WO2018049602A1 - 一种适用于杂波环境的多目标跟踪方法及跟踪系统 - Google Patents

一种适用于杂波环境的多目标跟踪方法及跟踪系统 Download PDF

Info

Publication number
WO2018049602A1
WO2018049602A1 PCT/CN2016/099036 CN2016099036W WO2018049602A1 WO 2018049602 A1 WO2018049602 A1 WO 2018049602A1 CN 2016099036 W CN2016099036 W CN 2016099036W WO 2018049602 A1 WO2018049602 A1 WO 2018049602A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
time
measurement
edge distribution
existing
Prior art date
Application number
PCT/CN2016/099036
Other languages
English (en)
French (fr)
Inventor
刘宗香
李良群
谢维信
Original Assignee
深圳大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳大学 filed Critical 深圳大学
Priority to PCT/CN2016/099036 priority Critical patent/WO2018049602A1/zh
Publication of WO2018049602A1 publication Critical patent/WO2018049602A1/zh
Priority to US16/236,603 priority patent/US10935653B2/en

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/66Radar-tracking systems; Analogous systems
    • G01S13/72Radar-tracking systems; Analogous systems for two-dimensional tracking, e.g. combination of angle and range tracking, track-while-scan radar
    • G01S13/723Radar-tracking systems; Analogous systems for two-dimensional tracking, e.g. combination of angle and range tracking, track-while-scan radar by using numerical data
    • G01S13/726Multiple target tracking
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/02Systems using reflection of radio waves, e.g. primary radar systems; Analogous systems
    • G01S13/50Systems of measurement based on relative movement of target
    • G01S13/52Discriminating between fixed and moving objects or between objects moving at different speeds
    • G01S13/538Discriminating between fixed and moving objects or between objects moving at different speeds eliminating objects that have not moved between successive antenna scans, e.g. area MTi

