CN106772353B - A kind of multi-object tracking method and system suitable for flicker noise - Google Patents
A kind of multi-object tracking method and system suitable for flicker noise Download PDFInfo
- Publication number
- CN106772353B CN106772353B CN201611090305.1A CN201611090305A CN106772353B CN 106772353 B CN106772353 B CN 106772353B CN 201611090305 A CN201611090305 A CN 201611090305A CN 106772353 B CN106772353 B CN 106772353B
- Authority
- CN
- China
- Prior art keywords
- target
- contract
- current time
- distribution
- existing probability
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO 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
- G01S13/00—Systems 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
- G01S13/66—Radar-tracking systems; Analogous systems
Abstract
The present invention is suitable for multi-sensor information fusion field, provides a kind of multi-object tracking method and system suitable for flicker noise, and the multi-object tracking method includes: prediction steps, updates step, generation step and extraction step.Technical solution provided by the invention efficiently solves the tracking problem of nonlinear motion target under flicker noise while guaranteeing data processing real-time.
Description
Technical field
The invention belongs to multi-sensor information fusion technology field more particularly to a kind of multiple targets suitable for flicker noise
Tracking and system.
Background technique
Edge distribution Bayesian filter is the effective ways suitable for multiple target tracking under clutter and noise, but filter
It is only applicable to the case where noise is Gaussian noise, is not suitable for the case where noise is flicker noise.How to solve under flicker noise
Multiple target tracking problem is the key technical problem for needing to explore and solve.
Summary of the invention
Technical problem to be solved by the present invention lies in provide a kind of multi-object tracking method suitable for flicker noise with
Tracking system, it is intended to solve the tracking problem of nonlinear motion target under flicker noise.
The invention is realized in this way a kind of multi-object tracking method suitable for flicker noise, comprising:
According to the distribution of the contract of each target of previous moment and existing probability and the time at current time and previous moment
Difference, the form parameter and scale parameter of gamma distribution are generated using didactic method, and then obtain current time each target
Prediction contract distribution and prediction existing probability;
Existing probability is distributed and predicted according to the prediction contract of the current time each target, utilizes variation Bayes side
The measurement data at method Sequential processing current time obtains the contract of novation distribution of current time each target and updates to exist generally
Rate;
It is distributed using the contract that the measurement data at current time generates newborn target, and existed for the newborn target is specified
The contract distribution of the newborn target and existing probability are distributed and are updated with the contract of novation at the current time respectively by probability
Existing probability merges, and obtains the contract distribution and existing probability of current time each target;
Reduce the target that existing probability is less than first threshold from the current time each target, and will be remaining after reduction
The contract distribution of lower target and existing probability are as filter recursive input next time, from target remaining after the reduction
The target that existing probability is greater than second threshold is extracted, the contract distribution of the target extracted is defeated as the current time
Out, state estimation of the mean value of the contract distribution exported as current target.
The present invention also provides a kind of multiple-target systems suitable for flicker noise, comprising:
Prediction module, for according to the distribution of the contract of each target of previous moment and existing probability and current time and before
The time difference at one moment, the form parameter and scale parameter of gamma distribution are generated using didactic method, and then obtain current
The prediction contract of moment each target is distributed and prediction existing probability;
Update module, for existing probability, benefit to be distributed and predicted according to the prediction contract of the current time each target
With the measurement data at variational Bayesian method Sequential processing current time, the contract of novation distribution of current time each target is obtained
With update existing probability;
Generation module, the contract for generating newborn target for the measurement data using current time are distributed, and are described new
Raw target specifies existing probability, by the contract distribution and the existing probability update with the current time respectively of the newborn target
Contract distribution and update existing probability merge, and obtain the contract distribution and existing probability of current time each target;
Extraction module, the mesh for being less than first threshold for reducing existing probability from the current time each target
Mark, and using the contract distribution of target remaining after reduction and existing probability as filter recursive input next time, from the sanction
Subtract and extract the target that existing probability is greater than second threshold in rear remaining target, the contract distribution of the target extracted is used as institute
The output for stating current time, state estimation of the mean value of the contract distribution exported as current target.
