTCN2015080051-appb-100033
    为给定的第j个新生目标边缘分布的方差,
    Figure PCTCN2015080051-appb-100034
    为第j个新生目标边缘分布的均值,
    Figure PCTCN2015080051-appb-100035
    直接由当前时刻的第j个测量数据
    Figure PCTCN2015080051-appb-100036
    产生,并且
    Figure PCTCN2015080051-appb-100037
    将当前时刻更新的边缘分布与当前时刻新生目标的边缘分布合并,生成当前时刻的边缘分布
    Figure PCTCN2015080051-appb-100038
    将当前时刻更新的边缘分布的存在概率与当前时刻新生目标的边缘分布的存在概率合并,生成当前时刻边缘分布的存在概率
    Figure PCTCN2015080051-appb-100039
    其中nk=nk-1+M。
  5. 根据权利要求4所述的目标跟踪方法,其特征在于,所述步骤4中,当前时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk,各边缘分布的存在概率为ρi,k,i=1,2,…,nk;步骤4包括:从合并后所生成的当前时刻的边缘分布中裁减掉存在概率小于第一阈值的边缘分布,裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输 入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。
  6. 一种传递边缘分布的测量驱动目标跟踪系统,其特征在于,包括:
    预测模块,当接收到新测量数据时,计算当前时刻与前一时刻的时间差,并根据该时间差与前一时刻的边缘分布及其存在概率得到当前时刻预测的边缘分布及其存在概率;
    更新模块,根据当前时刻预测的边缘分布及其存在概率,利用贝叶斯规则对当前时刻接收到的测量数据进行序贯处理,得到当前时刻更新的边缘分布及其存在概率;
    当前时刻边缘分布生成模块,利用当前时刻的测量数据生成当前时刻新生目标的边缘分布,并为其指定存在概率,同时,将当前时刻新生目标的边缘分布及其存在概率分别与当前时刻更新的边缘分布及其存在概率合并,生成当前时刻的边缘分布及其存在概率;
    边缘分布提取模块,从合并后所生成的当前时刻的边缘分布中将存在概率小于第一阈值的边缘分布裁减掉,并将裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出,并将各个输出边缘分布的均值与方差分别作为存活目标的状态估计与误差估计。
  7. 根据权利要求6所述的目标跟踪系统,其特征在于,在所述预测模块中,以k-1表示前一时刻,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;
    已知前一时刻的边缘分布为N(xi,k-1;mi,k-1,Pi,k-1),i=1,2,…,nk-1;前一时刻各边缘分布的存在概率为ρi,k-1,i=1,2,…,nk-1;其中,N为高斯分布,xi,k-1为前一时刻第i个边缘分布的状态,mi,k-1为前一时刻第i个边缘分布的均值,Pi,k-1为前一时刻第i个边缘分布的方差,nk-1为前一时刻目标的总数,i为索引号;
    由前一时刻的边缘分布,前一时刻各边缘分布的存在概率,以及当前时刻与前一时刻的时间差得到当前时刻各目标的预测边缘分布 为N(xi,k;mi,k|k-1,Pi,k|k-1),i=1,2,…,nk-1;当前时刻各目标预测边缘分布的存在概率为ρi,k|k-1=pS,k(tk,tk-1i,k-1,i=1,2,…,nk-1;其中,mi,k|k-1=Fk-1mi,k-1,为当前时刻第i个边缘分布的均值;Pi,k|k-1=Qk-1+Fk-1Pi,k-1Fk-1 T,为当前时刻第i个边缘分布的方差;
    Figure PCTCN2015080051-appb-100040
    为目标的幸存概率;Δt=tk-tk-1,为当前时刻与前一时刻的时间差;δ为已知常数;T为采样周期;
    Figure PCTCN2015080051-appb-100041
    为前一时刻的状态转移矩阵;Qk-1为前一时刻的过程噪声方差矩阵;上标T表示矩阵的转置。
  8. 根据权利要求7所述的目标跟踪系统,其特征在于,所述更新模块中,设当前时刻接收到的测量数据为yk=(y1,k,…,yM,k),其中,M为当前时刻接收到的测量数据总数;所述更新模块具体用于:
    取边缘分布
    Figure PCTCN2015080051-appb-100042
    i=1,2,…,nk-1,和存在概率
    Figure PCTCN2015080051-appb-100043
    i=1,2,…,nk-1;其中
    Figure PCTCN2015080051-appb-100044
    Figure PCTCN2015080051-appb-100045
    以及
    利用贝叶斯规则依次对第1至M个测量数据进行处理:设第j个测量数据在处理前的边缘分布为
    Figure PCTCN2015080051-appb-100046
    i=1,2,…,nk-1,第j个测量数据在处理前的各边缘分布的存在概率为
    Figure PCTCN2015080051-appb-100047
    i=1,2,…,nk-1,其中,1≤j≤M;由
    Figure PCTCN2015080051-appb-100048
    Figure PCTCN2015080051-appb-100049
    求得第j个测量数据更新的存在概率
    Figure PCTCN2015080051-appb-100050
    滤波器增益
    Figure PCTCN2015080051-appb-100051
    均值向量
    Figure PCTCN2015080051-appb-100052
    协方差矩阵
    Figure PCTCN2015080051-appb-100053
    其中,Hk为观测矩阵,Rk为观测噪声的方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度,yj,k为当前时刻接收到的第j个测量数据,I表示单位矩阵,上标T表示矩阵或向量的转置;
    如果
    Figure PCTCN2015080051-appb-100054
    则第j个测量数据处理后的第i个边缘分布为
    Figure PCTCN2015080051-appb-100055
    其存在概率为
    Figure PCTCN2015080051-appb-100056
    其中
    Figure PCTCN2015080051-appb-100057
    Figure PCTCN2015080051-appb-100058
    如果
    Figure PCTCN2015080051-appb-100059
    第j个测量数据处理后的第i个边缘分布为
    Figure PCTCN2015080051-appb-100060
    其存在概率为
    Figure PCTCN2015080051-appb-100061
    其中
    Figure PCTCN2015080051-appb-100062
    Figure PCTCN2015080051-appb-100063
    第M个测量数据处理后的第i个边缘分布为
    Figure PCTCN2015080051-appb-100064
    i=1,2,…,nk-1,其存在概率为
    Figure PCTCN2015080051-appb-100065
    i=1,2,…,nk-1
    得到的所述当前时刻更新的边缘分布为
    Figure PCTCN2015080051-appb-100066
    i=1,2,…,nk-1,各更新的边缘分布的存在概率为
    Figure PCTCN2015080051-appb-100067
    i=1,2,…,nk-1;其中
    Figure PCTCN2015080051-appb-100068
    Figure PCTCN2015080051-appb-100069
  9. 根据权利要求8所述的目标跟踪系统,其特征在于,所述当前时刻边缘分布生成模块具体用于:
    利用当前时刻的M个测量数据生成当前时刻新生目标的边缘分布
    Figure PCTCN2015080051-appb-100070
    j=1,2,…,M,同时,为各当前时刻新生目标的边缘分布指定存在概率
    Figure PCTCN2015080051-appb-100071
    j=1,2,…,M;以及
    将当前时刻更新的边缘分布与当前时刻新生目标的边缘分布合并,生成当前时刻的边缘分布
    Figure PCTCN2015080051-appb-100072
    将当前时刻更新的边缘分布的存在概率与当前时刻新生目标的边缘分布的存在概率合并,生成当前时刻边缘分布的存在概率
    Figure PCTCN2015080051-appb-100073
    其中nk=nk-1+M。
  10. 根据权利要求9所述的目标跟踪系统,其特征在于,当前时刻更新的边缘分布为N(xi,k;mi,k,Pi,k),i=1,2,…,nk,各边缘分布的存在概率为ρi,k,i=1,2,…,nk;所述边缘分布提取模块具体用于:从合并后所生成的当前时刻的边缘分布中裁减掉存在概率小于第一阈值的边缘分布,裁减后的边缘分布及其存在概率作为下一时刻滤波器递归的输入,同时,从裁减后的边缘分布中提取存在概率大于第二阈值的边缘分布作为当前时刻的输出。
PCT/CN2015/080051 2015-05-28 2015-05-28 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统 Ceased WO2016187870A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2015/080051 WO2016187870A1 (zh) 2015-05-28 2015-05-28 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2015/080051 WO2016187870A1 (zh) 2015-05-28 2015-05-28 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统