Definitions

  • the invention relates to the field of multi-sensor information fusion technology, in particular to a multi-target tracking method and a tracking system suitable for a clutter environment.
  • an effective multi-target tracking method for a clutter environment mainly includes: a target tracking method based on a Gaussian mixture probability hypothesis density filter and a measurement-driven target tracking method based on an edge distribution.
  • the main problem of these two target tracking methods is that the amount of calculation is large, and the state estimation cannot be provided in the first few time steps after the new target appears. How to effectively provide the new target in the first few time steps after its appearance State estimation and reduction of computation are a key technical problem that needs to be explored and solved in the multi-objective Bayesian filtering method.
  • an object of the present invention is to provide a multi-target tracking method and system thereof suitable for a clutter environment, which aims to solve the problem that the state estimation cannot be provided in the first few time steps after the emergence of a new target in the prior art. And the problem of large calculations.
  • the invention provides a multi-target tracking method suitable for a clutter environment, which mainly comprises:
  • the prediction step according to the edge distribution and the existence probability of each target at the previous moment, and the time difference between the current moment and the previous moment, predicting the edge distribution and the existence probability of the existing targets existing at the previous moment at the current moment;
  • N represents a Gaussian distribution
  • i 1,2,...N k-1
  • x i, k-1 is the state vector of the target i at time k-1
  • m i, k-1 and P i, k-1 respectively represent the state mean and covariance of the target i at time k-1
  • N k-1 is the former The total number of goals at a time;
  • the edge distribution and existence probabilities of i at time k are N(x i,k ;m i,k
  • k-1 F i,k
  • k-1 Q i,k-1 +F i,k
  • k-1 p S,k ( ⁇ t k ) ⁇ i,k-1 ,F i,k
  • the categorization step determines whether each measurement in the measurement set originates from an existing target at a previous moment according to the predicted edge distribution and the predicted existence probability of the current target at the current moment and the current set of measurement points. And classify them separately;
  • the updating step the predicted edge distribution and the predicted existence probability at the current time according to the existing targets existing at the previous moment, and the measurement of the current target from the existing target, using the Bayes rule to determine the existing targets existing at the previous moment Update edge distribution and update existence probability at the current time;
  • the clipping and extracting steps according to the updated edge distribution and the existing existence probability of the existing targets existing at the previous moment, the target having the existence probability less than the first threshold is cut off, and the edge of the target having the existence probability greater than the second threshold is extracted Distribution as the output of the current time;
  • the supplementary step the extraction of the new target at the current time edge distribution complements the output of the current time, and extracts the state estimation of the new target at the first two moments to supplement the output of the first two moments respectively;
  • a merging step combining an edge distribution and an existence probability of the remaining target after the clipping and the extracting step, respectively, and combining the edge distribution and the existence probability of the new target generated at the current time in the generating step to form a current moment
  • the classifying step specifically includes: predicting an edge distribution N(x i,k ;m i,k
  • the jth measure y j,k in the middle determines whether the measurement y j,k originates from the existing target, and classifies separately;
  • the step of determining whether the measurement y j,k originates from an existing target and separately classifying comprises:
  • Substep A seeking probability
  • H k is the observation matrix
  • R k is the observed noise variance matrix
  • p D,k is the detection probability of the target
  • ⁇ c,k is the clutter density
  • Substep B if The measurement y j,k is classified into other measurement classes;
  • the measurement y j,k is classified into the measurement originating from the existing target, in the measurement set
  • the updating step specifically includes: predicting edge distribution N(x i,k ;m i,k
  • the step of determining the updated edge distribution and the existence probability of each existing target at the current moment by using the Bayes rule includes:
  • the generating step specifically includes: using other measurements at time k Other measurements at time k-1 And other measurements at time k-2 Generate new targets and use the least squares method to estimate the state mean, covariance and edge distribution of the new target at the current time;
  • the other measurement using the k time Other measurements at time k-1 And other measurements at time k-2 The steps to generate new goals include:
  • Sub-step E from Take measurement From Take measurement From Take measurement Calculated
  • 2 represents the 2 norm of the vector
  • Substep F judgment condition v min ⁇ v f, e ⁇ v max , v min ⁇ v g, f ⁇ v max , a g, f, e ⁇ a max and c g, f, e ⁇ c min is satisfied,
  • v min , v max , a max and c min are four given parameters representing the minimum speed, the maximum speed, the maximum acceleration and the minimum value of the cosine of the included angle; if the four conditions are satisfied at the same time, the measurement is used And measurement The state average of the new target at time k by least squares Covariance And edge distribution among them
  • ⁇ w is the standard deviation of the measured noise; meanwhile, the probability of existence of the specified new target is taken as The state of the new target at k-1 is estimated to be among them The state of the new target at time k-2 is estimated as among them
  • the present invention also provides a multi-target tracking system suitable for use in a clutter environment, the system comprising:
  • a prediction module configured to predict an edge distribution and an existence probability of each target existing at a previous moment according to an edge distribution and an existence probability of each target at a previous moment, and a time difference between the current moment and a previous moment;
  • N represents a Gaussian distribution
  • i 1,2,...N k-1
  • x i, k-1 is the state vector of the target i at time k-1
  • m i, k-1 and P i, k-1 respectively represent the state mean and covariance of the target i at time k-1
  • N k-1 is the former The total number of goals at a time;
  • the edge distribution and existence probabilities of i at time k are N(x i,k ;m i,k
  • k-1 F i,k
  • k-1 Q i,k-1 +F i,k
  • k-1 p S,k ( ⁇ t k ) ⁇ i,k-1 ,F i,k
  • a classification module configured to determine, according to a predicted edge distribution and a predicted existence probability of each target existing at a previous moment, and a current set of measurements, whether each measurement in the measurement set originates from a previous moment Goals and categorize separately;
  • An update module configured to determine, according to the predicted edge distribution and the predicted existence probability of the current target at the current moment, and the current moment from the measurement of the existing target, and use the Bayes rule to determine each existing existing moment The updated edge distribution and update existence probability of the target at the current time;
  • the clipping and extracting module is configured to: according to the updated edge distribution and the update existence probability of each target existing at the previous moment, reduce the target whose existence probability is less than the first threshold, and extract the target whose existence probability is greater than the second threshold
  • the edge distribution is the output of the current moment
  • a supplemental module for extracting the edge distribution of the new target at the current time to supplement the output of the current time, and extracting the state estimates of the new target at the first two moments to supplement the outputs of the first two moments respectively;
  • a merging module configured to combine the edge distribution and the existence probability of the remaining target after the clipping and the extracting step, respectively, with the edge distribution and the existence probability of the new target generated at the current time in the generating step, to form The edge distribution and existence probability of each target at the current moment, and as the input of the next recursion.
  • the classification module is specifically configured to: according to the predicted edge distribution N of the target i existing at time k-1 at time k (x i, k ; m i, k
  • the jth measure y j,k in the middle determines whether the measurement y j,k originates from the existing target, and classifies separately;
  • the classification module includes:
  • H k is the observation matrix
  • R k is the observed noise variance matrix
  • p D,k is the detection probability of the target
  • ⁇ c,k is the clutter density
  • Second submodule for The measurement y j,k is classified into other measurement classes;
  • the measurement y j,k is classified into the measurement originating from the existing target, in the measurement set
  • the updating module is specifically configured to: according to the predicted edge distribution N(x i,k ;m i,k
  • the update module includes:
  • the generating module is specifically configured to: use other measurements at time k Other measurements at time k-1 And other measurements at time k-2 Generate new targets and use the least squares method to estimate the state mean, covariance and edge distribution of the new target at the current time;
  • the generating module includes:
  • the fifth submodule for Take measurement From Take measurement From Take measurement Calculated
  • 2 represents the 2 norm of the vector ,
  • v min ⁇ v f, e ⁇ v max , v min ⁇ v g, f ⁇ v max , a g, f, e ⁇ a max and c g, f, e ⁇ c min Satisfied where v min , v max , a max and c min are four given parameters representing the minimum speed, maximum speed, maximum acceleration and the minimum value of the cosine of the included angle; if the four conditions are satisfied at the same time, the measurement is used And measurement
  • the state average of the new target at time k by least squares Covariance And edge distribution among them ⁇ w is the standard deviation of the measured noise; meanwhile, the probability of existence of the specified new target is taken as The state of the new target at k-1 is estimated to be among them The state of the new target at time k-2 is estimated as among them
  • the technical solution provided by the invention effectively estimates the state estimation of the first three time steps after the occurrence of the new target by using the least square method by predicting, classifying, updating, reducing and extracting, generating, supplementing and merging the steps.
  • the existing method can not provide the new target state estimation problem in the first few time steps after the emergence of the new target, and has the characteristics of fast processing speed, and its calculation amount is obviously smaller than the existing method, and has strong practicability.
  • FIG. 1 is a flowchart of a multi-target tracking method applicable to a clutter environment according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram showing the internal structure of a multi-target tracking system suitable for use in a clutter environment according to an embodiment of the present invention
  • FIG. 3 is a measurement data diagram of a sensor provided by an embodiment of the present invention in 50 scanning cycles according to an embodiment of the present invention
  • FIG. 4 is a schematic diagram of an OSPA distance obtained by an experiment in a multi-target tracking method and a Gaussian mixture probability hypothesis density filtering method according to the present invention, which is applicable to a clutter environment according to an embodiment of the present invention;
  • FIG. 5 is a schematic diagram of an average SPAC distance obtained by 100 experiments in a multi-target tracking method and a Gaussian mixture probability hypothesis density filtering method according to the present invention, which is suitable for use in a clutter environment according to an embodiment of the present invention.
  • a multi-target tracking method suitable for a clutter environment provided by the present invention will be described in detail below.
  • FIG. 1 is a flowchart of a multi-target tracking method suitable for a clutter environment according to an embodiment of the present invention.
  • step S1 the prediction step, based on the edge distribution and the existence probability of each target at the previous moment, and the time difference between the current time and the previous time, predict the edge distribution and the existence probability of each target existing at the previous moment at the current time.
  • the predicting step S1 specifically includes:
  • N represents a Gaussian distribution
  • i 1,2,...N k-1