Compared with prior art, the present invention beneficial effect is: the embodiment of the present invention is made an uproar by being distributed modeling flashing with t
Sound obtains closure expression formula using variational Bayesian method, approaches multivariable with the product of each variation per minute edge distribution
Contract distribution efficiently solves so that the contract estimation of multivariable is turned to the iterative estimate to each variation per minute edge distribution
The tracking problem of nonlinear motion target under flicker noise, improves the tracking accuracy of multiple target.
Detailed description of the invention
Fig. 1 is a kind of flow chart of multi-object tracking method suitable for flicker noise provided in an embodiment of the present invention;
Fig. 2 is the measurement data of 70 scan periods of sensor provided in an embodiment of the present invention;
Fig. 3 is the result handled under flicker noise according to multi-object tracking method provided in an embodiment of the present invention;
Fig. 4 is the result handled under flicker noise according to the multi-object tracking method of UK-PHD filter;
Fig. 5 is to provide multi-object tracking method and UK-PHD filtering method according to the embodiment of the present invention to pass through 100 realities
The average OSPA tested is apart from schematic diagram;
Fig. 6 is that the embodiment of the present invention provides a kind of structural schematic diagram of multiple-target system suitable for flicker noise.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
The embodiment of the present invention is distributed according to the contract of each target of previous moment and existing probability prediction current time is each
The contract of target is distributed and existing probability;It is sequential using variational Bayesian method according to the distribution of the contract of prediction and existing probability
Each measurement at processing current time obtains the contract of novation distribution and existing probability of each target;The contract of update is distributed respectively
And existing probability and fresh target contract distribution and existing probability merge, generate the contract distribution of current time each target
And existing probability, so that the tracking that the embodiment of the present invention can be good at solving nonlinear motion target under flicker noise is asked
Topic.
Based on the above principles, the embodiment of the invention provides a kind of multiple targets suitable for flicker noise as shown in Figure 1
Tracking, comprising:
S101, according to the distribution of the contract of each target of previous moment and existing probability and current time and previous moment
Time difference generates the form parameter and scale parameter of gamma distribution using didactic method, and then it is each to obtain current time
The prediction contract of target is distributed and prediction existing probability;
S102 is distributed and is predicted according to the prediction contract of the current time each target existing probability, utilizes variation shellfish
The measurement data at this method Sequential processing current time of leaf, the contract of novation distribution and update for obtaining current time each target are deposited
In probability;
S103 is distributed using the contract that the measurement data at current time generates newborn target, and is referred to for the newborn target
Determine existing probability, the contract distribution of the newborn target and existing probability are distributed with the contract of novation at the current time respectively
And update existing probability and merge, obtain the contract distribution and existing probability of current time each target;
S104 reduces the target that existing probability is less than first threshold, and will cut out from the current time each target
The contract distribution for subtracting rear remaining target and existing probability are remaining after the reduction as filter recursive input next time
The target that existing probability is greater than second threshold is extracted in target, the contract distribution of the target extracted is used as the current time
Output, state estimation of the mean value as current target of the contract distribution exported.