Publications (1)

Publication Number Publication Date
WO2016187870A1 true WO2016187870A1 (zh) 2016-12-01

Family

ID=57392328

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/080051 Ceased WO2016187870A1 (zh) 2015-05-28 2015-05-28 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统

Country Status (1)

Country Link
WO (1) WO2016187870A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
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
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
CN104318059A (zh) * 2014-09-24 2015-01-28 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

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
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
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
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
CN104318059A (zh) * 2014-09-24 2015-01-28 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

Similar Documents

Publication Publication Date Title
CN105719312B (zh) 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN112154481B (zh) 基于多个测量假设的目标追踪
CN108226920B (zh) 一种基于预测值处理多普勒量测的机动目标跟踪系统及方法
CN106772353B (zh) 一种适用于闪烁噪声的多目标跟踪方法及系统
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
Li et al. Random finite set-based Bayesian filters using magnitude-adaptive target birth intensity
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
US10935653B2 (en) Multi-target tracking method and tracking system applicable to clutter environment
CN107462882B (zh) 一种适用于闪烁噪声的多机动目标跟踪方法及系统
CN104318059B (zh) 用于非线性高斯系统的目标跟踪方法和跟踪系统
WO2018010099A1 (zh) 一种用于跟踪转弯机动目标的方法及其系统
Yang et al. Adaptive probability hypothesis density filter based on variational Bayesian approximation for multi‐target tracking
CN112162244B (zh) 一种相关噪声和随机丢包环境下的事件触发目标跟踪方法
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
CN106291534B (zh) 一种改进的航迹确认方法
CN106896348B (zh) 一种基于概率数据关联的雷达信号分选方法
CN103247057A (zh) 目标-回波-道路网三元数据关联的道路目标多假设跟踪算法
Yang et al. Variational approximation for adaptive extended target tracking in clutter with random matrix
CN106405538A (zh) 一种适用于杂波环境的多目标跟踪方法及跟踪系统
CN106952290B (zh) 一种用于三维空间跟踪转弯机动目标的方法及系统
CN106168943A (zh) 一种用于跟踪转弯机动目标的方法及其系统
Xia et al. An efficient implementation of the extended object trajectory PMB filter using blocked Gibbs sampling
CN104063615B (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
CN109800721B (zh) 一种多目标跟踪方法及系统
WO2016187870A1 (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

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: 15892951

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: 15892951

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 02/02/2018)

122 Ep: pct application non-entry in european phase

Ref document number: 15892951

Country of ref document: EP

Kind code of ref document: A1