  • x i, K-1 is the state vector of the target i at time k-1
  • m i,k-1 and P i,k-1 respectively represent the state mean and covari
  • the edge distribution and existence probability of i at time k are N(x i,k ;m i,k
  • k-1 F i,k
  • k-1 Q i,k-1 +F i,k
  • k-1 p S,k ( ⁇ t k ) ⁇ i,k-1 ,F i,k
  • step S2 the classification step determines whether each measurement in the measurement set originates from the previous moment according to the predicted edge distribution and the predicted existence probability of the current target at the current time and the current existence measurement set. Existing goals are categorized separately.
  • the classifying step S2 specifically includes: predicting an edge distribution N(x i,k ;m i,k
  • the jth measure y j,k in the middle determines whether the measurement y j,k originates from the existing target, and classifies separately;
  • the step of determining whether the measurement y j,k originates from an existing target and separately classifying comprises:
  • Substep A seeking probability
  • H k is the observation matrix
  • R k is the observed noise variance matrix
  • p D,k is the detection probability of the target
  • ⁇ c,k is the clutter density
  • Substep B if The measurement y j,k is classified into other measurement classes;
  • the measurement y j,k is classified into the measurement originating from the existing target, in the measurement set
  • step S3 the updating step, the predicted edge distribution and the predicted existence probability of the respective targets existing at the previous moment according to the previous moment, and the current moment originating from the existing target are measured, and the Bayes rule is used to determine the previous moment.
  • Each existing target has an updated edge distribution at the current time and an update existence probability.
  • the updating step S3 specifically includes: predicting an edge distribution N(x i,k ;m i,k
  • the step of determining the updated edge distribution and the existence probability of each existing target at the current moment by using the Bayes rule includes:
  • step S4 the clipping and extracting steps, the updated edge distribution and the existing existence probability of the respective targets existing at the previous moment at the current moment, the target having the existence probability less than the first threshold is cut off, and the extraction probability is greater than the second
  • the edge distribution of the target of the threshold is taken as the output of the current time.
  • step S5 the generating step generates new targets using other measurements of the current time and other measurements of the first two moments, and estimates the state mean, covariance, and edge distribution of the new target at the current time using a least squares method.
  • the generating step S5 specifically includes: using other measurements at time k Other measurements at time k-1 And other measurements at time k-2 Generate new targets and use the least squares method to estimate the state mean, covariance and edge distribution of the new target at the current time;
  • the other measurement using the k time Other measurements at time k-1 And other measurements at time k-2 The steps to generate new goals include:
  • Sub-step E from Take measurement From Take measurement From Take measurement Calculated
  • 2 represents the 2 norm of the vector
  • step S6 the supplemental step extracts the edge distribution of the new target at the current time to supplement the output of the current time, and extracts the state estimates of the new target at the first two moments to supplement the outputs of the first two moments respectively.
  • step S7 the merging step, the edge distribution and the existence probability of the remaining target after the clipping in the clipping and extracting steps are respectively performed with the edge distribution and the existence probability of the new target generated at the current time in the generating step Merging, forming the edge distribution and existence probability of each target at the current moment, and as the input of the next recursion.
  • the invention provides a multi-target tracking method suitable for a clutter environment, which uses the steps of predicting, classifying, updating, reducing and extracting, generating, supplementing and merging, and using the least squares method to estimate the initial target of the new target after its appearance 3
  • the state estimation of time steps effectively solves the problem that the existing method can not provide new target state estimation in the first few time steps after the emergence of the new target, and has the characteristics of fast processing speed, and the calculation amount is obviously smaller than the existing method. , has a strong practicality.
  • FIG. 2 a schematic structural diagram of a multi-target tracking system 10 suitable for use in a clutter environment according to an embodiment of the present invention is shown.
  • the multi-target tracking system 10 suitable for the clutter environment mainly includes a prediction module 11, a classification module 12, an update module 13, a reduction and extraction module 14, a generation module 15, a supplementation module 16, and a merge module 17.
  • the prediction module 11 is configured to predict the edge distribution and the existence probability of each target existing at the previous moment according to the edge distribution and the existence probability of each target at the previous moment, and the time difference between the current moment and the previous moment.
  • the prediction module 11 is specifically configured to:
  • N represents a Gaussian distribution
  • i 1,2,...N k-1
  • x i, K-1 is the state vector of the target i at time k-1
  • m i,k-1 and P i,k-1 respectively represent the state mean and covari
  • the edge distribution and existence probabilities of i at time k are N(x i,k ;m i,k
  • k-1 F i,k
  • k-1 Q i,k-1 +F i,k
  • k-1 p S,k ( ⁇ t k ) ⁇ i,k-1 ,F i,k
  • the classification module 12 is configured to determine, according to the predicted edge distribution and the predicted existence probability of the current target at the current moment, and the measurement set of the current time, whether each measurement in the measurement set originates from the previous moment. of Goals and categorize separately.
  • the classification module 12 is specifically configured to: according to the predicted edge distribution N of the target i existing at time k-1 at time k (x i, k ; m i, k
  • the jth measure y j,k in the middle determines whether the measurement y j,k originates from the existing target, and classifies separately;
  • the classification module 12 includes: a first sub-module and a second sub-module.
  • H k is the observation matrix
  • R k is the observed noise variance matrix
  • p D,k is the detection probability of the target
  • ⁇ c,k is the clutter density.
  • Second submodule for The measurement y j,k is classified into other measurement classes;
  • the measurement y j,k is classified into the measurement originating from the existing target, in the measurement set
  • the updating module 13 is configured to determine, according to the predicted edge distribution and the predicted existence probability of the current target at the current moment, and the current moment from the measurement of the existing target, and use the Bayes rule to determine that the previous moment exists.
  • the update edge distribution and update existence probability of each target at the current time are configured to determine, according to the predicted edge distribution and the predicted existence probability of the current target at the current moment, and the current moment from the measurement of the existing target, and use the Bayes rule to determine that the previous moment exists.
  • the updating module 13 is specifically configured to: according to the predicted edge distribution N (x i, k ; m i, k
  • the update module 13 includes: a third submodule and a fourth submodule.
  • the third submodule for measuring with Bayesian rules Processing obtaining the target i corresponding to the measurement Probability of existence Mean vector And covariance matrix among them
  • the clipping and extraction module 14 is configured to: according to the updated edge distribution and the update existence probability of the current target existing at the previous moment, the target having the existence probability less than the first threshold is cut off, and the existence probability is greater than the second threshold.
  • the edge distribution of the target is used as the output of the current moment.
  • the generating module 15 is configured to generate a new target by using other measurements of the current time and other measurements of the first two moments thereof, and estimate a state mean, a covariance, and an edge distribution of the new target at the current time by using a least squares method.
  • the generating module 15 is specifically configured to: use other measurements at time k Other measurements at time k-1 And other measurements at time k-2 Generate new targets and use the least squares method to estimate the state mean, covariance and edge distribution of the new target at the current time;
  • the generating module 15 includes: a fifth sub-module and a sixth sub-module.
  • the fifth submodule for Take measurement From Take measurement From Take measurement Calculated
  • 2 represents the 2 norm of the vector
  • indicates an absolute value
  • ( ⁇ , ⁇ ) indicates the inner product of the two vectors.
  • v min ⁇ v f, e ⁇ v max , v min ⁇ v g, f ⁇ v max , a g, f, e ⁇ a max and c g, f, e ⁇ c min Satisfied where v min , v max , a max and c min are four given parameters representing the minimum speed, maximum speed, maximum acceleration and the minimum value of the cosine of the included angle; if the four conditions are satisfied at the same time, the measurement is used And measurement
  • the state average of the new target at time k by least squares Covariance And edge distribution among them ⁇ w is the standard deviation of the measured noise; meanwhile, the probability of existence of the specified new target is taken as The state of the new target at k-1 is estimated to be among them The state of the new target at time k-2 is estimated as among them
  • the supplementing module 16 is configured to extract the edge distribution of the new target at the current time to supplement the output of the current time, and extract the state estimation of the new target at the first two moments to supplement the output of the first two moments respectively.
  • a merging module 17 for dividing the edge distribution and the existence probability of the remaining target after the clipping and extracting steps are performed, The edge distribution and the existence probability of the new target generated at the current time are respectively combined with the new target generated in the generating step, and the edge distribution and the existence probability of each target at the current time are formed, and are used as the input of the next recursion.
  • the present invention provides a multi-target tracking system 10 suitable for use in a clutter environment, through the prediction module 11, the classification module 12, the update module 13, the reduction and extraction module 14, the generation module 15, the supplementation module 16, and the merge module 17.
  • Estimating the state of the first three time steps of the new target after the occurrence of the new target by using the least squares method effectively solving the problem that the existing method cannot provide the new target state estimation in the first few time steps after the new target appears,
  • the processing speed is fast, and the calculation amount is obviously smaller than the existing method, and has strong practicability.
  • the state of the target is composed of position and velocity, expressed as Where x and y represent positional components, respectively.
  • the superscript T represents the transpose of the vector
  • the simulated observation data of the sensor in 50 scan cycles in one experiment is shown in Fig. 3.
  • Figure 5 is the average OFPA distance obtained by performing 50 Monte Carlo experiments with the existing Gaussian mixture probability hypothesis density filter and the present invention.
  • Table 1 shows the average execution time of the existing Gaussian mixture probability hypothesis density filter and one experiment obtained in 50 experiments of the present invention, and the results show that the average execution time of the present invention is significantly smaller than the existing Gaussian mixture probability hypothesis density filter. Device.
  • the technical solution provided by the invention effectively estimates the state estimation of the first three time steps after the occurrence of the new target by using the least square method by predicting, classifying, updating, reducing and extracting, generating, supplementing and merging the steps.
  • the existing method can not provide the new target state estimation problem in the first few time steps after the emergence of the new target, and has the characteristics of fast processing speed, and its calculation amount is obviously smaller than the existing method, and has strong practicability.
  • each unit included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding function can be implemented; in addition, the specific name of each functional unit is also They are only used to facilitate mutual differentiation and are not intended to limit the scope of the present invention.