Specifically, in step s101, previous moment is indicated with k-1, k indicates current time, tk-1Indicate previous moment
Time, tkIndicate the time at current time, the measurement noise at the current time obeys the t distribution of ξ dimension, with S (zk;Hkxk,Rk,
rk) indicate the probability density function of current time measurement, wherein HkxkIndicate the mean value of t distribution, RkIndicate concentration matrix, rk
Indicate the freedom degree of t distribution, andThe multivariable contract of previous moment target i is distributed asThe existing probability of target i is ρi,k-1,
Wherein, N indicates Gaussian Profile, and g indicates Gamma distribution, xi,k-1Indicate the state vector of i-th of contract of previous moment distribution,
mi,k-1Indicate the mean value of Gaussian Profile in i-th of contract distribution of previous moment, Pi,k-1Indicate i-th of contract distribution of previous moment
The variance of middle Gaussian Profile,Indicate RkDiagonal entry,And γi,k-1It indicates in i-th of contract distribution of previous moment
The form parameter of Gamma distribution,And ηi,k-1Indicate the scale parameter of Gamma distribution in i-th of contract distribution of previous moment, ξ
For the dimension of state vector, i=1,2 ..., Nk-1, Nk-1For the sum of previous moment target;
According to the contract of each target of previous moment distribution and existing probability, current time and previous moment time difference,
Obtain the prediction contract distribution of current time each targetThe current time prediction of each target is deposited
It is ρ in probabilityi,k|k-1=Ps,k(tk-tk-1)ρi,k-1;Wherein, i=1,2 ..., Nk-1,When being current
The mean value of Gaussian Profile in i-th of contract distribution is carved,
For the variance of Gaussian Profile in i-th of contract distribution of current time, Sigma point xi,0=mi,k-1, The weight of Sigma pointL=1 ..., ξ,For mesh
Target probability of survival,γi,k|k-1=ργγi,k-1For Gamma distribution in i-th of contract distribution of current time
Form parameter,ηi,k|k-1=ρηηi,k-1For the ruler of Gamma distribution in i-th of contract distribution of current time
Parameter is spent, f is nonlinear function, Qk-1For the variance matrix of process noise, the transposition of subscript T representing matrix or vector, T is to adopt
Sample period, δ are known constant, ρα, ρβ, ργ, ρηFor propagation factor, value range be (0,1], rkIt indicates freedom degree, is known
Constant, ξ are the dimension of state vector, and c is a scale parameter.
In step s 102, if the observation that receives of current time integrates as yk=(y1,k,…,yM,k), wherein M is current
Reception to measurement data sum, then it is described according to the prediction contract of the current time each target distribution and prediction deposit
Current time each target is obtained more using the measurement data at variational Bayesian method Sequential processing current time in probability
New contract distribution and update existing probability, comprising:
S1021, the distribution of prediction contract and prediction existing probability using current time each target are each as current time
The initial contract of target is distributed and is initially present probability, i.e., initial contract distribution is taken asProbability is initially present to be taken asWherein i=1,2 ..., Nk-1,
S1022 successively carries out Sequential processing to the 1st to m-th measurement data using variational Bayesian method;
If the contract distribution of each target and existing probability are respectively before j-th of Measurement and Data ProcessingWithWherein, i=1,2 ..., Nk-1,1
≤j≤M;ByWithIt acquires and is measured with j-th
The existing probability of each target is when updateWherein
The contract of each target is distributed as when acquiring with j-th of measurement updaueWherein,Indicate gamma letter
Number, the mark of tr representing matrix,Indicate mean vector,Indicate association
Variance matrix,Indicate filter gain;Wherein Sigma point The form parameter of Gamma distribution isGamma
The scale parameter of distribution is HkFor observing matrix, RkTo see
Survey noise variance matrix, PD,kFor the detection probability of target, λc,kFor clutter density, I indicates unit matrix, yj,kFor current time
J-th of the measurement data received, subscript T are expressed as the transposition of matrix or vector, and ξ is the dimension of state vector;
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
S1023, the contract distribution of each target after m-th Measurement and Data Processing and existing probability are respectivelyWherein i=1,2 ..., Nk-1;M-th is surveyed
The contract distribution of each target and existing probability are divided respectively as the contract of novation of current time each target after amount data processing
Cloth, the contract of novation for thus obtaining current time each target are distributed as And the update existing probability of current time each targetWherein i=1 ..., Nk-1,
In step s 103, if the contract of novation of the current time each target is distributed asThe existing probability of each target is ρi,k;Wherein i=1,
2,...,Nk-1, it is distributed as using the contract that the M measurement data at the current time generates current time new life targetAnd the existing probability of specified current time each newborn target isWherein j=1,2 ..., M,For the covariance of Gaussian Profile in the contract distribution of given j-th newborn target,For
The mean value of Gaussian Profile in the contract distribution of j-th of newborn target,By j-th of measurement data y of the current timej,k=
[xj,k yj,k]TIt generates, and WithFor the conjunction of j-th of the current time newborn target
The form parameter of Gamma distribution in same distribution,WithFor gamma in the contract distribution of j-th of the current time newborn target
The scale parameter of distribution.
The contract of the contract of novation distribution of the current time each target and the current time new life target is distributed
It merges, the contract for obtaining current time each target is distributed as
The existing probability of the existing probability of the current time each target and the current time new life target is carried out
Merge, the existing probability for obtaining the current time each target isWherein Nk=Nk-1+
M。
The present embodiment is further explained below with reference to Fig. 2 to Fig. 5:
In the present embodiment, consider two-dimensional space [- 1000m, 1000m]×Nonlinear motion in [- 1000m, 1000m]
Target.The state of target is made of position, speed and turning rate, is expressed asWherein x and y distinguishes table
Show location components,WithVelocity component is respectively indicated, ω indicates turning rate, and subscript T indicates the transposition of vector, and state shifts square
Battle array beThe variance matrix of process noise isΔtk=tk-tk-1For current time with
The time difference of previous moment, σvAnd σwFor process noise standard deviation;The observational equation of radar are as follows:
Observation noise variance matrixσrAnd σθ
For observation noise standard deviation, noise v is measuredkAssuming that obeying rk=10 t distribution.
In order to generate emulation data, probability of survival p is takenS,k=1.0, detection probability pD,k=0.9, the standard deviation of process noise
σv=1ms-2, σw=0.1rads-2With the standard deviation sigma of observation noiser=2m, σθ=0.0003rad.The simulation observation once tested
Data are as shown in Figure 2.In order to handle emulation data, by the Unscented kalman Gaussian Mixture under the embodiment of the present invention and flicker noise
The relative parameters setting of probability hypothesis density filter is pS,k=1.0, pD,k=0.9, σv=1ms-2、σw=0.1rads-2, σr
=2m, σθ=0.0003rad, first threshold 10-3, second threshold 0.5, propagation factor ρα=ρβ=ργ=ρη=0.75, gal
Ma profile shape parameter initial value Scale Parameter of Gamma Distribution initial valueThe weight w of UK-PHD filter new life target under flicker noiseγ=0.1, the embodiment of the present invention
The existing probability ρ of newborn targetγ=0.1, the covariance of target is newly generatedFig. 3
It is to compare the result that filter and multi-object tracking method provided in an embodiment of the present invention generate with Fig. 4.By the embodiment of the present invention
It is handled with emulation data of the UK-PHD filter under existing flicker noise to Fig. 2,100 Monte Carlo experiments
Obtaining average OSPA (Optimal Subpattern Assignment, optimal Asia mode distribution), distance is as shown in Figure 5.It will show
Some is based on the UK-PHD filter of flicker noise compared with the present invention, and multi-object tracking method of the invention is under flicker noise
The tracking for not knowing, detecting uncertain nonlinear motion target for association can get more accurate and reliable dbjective state
The OSPA that estimation, its OSPA distance are obtained than existing this method is apart from small.
The present invention also provides embodiment as shown in FIG. 6, a kind of multiple-target system, comprising:
Prediction module 601, for according to the distribution of the contract of each target of previous moment and existing probability and current time
With the time difference of previous moment, the form parameter and scale parameter of gamma distribution are generated using didactic method, and then are obtained
The prediction contract of current time each target is distributed and prediction existing probability;
Update module 602, for existing probability to be distributed and predicted according to the prediction contract of the current time each target,
Using the measurement data at variational Bayesian method Sequential processing current time, the contract of novation point of current time each target is obtained
Cloth and update existing probability;
Generation module 603, the contract for generating newborn target for the measurement data using current time are distributed, and are described
Newborn target specifies existing probability, the contract distribution and existing probability by the newborn target respectively with the current time more
New contract distribution and update existing probability merge, and obtain the contract distribution and existing probability of current time each target;
Extraction module 604 is less than first threshold for reducing existing probability from the current time each target
Target, and using the contract distribution of target remaining after reduction and existing probability as filter recursive input next time, from described
The target that existing probability is greater than second threshold is extracted after reduction in remaining target, the contract of the target extracted is distributed conduct
The output at the current time, state estimation of the mean value of the contract distribution exported as current target.