Landscapes

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

Abstract

一种适用于杂波环境的多目标跟踪方法,其中,所述方法包括:预测步骤(S1)、分类步骤(S2)、更新步骤(S3)、裁减与提取步骤(S4)、生成步骤(S5)、补充步骤(S6)、合并步骤(S7)。还提供一种适用于杂波环境的多目标跟踪系统(10)。具有处理速度快的特点,同时有效地解决了现有方法在新目标出现后的前几个时间步不能提供新目标状态估计的问题。

Description

一种适用于杂波环境的多目标跟踪方法及跟踪系统 技术领域
本发明涉及多传感器信息融合技术领域,尤其涉及一种适用于杂波环境的多目标跟踪方法及跟踪系统。
背景技术
贝叶斯滤波技术能够提供一种强大的统计方法工具,用于协助解决杂波环境下以及测量数据具有不确定性情况下的多传感器信息的融合与处理。在现有技术中,用于杂波环境的多目标跟踪有效方法主要有:基于高斯混合概率假设密度滤波器的目标跟踪方法和传递边缘分布的测量驱动目标跟踪方法。这两种目标跟踪方法的主要问题是计算量较大,并且在新目标出现后的最初几个时间步中不能提供其状态估计,如何有效提供新目标在其出现后的最初几个时间步的状态估计,同时减少计算量是多目标贝叶斯滤波方法中需要探索和解决的一个关键技术问题。
发明内容
有鉴于此,本发明的目的在于提供一种适用于杂波环境的多目标跟踪方法及其系统,旨在解决现有技术中在新目标出现后的最初几个时间步中不能提供其状态估计以及计算量大的问题。
本发明提出一种适用于杂波环境的多目标跟踪方法,主要包括:
预测步骤、根据前一时刻各个目标的边缘分布和存在概率,以及当前时刻与前一时刻的时间差,预测前一时刻已存在的各个目标在当前时刻的边缘分布和存在概率;
其中,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,k-1时刻目标i的边缘分布和存在概率分别表示为N(xi,k-1;mi,k-1,Pi,k-1)和ρi,k-1,其中N表示高斯分布,i=1,2,…Nk-1,xi,k-1为k-1时刻目标i的状态向量,mi,k-1和Pi,k-1分别表示k-1时刻目标i的状态均值和协方差,Nk-1为前一时刻目标的总数;
由k-1时刻目标i的边缘分布N(xi,k-1;mi,k-1,Pi,k-1)和存在概率ρi,k-1,预测k-1时刻的目标i在k时刻的边缘分布和存在概率分别为N(xi,k;mi,k|k-1,Pi,k|k-1)和ρi,k|k-1,其中mi,k|k-1=Fi,k|k-1mi,k-1,Pi,k|k-1=Qi,k-1+Fi,k|k-1Pi,k-1(Fi,k|k-1)T,ρi,k|k-1=pS,k(Δtki,k-1,Fi,k|k-1为状态转移矩阵,上标T表示矩阵或向量的转置,Δtk=tk-tk-1为k时刻与k-1时刻的时间差,Qi,k-1为k-1时刻目标i的过程噪声协方差矩阵,pS,k(Δtk)为目标的幸存概率,且
Figure PCTCN2016099036-appb-000001
T为采样周期,δ为给定的常数,i=1,2,…Nk-1
分类步骤、根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻的测量集,确定出测量集中的每个测量是否源于前一时刻已存在的目标,并分别进行归类;
更新步骤、根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻源于已存在目标的测量,利用贝叶斯规则确定前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率;
裁减与提取步骤、根据前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率,将存在概率小于第一阈值的目标裁减掉,同时提取存在概率大于第二阈值的目标的边缘分布作为当前时刻的输出;
生成步骤、利用当前时刻的其它测量和其前两时刻的其它测量产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
补充步骤、提取新目标在当前时刻的边缘分布对当前时刻的输出进行补充,并提取新目标在前两个时刻的状态估计分别对前两个时刻的输出进行补充;
合并步骤、将在所述裁减与提取步骤中裁减后余下目标的边缘分布和存在概率,分别与在所述生成步骤中生成的新目标在当前时刻的边缘分布和存在概率进行合并,形成当前时刻各个目标的边缘分布和存在概率,并作为下一次递归的输入。
优选的,所述分类步骤具体包括:根据k-1时刻已存在的目标i在k时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及k时刻的测量集合
Figure PCTCN2016099036-appb-000002
中的第j个测量yj,k,确定测量yj,k是否源于已存在目标,并分别进行归类;
其中,所述确定测量yj,k是否源于已存在目标,并分别进行归类的步骤包括:
子步骤A、求取概率
Figure PCTCN2016099036-appb-000003
其中,Hk为观测矩阵,Rk为观测噪声方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度;
子步骤B、若
Figure PCTCN2016099036-appb-000004
将测量yj,k归入其它测量类;若
Figure PCTCN2016099036-appb-000005
将测量yj,k归入源于已存在目标的测量,在测量集合
Figure PCTCN2016099036-appb-000006
中的每个测量处理后,测量集合中yk的测量被分为两类,一类是源于已存在目标的测量,表示为
Figure PCTCN2016099036-appb-000007
另一类是其它测量,表示为
Figure PCTCN2016099036-appb-000008
其中M1,k和M2,k分别源于已存在目标测量的数目和其它测量的数目,且M1,k+M2,k=Mk
优选的,所述更新步骤具体包括:根据前一时刻已存在的各个目标在当前时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及当前时刻源于已存在目标的测量集合
Figure PCTCN2016099036-appb-000009
利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率;
其中,所述利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率的步骤包括:
子步骤C、利用贝叶斯规则对测量
Figure PCTCN2016099036-appb-000010
处理,得到目标i对应于测量
Figure PCTCN2016099036-appb-000011
的存在概率
Figure PCTCN2016099036-appb-000012
均值向量
Figure PCTCN2016099036-appb-000013
和协方差矩阵
Figure PCTCN2016099036-appb-000014
其中
Figure PCTCN2016099036-appb-000015
在所有的M1,k个测量处理后,各个目标对应于各测量的更新边缘分布和存在概率分别为
Figure PCTCN2016099036-appb-000016
Figure PCTCN2016099036-appb-000017
其中i=1,…,Nk-1,j=1,…,M1,k
子步骤D、设
Figure PCTCN2016099036-appb-000018
其中
Figure PCTCN2016099036-appb-000019
则k时刻目标i的更新边缘分布取为
Figure PCTCN2016099036-appb-000020
相应的存在概率取为
Figure PCTCN2016099036-appb-000021
其中i=1,…,Nk-1,当q=M1,k+1时有
Figure PCTCN2016099036-appb-000022
优选的,所述生成步骤具体包括:利用k时刻的其它测量
Figure PCTCN2016099036-appb-000023
k-1时刻的其它测量
Figure PCTCN2016099036-appb-000024
和k-2时刻的其它测量
Figure PCTCN2016099036-appb-000025
产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
其中,所述利用k时刻的其它测量
Figure PCTCN2016099036-appb-000026
k-1时刻的其它测量
Figure PCTCN2016099036-appb-000027
和k-2时刻的其它测量
Figure PCTCN2016099036-appb-000028
产生新目标的步骤包括:
子步骤E、从
Figure PCTCN2016099036-appb-000029
中取测量
Figure PCTCN2016099036-appb-000030
Figure PCTCN2016099036-appb-000031
中取测量
Figure PCTCN2016099036-appb-000032
Figure PCTCN2016099036-appb-000033
中取测量
Figure PCTCN2016099036-appb-000034
计算得到
Figure PCTCN2016099036-appb-000035
Figure PCTCN2016099036-appb-000036
Figure PCTCN2016099036-appb-000037
其中e=1,…,M2,k-2,f=1,…,M2,k-1,g=1,…,M2,k,||·||2表示向量的2范数,|·|表示取绝对值,(·,·)表示两向量的内积;
子步骤F、判断条件vmin≤vf,e≤vmax、vmin≤vg,f≤vmax、ag,f,e≤amax和cg,f,e≥cmin是否满足,其中vmin、vmax、amax和cmin为4个给定的参数,分别表示最小速度、最大速度、最大加速度和夹角余弦的最小值;若4个条件同时满足,利用测量
Figure PCTCN2016099036-appb-000038
和测量
Figure PCTCN2016099036-appb-000039
由最小二乘法得到一个新目标的在k时刻的状态均值
Figure PCTCN2016099036-appb-000040
协方差
Figure PCTCN2016099036-appb-000041
和边缘分布
Figure PCTCN2016099036-appb-000042
其中
Figure PCTCN2016099036-appb-000043
Figure PCTCN2016099036-appb-000044
Figure PCTCN2016099036-appb-000045
σw为测量噪声的标准差;同时,指定新目标的存在概率取为
Figure PCTCN2016099036-appb-000046
新目标在k-1时刻的状态估计为
Figure PCTCN2016099036-appb-000047
其中
Figure PCTCN2016099036-appb-000048
新目标在k-2时刻的状态估计为
Figure PCTCN2016099036-appb-000049
其中
Figure PCTCN2016099036-appb-000050
另一方面,本发明还提供一种适用于杂波环境的多目标跟踪系统,所述系统包括:
预测模块,用于根据前一时刻各个目标的边缘分布和存在概率,以及当前时刻与前一时刻的时间差,预测前一时刻已存在的各个目标在当前时刻的边缘分布和存在概率;
其中,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,k-1时刻目标i的边缘分布和存在概率分别表示为N(xi,k-1;mi,k-1,Pi,k-1)和ρi,k-1,其中N表示高斯分布,i=1,2,…Nk-1,xi,k-1为k-1时刻目标i的状态向量,mi,k-1和Pi,k-1分别表示k-1时刻目标i的状态均值和协方差,Nk-1为前一时刻目标的总数;
由k-1时刻目标i的边缘分布N(xi,k-1;mi,k-1,Pi,k-1)和存在概率ρi,k-1,预测k-1时刻的目标i在k时刻的边缘分布和存在概率分别为N(xi,k;mi,k|k-1,Pi,k|k-1)和ρi,k|k-1,其中mi,k|k-1=Fi,k|k-1mi,k-1,Pi,k|k-1=Qi,k-1+Fi,k|k-1Pi,k-1(Fi,k|k-1)T,ρi,k|k-1=pS,k(Δtki,k-1,Fi,k|k-1为状态转移矩阵,上标T表示矩阵或向量的转置,Δtk=tk-tk-1为k时刻与k-1时刻的时间差,Qi,k-1为k-1时刻目标i的过程噪声协方差矩阵,pS,k(Δtk)为目标的幸存概率,且
Figure PCTCN2016099036-appb-000051
T为采样周期,δ为给定的常数,i=1,2,…Nk-1
分类模块,用于根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻的测量集,确定出测量集中的每个测量是否源于前一时刻已存在的目标,并分别进行归类;
更新模块,用于根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻源于已存在目标的测量,利用贝叶斯规则确定前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率;
裁减与提取模块,用于根据前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率,将存在概率小于第一阈值的目标裁减掉,同时提取存在概率大于第二阈值的目标的边缘分布作为当前时刻的输出;
生成模块,用于利用当前时刻的其它测量和其前两时刻的其它测量产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
补充模块,用于提取新目标在当前时刻的边缘分布对当前时刻的输出进行补充,并提取新目标在前两个时刻的状态估计分别对前两个时刻的输出进行补充;
合并模块,用于将在所述裁减与提取步骤中裁减后余下目标的边缘分布和存在概率,分别与在所述生成步骤中生成的新目标在当前时刻的边缘分布和存在概率进行合并,形成当前时刻各个目标的边缘分布和存在概率,并作为下一次递归的输入。
优选的,所述分类模块具体用于:根据k-1时刻已存在的目标i在k时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及k时刻的测量集合
Figure PCTCN2016099036-appb-000052
中的第j个测量yj,k,确定测量yj,k是否源于已存在目标,并分别进行归类;
其中,所述分类模块包括:
第一子模块,用于求取概率
Figure PCTCN2016099036-appb-000053
其中,Hk为观测矩阵,Rk为观测噪声方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度;
第二子模块,用于若
Figure PCTCN2016099036-appb-000054
将测量yj,k归入其它测量类;若
Figure PCTCN2016099036-appb-000055
将测量yj,k归入源于已存在目标的测量,在测量集合
Figure PCTCN2016099036-appb-000056
中的每个测量处理后,测量集合中yk的测量被分为两类,一类是源于已存在目标的测量,表示为