Further, prediction module 601 is specifically used for:
Previous moment is indicated with k-1, and k indicates current time, tk-1Indicate the time of previous moment, tkIndicate current time
Time, the current time measurement noise obey ξ dimension t distribution, with S (zk;Hkxk,Rk,rk) indicate the current time
The probability density function of measurement, wherein HkxkIndicate the mean value of t distribution, RkIndicate concentration matrix, rkIndicate the freedom degree of t distribution,
AndThe multivariable contract of previous moment target i is distributed asThe existing probability of target i is ρi,k-1,
Wherein, N indicates Gaussian Profile, and g indicates Gamma distribution, xi,k-1Indicate the state vector of i-th of contract of previous moment distribution,
mi,k-1Indicate the mean value of Gaussian Profile in i-th of contract distribution of previous moment, Pi,k-1Indicate i-th of contract distribution of previous moment
The variance of middle Gaussian Profile,Indicate RkDiagonal entry,And γi,k-1It indicates in i-th of contract distribution of previous moment
The form parameter of Gamma distribution,And ηi,k-1Indicate the scale parameter of Gamma distribution in i-th of contract distribution of previous moment, ξ
For the dimension of state vector, i=1,2 ..., Nk-1, Nk-1For the sum of previous moment target;
According to the contract of each target of previous moment distribution and existing probability, current time and previous moment time difference,
Obtain the prediction contract distribution of current time each targetThe current time prediction of each target is deposited
It is ρ in probabilityi,k|k-1=Ps,k(tk-tk-1)ρi,k-1;Wherein, i=1,2 ..., Nk-1,When being current
The mean value of Gaussian Profile in i-th of contract distribution is carved,
For the variance of Gaussian Profile in i-th of contract distribution of current time, Sigma point xi,0=mi,k-1, The weight of Sigma pointL=1 ..., ξ,For mesh
Target probability of survival,γi,k|k-1=ργγi,k-1For Gamma distribution in i-th of contract distribution of current time
Form parameter,ηi,k|k-1=ρηηi,k-1For the scale of Gamma distribution in i-th of contract distribution of current time
Parameter, f are nonlinear function, Qk-1For the variance matrix of process noise, the transposition of subscript T representing matrix or vector, T is sampling
Period, δ are known constant, ρα, ρβ, ργ, ρηFor propagation factor, value range be (0,1], rkIt indicates freedom degree, is known normal
Number, ξ are the dimension of state vector, and c is a scale parameter.
Further, update module 602 is specifically used for:
Using the prediction contract distribution of current time each target and predict existing probability as current time each target
Initial contract is distributed and is initially present probability, i.e., initial contract distribution is taken asProbability is initially present to be taken asWherein i=1,2 ..., Nk-1,
Sequential processing is successively carried out to the 1st to m-th measurement data using variational Bayesian method;
If the contract distribution of each target and existing probability are respectively before j-th of Measurement and Data ProcessingWithWherein, i=1,2 ..., Nk-1,1
≤j≤M;ByWithIt acquires and is measured with j-th
The existing probability of each target is when updateWherein
The contract of each target is distributed as when acquiring with j-th of measurement updaueWherein,Indicate gamma letter
Number, the mark of tr representing matrix,Indicate mean vector,Indicate association
Variance matrix,Indicate filter gain;Wherein Sigma point The form parameter of Gamma distribution isGamma
The scale parameter of distribution is HkFor observing matrix, RkTo see
Survey noise variance matrix, PD,kFor the detection probability of target, λc,kFor clutter density, I indicates unit matrix, yj,kFor current time
J-th of the measurement data received, subscript T are expressed as the transposition of matrix or vector, and ξ is the dimension of state vector;
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
The contract of each target after m-th Measurement and Data Processing is distributed and existing probability is respectivelyWithWherein i=1,2 ..., Nk-1;
The contract distribution of each target after m-th Measurement and Data Processing and existing probability is each as current time
The contract of novation of target is distributed, and the contract of novation for thus obtaining current time each target is distributed as And the update existing probability of current time each targetWherein i=1 ..., Nk-1,
Further, generation module 603 is specifically used for:
If the contract of novation of current time each target is distributed asThe existing probability of each target is ρi,k;Wherein i=1,
2,...,Nk-1, it is distributed as using the contract that the M measurement data at the current time generates current time new life targetAnd the existing probability of specified current time each newborn target isWherein j=1,2 ..., M,For the covariance of Gaussian Profile in the contract distribution of given j-th newborn target,For
The mean value of Gaussian Profile in the contract distribution of j-th of newborn target,By j-th of measurement data y of the current timej,k=
[xj,k yj,k]TIt generates, and WithFor the conjunction of j-th of the current time newborn target
The form parameter of Gamma distribution in same distribution,WithFor gamma in the contract distribution of j-th of the current time newborn target
The scale parameter of distribution.
The contract of the contract of novation distribution of the current time each target and the current time new life target is distributed
It merges, the contract for obtaining current time each target is distributed as
The existing probability of the current time each target and the existing probability of the current time new life target are merged, obtained
The existing probability of current time each target isWherein Nk=Nk-1+M。
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (6)
1. a kind of multi-object tracking method suitable for flicker noise characterized by comprising
According to the distribution of the contract of each target of previous moment and existing probability and the time difference at current time and previous moment, adopt
The form parameter and scale parameter of gamma distribution are generated with didactic method, and then obtain the prediction of current time each target
Contract distribution and prediction existing probability;
Existing probability is distributed and predicted according to the prediction contract of the current time each target, utilizes variational Bayesian method sequence
The measurement data for passing through processing current time obtains the contract of novation distribution of current time each target and updates existing probability;
The contract distribution of newborn target is generated using the measurement data at current time, and is existed generally for the newborn target is specified
The contract distribution of the newborn target and existing probability are distributed with the contract of novation at the current time and are updated respectively and deposited by rate
It is merged in probability, obtains the contract distribution and existing probability of current time each target;
Reduce the target that existing probability is less than first threshold from the current time each target, and remaining mesh after reducing
The distribution of target contract and existing probability are extracted from target remaining after the reduction as filter recursive input next time
Existing probability is greater than the target of second threshold, and the contract of the target extracted is distributed the output as the current time, institute
State estimation of the mean value of the contract distribution of output as current target;
Wherein, described according to the distribution of the contract of each target of previous moment and existing probability and current time and previous moment
Time difference generates the form parameter and scale parameter of gamma distribution using didactic method, and then it is each to obtain current time
The prediction contract of target is distributed and prediction existing probability, comprising:
Previous moment is indicated with k-1, and k indicates current time, tk-1Indicate the time of previous moment, tkIndicate current time when
Between, the measurement noise at the current time obeys the t distribution of ξ dimension, with S (zk;Hkxk,Rk,rk) indicate the current time measurement
Probability density function, wherein HkxkIndicate the mean value of t distribution, RkIndicate concentration matrix, rkIndicate the freedom degree of t distribution, andThe multivariable contract of previous moment target i is distributed asThe existing probability of target i is ρi,k-1,
Wherein, N indicates Gaussian Profile, and g indicates Gamma distribution, xi,k-1Indicate the state vector of i-th of contract of previous moment distribution,
mi,k-1Indicate the mean value of Gaussian Profile in i-th of contract distribution of previous moment, Pi,k-1Indicate i-th of contract distribution of previous moment
The variance of middle Gaussian Profile,Indicate RkDiagonal entry,And γi,k-1It indicates in i-th of contract distribution of previous moment
The form parameter of Gamma distribution,And ηi,k-1Indicate the scale parameter of Gamma distribution in i-th of contract distribution of previous moment, ξ
For the dimension of state vector, i=1,2 ..., Nk-1, Nk-1For the sum of previous moment target;
According to the time difference of the contract of each target of previous moment distribution and existing probability, current time and previous moment, obtain
The prediction contract of current time each target is distributedThe current time prediction of each target is deposited
It is ρ in probabilityi,k|k-1=Ps,k(tk-tk-1)ρi,k-1;Wherein, i=1,2 ..., Nk-1,When being current
The mean value of Gaussian Profile in i-th of contract distribution is carved,
For the variance of Gaussian Profile in i-th of contract distribution of current time, Sigma point xi,0=mi,k-1, The weight of Sigma pointL=1 ..., ξ,For mesh
Target probability of survival,γi,k|k-1=ργγi,k-1For Gamma distribution in i-th of contract distribution of current time
Form parameter,ηi,k|k-1=ρηηi,k-1For the scale of Gamma distribution in i-th of contract distribution of current time
Parameter, f are nonlinear function, Qk-1For the variance matrix of process noise, the transposition of subscript T representing matrix or vector, T is sampling
Period, δ are known constant, ρα, ρβ, ργ, ρηFor propagation factor, value range be (0,1], rkIt indicates freedom degree, is known normal
Number, ξ are the dimension of state vector, and c is a scale parameter.