Figure PCTCN2016099036-appb-000057
另一类是其它测量,表示为
Figure PCTCN2016099036-appb-000058
其中M1,k和M2,k分别源于已存在目标测量的数目和其它测量的数目,且M1,k+M2,k=Mk
优选的,所述更新模块具体用于:根据前一时刻已存在的各个目标在当前时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及当前时刻源于已存在目标的测量 集合
Figure PCTCN2016099036-appb-000059
利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率;
其中,所述更新模块包括:
第三子模块,用于利用贝叶斯规则对测量
Figure PCTCN2016099036-appb-000060
处理,得到目标i对应于测量
Figure PCTCN2016099036-appb-000061
的存在概率
Figure PCTCN2016099036-appb-000062
均值向量
Figure PCTCN2016099036-appb-000063
和协方差矩阵
Figure PCTCN2016099036-appb-000064
其中
Figure PCTCN2016099036-appb-000065
在所有的M1,k个测量处理后,各个目标对应于各测量的更新边缘分布和存在概率分别为
Figure PCTCN2016099036-appb-000066
Figure PCTCN2016099036-appb-000067
其中i=1,…,Nk-1,j=1,…,M1,k
第四子模块,用于设
Figure PCTCN2016099036-appb-000068
其中
Figure PCTCN2016099036-appb-000069
则k时刻目标i的更新边缘分布取为
Figure PCTCN2016099036-appb-000070
相应的存在概率取为
Figure PCTCN2016099036-appb-000071
其中i=1,…,Nk-1,当q=M1,k+1时有
Figure PCTCN2016099036-appb-000072
优选的,所述生成模块具体用于:利用k时刻的其它测量
Figure PCTCN2016099036-appb-000073
k-1时刻的其它测量
Figure PCTCN2016099036-appb-000074
和k-2时刻的其它测量
Figure PCTCN2016099036-appb-000075
产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
其中,所述生成模块包括:
第五子模块,用于从
Figure PCTCN2016099036-appb-000076
中取测量
Figure PCTCN2016099036-appb-000077
Figure PCTCN2016099036-appb-000078
中取测量
Figure PCTCN2016099036-appb-000079
Figure PCTCN2016099036-appb-000080
中取测量
Figure PCTCN2016099036-appb-000081
计算得到
Figure PCTCN2016099036-appb-000082
Figure PCTCN2016099036-appb-000083
其中e=1,…,M2,k-2,f=1,…,M2,k-1,g=1,…,M2,k,||·||2表示向量的2范数,|·|表示取绝对值,(·,·)表示两向量的内积;
第六子模块,用于判断条件vmin≤vf,e≤vmax、vmin≤vg,f≤vmax、ag,f,e≤amax和cg,f,e≥cmin是否满足,其中vmin、vmax、amax和cmin为4个给定的参数,分别表示最小速度、最大速度、最大加速度和夹角余弦的最小值;若4个条件同时满足,利用测量
Figure PCTCN2016099036-appb-000084
和测量
Figure PCTCN2016099036-appb-000085
由最小二乘法得到一个新目标的在k时刻的状态均值
Figure PCTCN2016099036-appb-000086
协方差
Figure PCTCN2016099036-appb-000087
和边缘分布
Figure PCTCN2016099036-appb-000088
其中
Figure PCTCN2016099036-appb-000089
Figure PCTCN2016099036-appb-000090
Figure PCTCN2016099036-appb-000091
σw为测量噪声的标准差;同时,指定新目标的存在概率取为
Figure PCTCN2016099036-appb-000092
新目标在k-1时刻的 状态估计为
Figure PCTCN2016099036-appb-000093
其中
Figure PCTCN2016099036-appb-000094
新目标在k-2时刻的状态估计为
Figure PCTCN2016099036-appb-000095
其中
Figure PCTCN2016099036-appb-000096
本发明提供的技术方案,通过预测、分类、更新、裁减与提取、生成、补充、合并这些步骤,利用最小二乘法估计新目标在其出现后的最初3个时间步的状态估计,有效地解决了现有方法在新目标出现后的前几个时间步不能提供新目标状态估计的问题,具有处理速度快的特点,且其计算量明显小于现有方法,具有很强的实用性。
附图说明
图1为本发明一实施方式中适用于杂波环境的多目标跟踪方法流程图;
图2为本发明一实施方式中适用于杂波环境的多目标跟踪系统的内部结构示意图;
图3为本发明一实施方式中利用本发明实施例提供的传感器在50个扫描周期的测量数据图;
图4为本发明一实施方式中根据本发明适用于杂波环境下的多目标跟踪方法与高斯混合概率假设密度滤波方法经过一次实验得到的OSPA距离示意图;
图5为本发明一实施方式中根据本发明用于适用于杂波环境下的多目标跟踪方法与高斯混合概率假设密度滤波方法在经过100次实验得到的平均OSPA距离示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本发明进行进一步详细说明。应当理解,此处所描述的具体实施例仅仅用以解释本发明,并不用于限定本发明。
以下将对本发明所提供的一种适用于杂波环境的多目标跟踪方法进行详细说明。
请参阅图1,为本发明一实施方式中适用于杂波环境的多目标跟踪方法流程图。
在步骤S1中,预测步骤、根据前一时刻各个目标的边缘分布和存在概率,以及当前时刻与前一时刻的时间差,预测前一时刻已存在的各个目标在当前时刻的边缘分布和存在概率。
在本实施方式中,所述预测步骤S1具体包括:
以k-1表示前一时刻,k表示当前时刻,tk-1表示前一时刻的时间,tk表示当前时刻的时间,k-1时刻目标i的边缘分布和存在概率分别表示为N(xi,k-1;mi,k-1,Pi,k-1)和ρi,k-1,其中N表示高斯分布,i=1,2,…Nk-1,xi,k-1为k-1时刻目标i的状态向量,mi,k-1和Pi,k-1分别表示k-1时刻目标i的状态均值和协方差,Nk-1为前一时刻目标的总数;
由k-1时刻目标i的边缘分布N(xi,k-1;mi,k-1,Pi,k-1)和存在概率ρi,k-1,预测k-1时刻的目标i在k时刻的边缘分布和存在概率分别为N(xi,k;mi,k|k-1,Pi,k|k-1)和ρi,k|k-1,其中 mi,k|k-1=Fi,k|k-1mi,k-1,Pi,k|k-1=Qi,k-1+Fi,k|k-1Pi,k-1(Fi,k|k-1)T,ρi,k|k-1=pS,k(Δtki,k-1,Fi,k|k-1为状态转移矩阵,上标T表示矩阵或向量的转置,Δtk=tk-tk-1为k时刻与k-1时刻的时间差,Qi,k-1为k-1时刻目标i的过程噪声协方差矩阵,pS,k(Δtk)为目标的幸存概率,且
Figure PCTCN2016099036-appb-000097
T为采样周期,δ为给定的常数,i=1,2,…Nk-1
在步骤S2中,分类步骤、根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻的测量集,确定出测量集中的每个测量是否源于前一时刻已存在的目标,并分别进行归类。
在本实施方式中,所述分类步骤S2具体包括:根据k-1时刻已存在的目标i在k时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及k时刻的测量集合
Figure PCTCN2016099036-appb-000098
中的第j个测量yj,k,确定测量yj,k是否源于已存在目标,并分别进行归类;
其中,所述确定测量yj,k是否源于已存在目标,并分别进行归类的步骤包括:
子步骤A、求取概率
Figure PCTCN2016099036-appb-000099
其中,Hk为观测矩阵,Rk为观测噪声方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度;
子步骤B、若
Figure PCTCN2016099036-appb-000100
将测量yj,k归入其它测量类;若
Figure PCTCN2016099036-appb-000101
将测量yj,k归入源于已存在目标的测量,在测量集合
Figure PCTCN2016099036-appb-000102
中的每个测量处理后,测量集合中yk的测量被分为两类,一类是源于已存在目标的测量,表示为
Figure PCTCN2016099036-appb-000103
另一类是其它测量,表示为
Figure PCTCN2016099036-appb-000104
其中M1,k和M2,k分别源于已存在目标测量的数目和其它测量的数目,且M1,k+M2,k=Mk
在步骤S3中,更新步骤、根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻源于已存在目标的测量,利用贝叶斯规则确定前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率。
在本实施方式中,所述更新步骤S3具体包括:根据前一时刻已存在的各个目标在当前时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及当前时刻源于已存在目标的测量集合
Figure PCTCN2016099036-appb-000105
利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率;
其中,所述利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率的步骤包括:
子步骤C、利用贝叶斯规则对测量
Figure PCTCN2016099036-appb-000106
处理,得到目标i对应于测量
Figure PCTCN2016099036-appb-000107
的存在概率
Figure PCTCN2016099036-appb-000108
均值向量
Figure PCTCN2016099036-appb-000109
和协方差矩阵
Figure PCTCN2016099036-appb-000110
其中
Figure PCTCN2016099036-appb-000111
在所有的M1,k个测量处理后,各个目标对应于各测量的更新边缘分布和存在概率分别为
Figure PCTCN2016099036-appb-000112
Figure PCTCN2016099036-appb-000113
其中i=1,…,Nk-1,j=1,…,M1,k
子步骤D、设
Figure PCTCN2016099036-appb-000114
其中
Figure PCTCN2016099036-appb-000115
则k时刻目标i的更新边缘 分布取为
Figure PCTCN2016099036-appb-000116
相应的存在概率取为
Figure PCTCN2016099036-appb-000117
其中i=1,…,Nk-1,当q=M1,k+1时有
Figure PCTCN2016099036-appb-000118
在步骤S4中,裁减与提取步骤、根据前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率,将存在概率小于第一阈值的目标裁减掉,同时提取存在概率大于第二阈值的目标的边缘分布作为当前时刻的输出。
在步骤S5中,生成步骤、利用当前时刻的其它测量和其前两时刻的其它测量产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布。
在本实施方式中,所述生成步骤S5具体包括:利用k时刻的其它测量
Figure PCTCN2016099036-appb-000119
k-1时刻的其它测量
Figure PCTCN2016099036-appb-000120
和k-2时刻的其它测量
Figure PCTCN2016099036-appb-000121
产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
其中,所述利用k时刻的其它测量
Figure PCTCN2016099036-appb-000122
k-1时刻的其它测量
Figure PCTCN2016099036-appb-000123
和k-2时刻的其它测量
Figure PCTCN2016099036-appb-000124
产生新目标的步骤包括:
子步骤E、从
Figure PCTCN2016099036-appb-000125
中取测量
Figure PCTCN2016099036-appb-000126
Figure PCTCN2016099036-appb-000127
中取测量
Figure PCTCN2016099036-appb-000128
Figure PCTCN2016099036-appb-000129
中取测量
Figure PCTCN2016099036-appb-000130
计算得到
Figure PCTCN2016099036-appb-000131
Figure PCTCN2016099036-appb-000132
Figure PCTCN2016099036-appb-000133
其中e=1,…,M2,k-2,f=1,…,M2,k-1,g=1,…,M2,k,||·||2表示向量的2范数,|·|表示取绝对值,(·,·)表示两向量的内积;
子步骤F、判断条件vmin≤vf,e≤vmax、vmin≤vg,f≤vmax、ag,f,e≤amax和cg,f,e≥cmin是否满足,其中vmin、vmax、amax和cmin为4个给定的参数,分别表示最小速度、最大速度、最大加速度和夹角余弦的最小值;若4个条件同时满足,利用测量
Figure PCTCN2016099036-appb-000134
和测量
Figure PCTCN2016099036-appb-000135
由最小二乘法得到一个新目标的在k时刻的状态均值
Figure PCTCN2016099036-appb-000136
协方差
Figure PCTCN2016099036-appb-000137
和边缘分布
Figure PCTCN2016099036-appb-000138
其中
Figure PCTCN2016099036-appb-000139
Figure PCTCN2016099036-appb-000140
Figure PCTCN2016099036-appb-000141
σw为测量噪声的标准差;同时,指定新目标的存在概率取为
Figure PCTCN2016099036-appb-000142
新目标在k-1时刻的 状态估计为
Figure PCTCN2016099036-appb-000143
其中
Figure PCTCN2016099036-appb-000144
新目标在k-2时刻的状态估计为
Figure PCTCN2016099036-appb-000145
其中
Figure PCTCN2016099036-appb-000146
在步骤S6中,补充步骤、提取新目标在当前时刻的边缘分布对当前时刻的输出进行补充,并提取新目标在前两个时刻的状态估计分别对前两个时刻的输出进行补充。
在步骤S7中,合并步骤、将在所述裁减与提取步骤中裁减后余下目标的边缘分布和存在概率,分别与在所述生成步骤中生成的新目标在当前时刻的边缘分布和存在概率进行合并,形成当前时刻各个目标的边缘分布和存在概率,并作为下一次递归的输入。
本发明提供的一种适用于杂波环境的多目标跟踪方法,通过预测、分类、更新、裁减与提取、生成、补充、合并这些步骤,利用最小二乘法估计新目标在其出现后的最初3个时间步的状态估计,有效地解决了现有方法在新目标出现后的前几个时间步不能提供新目标状态估计的问题,具有处理速度快的特点,且其计算量明显小于现有方法,具有很强的实用性。
请参阅图2,所示为本发明一实施方式中适用于杂波环境的多目标跟踪系统10的结构示意图。
在本实施方式中,适用于杂波环境的多目标跟踪系统10,主要包括预测模块11、分类模块12、更新模块13、裁减与提取模块14、生成模块15、补充模块16以及合并模块17。
预测模块11,用于根据前一时刻各个目标的边缘分布和存在概率,以及当前时刻与前一时刻的时间差,预测前一时刻已存在的各个目标在当前时刻的边缘分布和存在概率。
在本实施方式中,所述预测模块11具体用于:
以k-1表示前一时刻,k表示当前时刻,tk-1表示前一时刻的时间,tk表示当前时刻的时间,k-1时刻目标i的边缘分布和存在概率分别表示为N(xi,k-1;mi,k-1,Pi,k-1)和ρi,k-1,其中N表示高斯分布,i=1,2,…Nk-1,xi,k-1为k-1时刻目标i的状态向量,mi,k-1和Pi,k-1分别表示k-1时刻目标i的状态均值和协方差,Nk-1为前一时刻目标的总数;
由k-1时刻目标i的边缘分布N(xi,k-1;mi,k-1,Pi,k-1)和存在概率ρi,k-1,预测k-1时刻的目标i在k时刻的边缘分布和存在概率分别为N(xi,k;mi,k|k-1,Pi,k|k-1)和ρi,k|k-1,其中mi,k|k-1=Fi,k|k-1mi,k-1,Pi,k|k-1=Qi,k-1+Fi,k|k-1Pi,k-1(Fi,k|k-1)T,ρi,k|k-1=pS,k(Δtki,k-1,Fi,k|k-1为状态转移矩阵,上标T表示矩阵或向量的转置,Δtk=tk-tk-1为k时刻与k-1时刻的时间差,Qi,k-1为k-1时刻目标i的过程噪声协方差矩阵,pS,k(Δtk)为目标的幸存概率,且
Figure PCTCN2016099036-appb-000147
T为采样周期,δ为给定的常数,i=1,2,…Nk-1
分类模块12,用于根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻的测量集,确定出测量集中的每个测量是否源于前一时刻已存在的 目标,并分别进行归类。
在本实施方式中,所述分类模块12具体用于:根据k-1时刻已存在的目标i在k时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及k时刻的测量集合
Figure PCTCN2016099036-appb-000148
中的第j个测量yj,k,确定测量yj,k是否源于已存在目标,并分别进行归类;
其中,所述分类模块12包括:第一子模块和第二子模块。
第一子模块,用于求取概率
Figure PCTCN2016099036-appb-000149
其中,Hk为观测矩阵,Rk为观测噪声方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度。
第二子模块,用于若
Figure PCTCN2016099036-appb-000150
将测量yj,k归入其它测量类;若
Figure PCTCN2016099036-appb-000151
将测量yj,k归入源于已存在目标的测量,在测量集合
Figure PCTCN2016099036-appb-000152
中的每个测量处理后,测量集合中yk的测量被分为两类,一类是源于已存在目标的测量,表示为
Figure PCTCN2016099036-appb-000153
另一类是其它测量,表示为
Figure PCTCN2016099036-appb-000154
其中M1,k和M2,k分别源于已存在目标测量的数目和其它测量的数目,且M1,k+M2,k=Mk
更新模块13,用于根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻源于已存在目标的测量,利用贝叶斯规则确定前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率。
在本实施方式中,所述更新模块13具体用于:根据前一时刻已存在的各个目标在当前时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及当前时刻源于已存在目标的测量集合
Figure PCTCN2016099036-appb-000155
利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率;
其中,所述更新模块13包括:第三子模块和第四子模块。
第三子模块,用于利用贝叶斯规则对测量
Figure PCTCN2016099036-appb-000156
处理,得到目标i对应于测量
Figure PCTCN2016099036-appb-000157
的存在概率
Figure PCTCN2016099036-appb-000158
均值向量
Figure PCTCN2016099036-appb-000159
和协方差矩阵
Figure PCTCN2016099036-appb-000160
其中
Figure PCTCN2016099036-appb-000161
在所有的M1,k个测量处理后,各个目标对应于各测量的更新边缘分布和存在概率分别为
Figure PCTCN2016099036-appb-000162
Figure PCTCN2016099036-appb-000163
其中i=1,…,Nk-1,j=1,…,M1,k
第四子模块,用于设
Figure PCTCN2016099036-appb-000164
其中
Figure PCTCN2016099036-appb-000165
则k时刻目标i的更新边缘分布取为
Figure PCTCN2016099036-appb-000166
相应的存在概率取为
Figure PCTCN2016099036-appb-000167
其中i=1,…,Nk-1,当q=M1,k+1时有
Figure PCTCN2016099036-appb-000168
裁减与提取模块14,用于根据前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率,将存在概率小于第一阈值的目标裁减掉,同时提取存在概率大于第二阈值的目标的边缘分布作为当前时刻的输出。
生成模块15,用于利用当前时刻的其它测量和其前两时刻的其它测量产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布。
在本实施方式中,所述生成模块15具体用于:利用k时刻的其它测量
Figure PCTCN2016099036-appb-000169
k-1时刻的其它测量
Figure PCTCN2016099036-appb-000170
和k-2时刻的其它测量
Figure PCTCN2016099036-appb-000171
产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
其中,所述生成模块15包括:第五子模块和第六子模块。
第五子模块,用于从
Figure PCTCN2016099036-appb-000172
中取测量
Figure PCTCN2016099036-appb-000173
Figure PCTCN2016099036-appb-000174
中取测量
Figure PCTCN2016099036-appb-000175
Figure PCTCN2016099036-appb-000176
中取测量
Figure PCTCN2016099036-appb-000177
计算得到
Figure PCTCN2016099036-appb-000178
Figure PCTCN2016099036-appb-000179
其中e=1,…,M2,k-2,f=1,…,M2,k-1,g=1,…,M2,k,||·||2表示向量的2范数,|·|表示取绝对值,(·,·)表示两向量的内积。
第六子模块,用于判断条件vmin≤vf,e≤vmax、vmin≤vg,f≤vmax、ag,f,e≤amax和cg,f,e≥cmin是否满足,其中vmin、vmax、amax和cmin为4个给定的参数,分别表示最小速度、最大速度、最大加速度和夹角余弦的最小值;若4个条件同时满足,利用测量
Figure PCTCN2016099036-appb-000180
和测量
Figure PCTCN2016099036-appb-000181
由最小二乘法得到一个新目标的在k时刻的状态均值
Figure PCTCN2016099036-appb-000182
协方差
Figure PCTCN2016099036-appb-000183
和边缘分布
Figure PCTCN2016099036-appb-000184
其中
Figure PCTCN2016099036-appb-000185
Figure PCTCN2016099036-appb-000186
Figure PCTCN2016099036-appb-000187
σw为测量噪声的标准差;同时,指定新目标的存在概率取为
Figure PCTCN2016099036-appb-000188
新目标在k-1时刻的状态估计为
Figure PCTCN2016099036-appb-000189
其中
Figure PCTCN2016099036-appb-000190
新目标在k-2时刻的状态估计为
Figure PCTCN2016099036-appb-000191
其中
Figure PCTCN2016099036-appb-000192
补充模块16,用于提取新目标在当前时刻的边缘分布对当前时刻的输出进行补充,并提取新目标在前两个时刻的状态估计分别对前两个时刻的输出进行补充。
合并模块17,用于将在所述裁减与提取步骤中裁减后余下目标的边缘分布和存在概率, 分别与在所述生成步骤中生成的新目标在当前时刻的边缘分布和存在概率进行合并,形成当前时刻各个目标的边缘分布和存在概率,并作为下一次递归的输入。
本发明提供的一种适用于杂波环境的多目标跟踪系统10,通过预测模块11、分类模块12、更新模块13、裁减与提取模块14、生成模块15、补充模块16以及合并模块17这些模块,利用最小二乘法估计新目标在其出现后的最初3个时间步的状态估计,有效地解决了现有方法在新目标出现后的前几个时间步不能提供新目标状态估计的问题,具有处理速度快的特点,且其计算量明显小于现有方法,具有很强的实用性。
以下通过将本发明的适用于杂波环境的多目标跟踪系统10与现有的高斯混合概率假设密度滤波器进行对比来说明本发明的有益效果。
作为本发明的一个实例,考虑二维空间[-1000m,1000m]×[-1000m,1000m]中运动的目标,目标的状态由位置和速度构成,表示为
Figure PCTCN2016099036-appb-000193
其中x和y分别表示位置分量,
Figure PCTCN2016099036-appb-000194
Figure PCTCN2016099036-appb-000195
分别表示速度分量,上标T表示向量的转置;过程噪声方差矩阵为
Figure PCTCN2016099036-appb-000196
其中,Δtk=tk-tk-1为当前时刻与前一时刻的时间差,σv为过程噪声标准差;观测噪声方差矩阵
Figure PCTCN2016099036-appb-000197
σw为观测噪声的标准差;参数δ取为δ=2.5,最小速度vmin、最大速度vmax、最大加速度amax和夹角余弦的最小值cmin分别取为vmin=30ms-1、vmax=80ms-1、amax=10ms-2和cmin=0.94。
为了产生仿真数据,取幸存概率pS,k=1.0、检测概率pD,k=0.95、杂波密度λc,k=2.5×10-6m-2、过程噪声的标准差σv,=1ms-2、观测噪声的标准差σw=2m和传感器的扫描周期T=1s。一次实验中传感器在50个扫描周期的仿真观测数据如图3所示。
为了处理仿真数据,将本发明与高斯混合概率假设密度滤波器的相关参数设置为pS,k=1.0、pD,k=0.95、λc,k=2.5×10-6m-2、σw=2m、σv=1ms-2、第一阈值为10-3、第二阈值为0.5、高斯混合概率假设密度滤波器新目标的权重为wγ=0.1,新目标的协方差为
Figure PCTCN2016099036-appb-000198
图4为用现有的高斯混合概率假设密度滤波器与本发明对图3中的数据处理得到的最优亚模式分配(Optimal Subpattern Assignment,OSPA)距离。图5为用现有的高斯混合概率假设密度滤波器与本发明分别进行50次Monte Carlo实验得到的平均OSPA距离。
现有的高斯混合概率假设密度滤波器与本发明的实验结果比较表明,本发明的方法可以获得更为精确和可靠的目标状态估计、其OSPA距离比现有的这种方法得到的OSPA距离要小,特别在多目标出现的最初时刻(t=1s至t=16s),OSPA距离减小更为明显。
表1
Figure PCTCN2016099036-appb-000199
表1显示了现有的高斯混合概率假设密度滤波器与本发明在50次实验中得到的一次实验的平均执行时间,结果表明本发明的平均执行时间明显小于现有的高斯混合概率假设密度滤波器。
本发明提供的技术方案,通过预测、分类、更新、裁减与提取、生成、补充、合并这些步骤,利用最小二乘法估计新目标在其出现后的最初3个时间步的状态估计,有效地解决了现有方法在新目标出现后的前几个时间步不能提供新目标状态估计的问题,具有处理速度快的特点,且其计算量明显小于现有方法,具有很强的实用性。
值得注意的是,上述实施例中,所包括的各个单元只是按照功能逻辑进行划分的,但并不局限于上述的划分,只要能够实现相应的功能即可;另外,各功能单元的具体名称也只是为了便于相互区分,并不用于限制本发明的保护范围。
另外,本领域普通技术人员可以理解实现上述各实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成,相应的程序可以存储于一计算机可读取存储介质中,所述的存储介质,如ROM/RAM、磁盘或光盘等。
以上所述仅为本发明的较佳实施例而已,并不用以限制本发明,凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等,均应包含在本发明的保护范围之内。

Claims (8)

  1. 一种适用于杂波环境的多目标跟踪方法,其特征在于,所述方法包括:
    预测步骤、根据前一时刻各个目标的边缘分布和存在概率,以及当前时刻与前一时刻的时间差,预测前一时刻已存在的各个目标在当前时刻的边缘分布和存在概率;
    其中,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,k-1时刻目标i的边缘分布和存在概率分别表示为N(xi,k-1;mi,k-1,Pi,k-1)和ρi,k-1,其中N表示高斯分布,i=1,2,…Nk-1,xi,k-1为k-1时刻目标i的状态向量,mi,k-1和Pi,k-1分别表示k-1时刻目标i的状态均值和协方差,Nk-1为前一时刻目标的总数;
    由k-1时刻目标i的边缘分布N(xi,k-1;mi,k-1,Pi,k-1)和存在概率ρi,k-1,预测k-1时刻的目标i在k时刻的边缘分布和存在概率分别为N(xi,k;mi,k|k-1,Pi,k|k-1)和ρi,k|k-1,其中mi,k|k-1=Fi,k|k-1mi,k-1,Pi,k|k-1=Qi,k-1+Fi,k|k-1Pi,k-1(Fi,k|k-1)T,ρi,k|k-1=pS,k(Δtki,k-1,Fi,k|k-1为状态转移矩阵,上标T表示矩阵或向量的转置,Δtk=tk-tk-1为k时刻与k-1时刻的时间差,Qi,k-1为k-1时刻目标i的过程噪声协方差矩阵,pS,k(Δtk)为目标的幸存概率,且
    Figure PCTCN2016099036-appb-100001
    T为采样周期,δ为给定的常数,i=1,2,…Nk-1
    分类步骤、根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻的测量集,确定出测量集中的每个测量是否源于前一时刻已存在的目标,并分别进行归类;
    更新步骤、根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻源于已存在目标的测量,利用贝叶斯规则确定前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率;
    裁减与提取步骤、根据前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率,将存在概率小于第一阈值的目标裁减掉,同时提取存在概率大于第二阈值的目标的边缘分布作为当前时刻的输出;
    生成步骤、利用当前时刻的其它测量和其前两时刻的其它测量产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
    补充步骤、提取新目标在当前时刻的边缘分布对当前时刻的输出进行补充,并提取新目标在前两个时刻的状态估计分别对前两个时刻的输出进行补充;
    合并步骤、将在所述裁减与提取步骤中裁减后余下目标的边缘分布和存在概率,分别与在所述生成步骤中生成的新目标在当前时刻的边缘分布和存在概率进行合并,形成当前时刻各个目标的边缘分布和存在概率,并作为下一次递归的输入。
  2. 如权利要求1所述的适用于杂波环境的多目标跟踪方法,其特征在于,所述分类步骤具体包括:根据k-1时刻已存在的目标i在k时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及k时刻的测量集合
    Figure PCTCN2016099036-appb-100002
    中的第j个测量yj,k,确定测量yj,k是否源于已存在目标,并分别进行归类;
    其中,所述确定测量yj,k是否源于已存在目标,并分别进行归类的步骤包括:
    子步骤A、求取概率
    Figure PCTCN2016099036-appb-100003
    其中,Hk为观测矩阵,Rk为观测噪声方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度;
    子步骤B、若
    Figure PCTCN2016099036-appb-100004
    将测量yj,k归入其它测量类;若
    Figure PCTCN2016099036-appb-100005
    将测量yj,k归入源于 已存在目标的测量,在测量集合
    Figure PCTCN2016099036-appb-100006
    中的每个测量处理后,测量集合中yk的测量被分为两类,一类是源于已存在目标的测量,表示为
    Figure PCTCN2016099036-appb-100007
    另一类是其它测量,表示为
    Figure PCTCN2016099036-appb-100008
    其中M1,k和M2,k分别源于已存在目标测量的数目和其它测量的数目,且M1,k+M2,k=Mk
  3. 如权利要求2所述的适用于杂波环境的多目标跟踪方法,其特征在于,所述更新步骤具体包括:根据前一时刻已存在的各个目标在当前时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及当前时刻源于已存在目标的测量集合
    Figure PCTCN2016099036-appb-100009
    利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率;
    其中,所述利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率的步骤包括:
    子步骤C、利用贝叶斯规则对测量
    Figure PCTCN2016099036-appb-100010
    处理,得到目标i对应于测量
    Figure PCTCN2016099036-appb-100011
    的存在概率
    Figure PCTCN2016099036-appb-100012
    均值向量
    Figure PCTCN2016099036-appb-100013
    和协方差矩阵
    Figure PCTCN2016099036-appb-100014
    其中
    Figure PCTCN2016099036-appb-100015
    在所有的M1,k个测量处理后,各个目标对应于各测量的更新边缘分布和存在概率分别为
    Figure PCTCN2016099036-appb-100016
    Figure PCTCN2016099036-appb-100017
    其中i=1,…,Nk-1,j=1,…,M1,k
    子步骤D、设
    Figure PCTCN2016099036-appb-100018
    其中
    Figure PCTCN2016099036-appb-100019
    则k时刻目标i的更新边缘分布取为
    Figure PCTCN2016099036-appb-100020
    相应的存在概率取为
    Figure PCTCN2016099036-appb-100021
    其中i=1,…,Nk-1,当q=M1,k+1时确
    Figure PCTCN2016099036-appb-100022
  4. 如权利要求3所述的适用于杂波环境的多目标跟踪方法,其特征在于,所述生成步骤具体包括:利用k时刻的其它测量
    Figure PCTCN2016099036-appb-100023
    k-1时刻的其它测量
    Figure PCTCN2016099036-appb-100024
    和k-2时刻的其它测量
    Figure PCTCN2016099036-appb-100025
    产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
    其中,所述利用k时刻的其它测量
    Figure PCTCN2016099036-appb-100026
    k-1时刻的其它测量
    Figure PCTCN2016099036-appb-100027
    和k-2时刻的其它测量
    Figure PCTCN2016099036-appb-100028
    产生新目标的步骤包括:
    子步骤E、从
    Figure PCTCN2016099036-appb-100029
    中取测量
    Figure PCTCN2016099036-appb-100030
    Figure PCTCN2016099036-appb-100031
    中取测量
    Figure PCTCN2016099036-appb-100032
    Figure PCTCN2016099036-appb-100033
    中取测量
    Figure PCTCN2016099036-appb-100034
    计算得到
    Figure PCTCN2016099036-appb-100035
    Figure PCTCN2016099036-appb-100036
    其中e=1,…,M2,k-2,f=1,…,M2,k-1,g=1,…,M2,k,||·||2表示向量的2范数,|·|表示取绝对值,(·,·)表示两向量的内积;
    子步骤F、判断条件vmin≤vf,e≤vmax、vmin≤vg,f≤vmax、ag,f,e≤amax和cg,f,e≥cmin是否满足,其中vmin、vmax、amax和cmin为4个给定的参数,分别表示最小速度、最大速度、最大加 速度和夹角余弦的最小值;若4个条件同时满足,利用测量
    Figure PCTCN2016099036-appb-100037
    和测量
    Figure PCTCN2016099036-appb-100038
    由最小二乘法得到一个新目标的在k时刻的状态均值
    Figure PCTCN2016099036-appb-100039
    协方差
    Figure PCTCN2016099036-appb-100040
    和边缘分布
    Figure PCTCN2016099036-appb-100041
    其中
    Figure PCTCN2016099036-appb-100042
    Figure PCTCN2016099036-appb-100043
    σw为测量噪声的标准差;同时,指定新目标的存在概率取为
    Figure PCTCN2016099036-appb-100044
    新目标在k-1时刻的状态估计为
    Figure PCTCN2016099036-appb-100045
    其中
    Figure PCTCN2016099036-appb-100046
    新目标在k-2时刻的状态估计为
    Figure PCTCN2016099036-appb-100047
    其中
    Figure PCTCN2016099036-appb-100048
  5. 一种适用于杂波环境的多目标跟踪系统,其特征在于,所述系统包括:
    预测模块,用于根据前一时刻各个目标的边缘分布和存在概率,以及当前时刻与前一时刻的时间差,预测前一时刻已存在的各个目标在当前时刻的边缘分布和存在概率;
    其中,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,k-1时刻目标i的边缘分布和存在概率分别表示为N(xi,k-1;mi,k-1,Pi,k-1)和ρi,k-1,其中N表示高斯分布,i=1,2,…Nk-1,xi,k-1为k-1时刻目标i的状态向量,mi,k-1和Pi,k-1分别表示k-1时刻目标i的状态均值和协方差,Nk-1为前一时刻目标的总数;
    由k-1时刻目标i的边缘分布N(xi,k-1;mi,k-1,Pi,k-1)和存在概率ρi,k-1,预测k-1时刻的目标i在k时刻的边缘分布和存在概率分别为N(xi,k;mi,k|k-1,Pi,k|k-1)和ρi,k|k-1,其中mi,k|k-1=Fi,k|k-1mi,k-1,Pi,k|k-1=Qi,k-1+Fi,k|k-1Pi,k-1(Fi,k|k-1)T,ρi,k|k-1=pS,k(Δtki,k-1,Fi,k|k-1为状态转移矩阵,上标T表示矩阵或向量的转置,Δtk=tk-tk-1为k时刻与k-1时刻的时间差,Qi,k-1为k-1时刻目标i的过程噪声协方差矩阵,pS,k(Δtk)为目标的幸存概率,且
    Figure PCTCN2016099036-appb-100049
    T为采样周期,δ为给定的常数,i=1,2,…Nk-1
    分类模块,用于根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻的测量集,确定出测量集中的每个测量是否源于前一时刻已存在的目标,并分别进行归类;
    更新模块,用于根据前一时刻已存在的各个目标在当前时刻的预测边缘分布和预测存在概率,以及当前时刻源于已存在目标的测量,利用贝叶斯规则确定前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率;
    裁减与提取模块,用于根据前一时刻已存在的各个目标在当前时刻的更新边缘分布和更新存在概率,将存在概率小于第一阈值的目标裁减掉,同时提取存在概率大于第二阈值的目标的边缘分布作为当前时刻的输出;
    生成模块,用于利用当前时刻的其它测量和其前两时刻的其它测量产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
    补充模块,用于提取新目标在当前时刻的边缘分布对当前时刻的输出进行补充,并提取新目标在前两个时刻的状态估计分别对前两个时刻的输出进行补充;
    合并模块,用于将在所述裁减与提取步骤中裁减后余下目标的边缘分布和存在概率,分别与在所述生成步骤中生成的新目标在当前时刻的边缘分布和存在概率进行合并,形成当前时刻各个目标的边缘分布和存在概率,并作为下一次递归的输入。
  6. 如权利要求5所述的适用于杂波环境的多目标跟踪系统,其特征在于,所述分类模块具体用于:根据k-1时刻已存在的目标i在k时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及k时刻的测量集合
    Figure PCTCN2016099036-appb-100050
    中的第j个测量yj,k,确定测量yj,k是否源于已存在目标,并分别进行归类;
    其中,所述分类模块包括:
    第一子模块,用于求取概率
    Figure PCTCN2016099036-appb-100051
    其中,Hk为观测矩阵,Rk为观测噪声方差矩阵,pD,k为目标的检测概率,λc,k为杂波密度;
    第二子模块,用于若
    Figure PCTCN2016099036-appb-100052
    将测量yj,k归入其它测量类;若
    Figure PCTCN2016099036-appb-100053
    将测量yj,k归入源于已存在目标的测量,在测量集合
    Figure PCTCN2016099036-appb-100054
    中的每个测量处理后,测量集合中yk的测量被分为两类,一类是源于已存在目标的测量,表示为
    Figure PCTCN2016099036-appb-100055
    另一类是其它测量,表示为
    Figure PCTCN2016099036-appb-100056
    其中M1,k和M2,k分别源于已存在目标测量的数目和其它测量的数目,且M1,k+M2,k=Mk
  7. 如权利要求6所述的适用于杂波环境的多目标跟踪系统,其特征在于,所述更新模块具体用于:根据前一时刻已存在的各个目标在当前时刻的预测边缘分布N(xi,k;mi,k|k-1,Pi,k|k-1)和预测存在概率ρi,k|k-1,以及当前时刻源于已存在目标的测量集合
    Figure PCTCN2016099036-appb-100057
    利用贝叶斯规则确定当前时刻各个已存在目标的更新边缘分布和存在概率;
    其中,所述更新模块包括:
    第三子模块,用于利用贝叶斯规则对测量
    Figure PCTCN2016099036-appb-100058
    处理,得到目标i对应于测量
    Figure PCTCN2016099036-appb-100059
    的存在概率
    Figure PCTCN2016099036-appb-100060
    均值向量
    Figure PCTCN2016099036-appb-100061
    和协方差矩阵
    Figure PCTCN2016099036-appb-100062
    其中
    Figure PCTCN2016099036-appb-100063
    在所有的M1,k个测量处理后,各个目标对应于各测量的更新边缘分布和存在概率分别为
    Figure PCTCN2016099036-appb-100064
    其中i=1,…,Nk-1,j=1,…,M1,k
    第四子模块,用于设
    Figure PCTCN2016099036-appb-100065
    其中
    Figure PCTCN2016099036-appb-100066
    则k时刻目标i的更新边缘分布取为
    Figure PCTCN2016099036-appb-100067
    相应的存在概率取为
    Figure PCTCN2016099036-appb-100068
    其中i=1,…,Nk-1,当q=M1,k+1时有
    Figure PCTCN2016099036-appb-100069
  8. 如权利要求7所述的适用于杂波环境的多目标跟踪系统,其特征在于,所述生成模块具体用于:利用k时刻的其它测量
    Figure PCTCN2016099036-appb-100070
    k-1时刻的其它测量
    Figure PCTCN2016099036-appb-100071
    和k-2时刻的其它测量
    Figure PCTCN2016099036-appb-100072
    产生新目标,并利用最小二乘法估计新目标在当前时刻的状态均值、协方差和边缘分布;
    其中,所述生成模块包括:
    第五子模块,用于从
    Figure PCTCN2016099036-appb-100073
    中取测量
    Figure PCTCN2016099036-appb-100074
    Figure PCTCN2016099036-appb-100075
    中取测量
    Figure PCTCN2016099036-appb-100076
    Figure PCTCN2016099036-appb-100077
    中取测量
    Figure PCTCN2016099036-appb-100078
    计算得到
    Figure PCTCN2016099036-appb-100079
    Figure PCTCN2016099036-appb-100080
    其中e=1,…,M2,k-2,f=1,…,M2,k-1,g=1,…,M2,k,||·||2表示向量的2范数,|·|表示取绝对值,(·,·)表示两向量的内积;
    第六子模块,用于判断条件vmin≤vf,e≤vmax、vmin≤vg,f≤vmax、ag,f,e≤amax和cg,f,e≥cmin是否满足,其中vmin、vmax、amax和cmin为4个给定的参数,分别表示最小速度、最大速度、最大加速度和夹角余弦的最小值;若4个条件同时满足,利用测量
    Figure PCTCN2016099036-appb-100081
    和测量
    Figure PCTCN2016099036-appb-100082
    由最小二乘法得到一个新目标的在k时刻的状态均值
    Figure PCTCN2016099036-appb-100083
    协方差
    Figure PCTCN2016099036-appb-100084
    和边缘分布
    Figure PCTCN2016099036-appb-100085
    其中
    Figure PCTCN2016099036-appb-100086
    Figure PCTCN2016099036-appb-100087
    σw为测量噪声的标准差;同时,指定新目标的存在概率取为
    Figure PCTCN2016099036-appb-100088
    新目标在k-1时刻的状态估计为
    Figure PCTCN2016099036-appb-100089
    其中
    Figure PCTCN2016099036-appb-100090
    新目标在k-2时刻的 状态估计为
    Figure PCTCN2016099036-appb-100091
    其中
    Figure PCTCN2016099036-appb-100092
PCT/CN2016/099036 2016-09-14 2016-09-14 一种适用于杂波环境的多目标跟踪方法及跟踪系统 WO2018049602A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2016/099036 WO2018049602A1 (zh) 2016-09-14 2016-09-14 一种适用于杂波环境的多目标跟踪方法及跟踪系统
US16/236,603 US10935653B2 (en) 2016-09-14 2018-12-30 Multi-target tracking method and tracking system applicable to clutter environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2016/099036 WO2018049602A1 (zh) 2016-09-14 2016-09-14 一种适用于杂波环境的多目标跟踪方法及跟踪系统

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/236,603 Continuation US10935653B2 (en) 2016-09-14 2018-12-30 Multi-target tracking method and tracking system applicable to clutter environment

Publications (1)

Publication Number Publication Date
WO2018049602A1 true WO2018049602A1 (zh) 2018-03-22

Family

ID=61618571

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/099036 WO2018049602A1 (zh) 2016-09-14 2016-09-14 一种适用于杂波环境的多目标跟踪方法及跟踪系统

Country Status (2)

Country Link
US (1) US10935653B2 (zh)
WO (1) WO2018049602A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109508000A (zh) * 2018-12-16 2019-03-22 西南电子技术研究所(中国电子科技集团公司第十研究所) 异构多传感器多目标跟踪方法
CN109670142A (zh) * 2019-01-07 2019-04-23 杭州电子科技大学 一种基于随机有限集的可分辨机动群目标状态估计方法
CN111340853A (zh) * 2019-12-04 2020-06-26 杭州电子科技大学 基于ospa迭代的多传感器gmphd自适应融合方法
US10935653B2 (en) 2016-09-14 2021-03-02 Shenzhen University Multi-target tracking method and tracking system applicable to clutter environment

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111488552B (zh) * 2020-04-24 2023-03-21 商丘师范学院 基于高斯混合概率假设密度的紧邻多目标跟踪方法
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
CN112711020A (zh) * 2020-12-08 2021-04-27 中国人民解放军海军航空大学 基于循环阈值距离分割的群目标跟踪方法
CN113534131B (zh) * 2021-07-20 2023-08-08 西安邮电大学 基于gm-phd的运动参数估计组合平滑滤波方法

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101975575A (zh) * 2010-10-15 2011-02-16 西安电子科技大学 基于粒子滤波的被动传感器多目标跟踪方法
US20130006576A1 (en) * 2010-03-15 2013-01-03 Bae Systems Plc Target tracking
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
US20140324339A1 (en) * 2013-04-30 2014-10-30 BASELABS GmbH Method and apparatus for the tracking of multiple objects
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
CN104867163A (zh) * 2015-05-28 2015-08-26 深圳大学 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统
CN105719312A (zh) * 2016-01-19 2016-06-29 深圳大学 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN106405538A (zh) * 2016-09-13 2017-02-15 深圳大学 一种适用于杂波环境的多目标跟踪方法及跟踪系统

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8154600B2 (en) * 2007-04-20 2012-04-10 Utc Fire & Security Americas Corporation, Inc. Method and system for distributed multiple target tracking
US8610775B2 (en) * 2008-07-30 2013-12-17 Bae Systems Plc Estimating states of a plurality of targets tracked by a plurality of sensors
US8295548B2 (en) * 2009-06-22 2012-10-23 The Johns Hopkins University Systems and methods for remote tagging and tracking of objects using hyperspectral video sensors
EP2345998B1 (en) * 2009-12-01 2019-11-20 Honda Research Institute Europe GmbH Multi-object tracking with a knowledge-based, autonomous adaptation of the tracking modeling level
US9240053B2 (en) * 2010-03-15 2016-01-19 Bae Systems Plc Target tracking
US8909589B2 (en) * 2012-09-12 2014-12-09 Numerica Corp. Methods and systems for updating a predicted location of an object in a multi-dimensional space
WO2018049602A1 (zh) 2016-09-14 2018-03-22 深圳大学 一种适用于杂波环境的多目标跟踪方法及跟踪系统
US10254758B2 (en) * 2017-01-18 2019-04-09 Ford Global Technologies, Llc Object tracking by unsupervised learning

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130006576A1 (en) * 2010-03-15 2013-01-03 Bae Systems Plc Target tracking
CN101975575A (zh) * 2010-10-15 2011-02-16 西安电子科技大学 基于粒子滤波的被动传感器多目标跟踪方法
US20140324339A1 (en) * 2013-04-30 2014-10-30 BASELABS GmbH Method and apparatus for the tracking of multiple objects
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
CN104867163A (zh) * 2015-05-28 2015-08-26 深圳大学 一种传递边缘分布的测量驱动目标跟踪方法与跟踪系统
CN105719312A (zh) * 2016-01-19 2016-06-29 深圳大学 基于序贯贝叶斯滤波的多目标跟踪方法及跟踪系统
CN106405538A (zh) * 2016-09-13 2017-02-15 深圳大学 一种适用于杂波环境的多目标跟踪方法及跟踪系统

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10935653B2 (en) 2016-09-14 2021-03-02 Shenzhen University Multi-target tracking method and tracking system applicable to clutter environment
CN109508000A (zh) * 2018-12-16 2019-03-22 西南电子技术研究所(中国电子科技集团公司第十研究所) 异构多传感器多目标跟踪方法
CN109508000B (zh) * 2018-12-16 2021-09-03 西南电子技术研究所(中国电子科技集团公司第十研究所) 异构多传感器多目标跟踪方法
CN109670142A (zh) * 2019-01-07 2019-04-23 杭州电子科技大学 一种基于随机有限集的可分辨机动群目标状态估计方法
CN109670142B (zh) * 2019-01-07 2023-05-02 杭州电子科技大学 一种基于随机有限集的可分辨机动群目标状态估计方法
CN111340853A (zh) * 2019-12-04 2020-06-26 杭州电子科技大学 基于ospa迭代的多传感器gmphd自适应融合方法
CN111340853B (zh) * 2019-12-04 2023-06-02 杭州电子科技大学 基于ospa迭代的多传感器gmphd自适应融合方法

Also Published As

Publication number Publication date
US10935653B2 (en) 2021-03-02
US20200025909A1 (en) 2020-01-23

Similar Documents

Publication Publication Date Title
WO2018049602A1 (zh) 一种适用于杂波环境的多目标跟踪方法及跟踪系统
Hue et al. Tracking multiple objects with particle filtering
Clark et al. Multi-target state estimation and track continuity for the particle PHD filter
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
Li et al. Unscented Kalman filter for visual curve tracking
Han et al. Incremental density approximation and kernel-based bayesian filtering for object tracking
Dong et al. Maneuvering multi-target tracking based on variable structure multiple model GMCPHD filter
Vo et al. A random finite set conjugate prior and application to multi-target tracking
Beard et al. A generalised labelled multi-Bernoulli filter for extended multi-target tracking
Liang-Qun et al. Intuitionistic fuzzy joint probabilistic data association filter and its application to multitarget tracking
CN106405538B (zh) 一种适用于杂波环境的多目标跟踪方法及跟踪系统
Yang et al. Adaptive probability hypothesis density filter based on variational Bayesian approximation for multi‐target tracking
Punchihewa et al. A generalized labeled multi-Bernoulli filter for maneuvering targets
WO2018010099A1 (zh) 一种用于跟踪转弯机动目标的方法及其系统
Chen et al. Integrated Bayesian clutter estimation with JIPDA/MHT trackers
Attari et al. An SVSF-based generalized robust strategy for target tracking in clutter
Yao et al. Image moment-based extended object tracking for complex motions
Maharani et al. Particle filter based single shot multibox detector for human moving prediction
Yoon et al. Gaussian mixture importance sampling function for unscented SMC-PHD filter
Chen et al. Multi-target joint detection, tracking and classification with merged measurements using generalized labeled multi-Bernoulli filter
Chen et al. Joint nearest neighbor data association based on interacting multiple model Kalman filtering
Wei et al. An information value function for nonparametric Gaussian processes
Fu et al. Observation bootstrapping CDK-GMPHD filter based on consensus fusion strategy
Zhang et al. Model‐switched Gaussian sum cubature Kalman filter for attitude angle‐aided three‐dimensional target tracking
US9514361B2 (en) Classification of range profiles

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: 16915987

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

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

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 12.07.2019)

122 Ep: pct application non-entry in european phase

Ref document number: 16915987

Country of ref document: EP

Kind code of ref document: A1