2. multi-object tracking method as described in claim 1, which is characterized in that set the observation that current time receives and integrate as yk
=(y1,k,…,yM,k), wherein M is to receive measurement data sum current time, then described each according to the current time
The prediction contract of target is distributed and prediction existing probability, utilizes the measurement number at variational Bayesian method Sequential processing current time
According to the contract of novation for obtaining current time each target is distributed and updates existing probability, comprising:
Using the prediction contract distribution of current time each target and predict existing probability as the initial of current time each target
Contract is distributed and is initially present probability, i.e., initial contract distribution is taken asProbability is initially present to be taken asWherein i=1,2 ..., Nk-1,
Sequential processing is successively carried out to the 1st to m-th measurement data using variational Bayesian method;
If the contract distribution of each target and existing probability are respectively before j-th of Measurement and Data ProcessingWithWherein, i=1,2 ..., Nk-1,1
≤j≤M;ByWithIt acquires and is measured with j-th
The existing probability of each target is when updateWherein
The contract of each target is distributed as when acquiring with j-th of measurement updaueWherein,Indicate gamma letter
Number, the mark of tr representing matrix,Indicate mean vector,It indicates
Covariance matrix,Indicate filter gain;Wherein Sigma point The form parameter of Gamma distribution isGal
Ma distribution scale parameter be HkFor observing matrix, RkTo see
Survey noise variance matrix, PD,kFor the detection probability of target, λc,kFor clutter density, I indicates unit matrix, yj,kFor current time
J-th of the measurement data received, subscript T are expressed as the transposition of matrix or vector, and ξ is the dimension of state vector;
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
The contract of each target after m-th Measurement and Data Processing is distributed and existing probability is respectivelyWithWherein i=1,2 ..., Nk-1;
Using the contract distribution of each target after m-th Measurement and Data Processing and existing probability as current time each target
Contract of novation distribution, the contract of novation for thus obtaining current time each target is distributed as And the update existing probability of current time each targetWherein i=1 ..., Nk-1,
3. multi-object tracking method as described in claim 1, which is characterized in that set the observation collection that the current time receives
For yk=(y1,k,…,yM,k), wherein M is to receive measurement data sum, the measurement using current time current time
Data generate the contract distribution of newborn target, and specify existing probability for the newborn target, by the contract of the newborn target
Distribution and existing probability are distributed and are updated with the contract of novation at the current time respectively existing probability and merge, and obtain current
The contract distribution of moment each target and existing probability include:
If the contract of novation of current time each target is distributed asThe existing probability of each target is ρi,k;Wherein i=1,
2,...,Nk-1, it is distributed as using the contract that the M measurement data at the current time generates current time new life targetAnd the existing probability of specified current time each newborn target isWherein j=1,2 ..., M,For the covariance of Gaussian Profile in the contract distribution of given j-th newborn target,For
The mean value of Gaussian Profile in the contract distribution of j-th of newborn target,By j-th of measurement data y of the current timej,k=
[xj,k yj,k]TIt generates, and WithFor the conjunction of j-th of the current time newborn target
The form parameter of Gamma distribution in same distribution,WithFor gamma in the contract distribution of j-th of the current time newborn target
The scale parameter of distribution;
The contract of the contract of novation distribution of the current time each target and the current time new life target is distributed and is carried out
Merge, the contract for obtaining current time each target is distributed as
The existing probability of the current time each target and the existing probability of the current time new life target are merged, obtained
The existing probability of current time each target isWherein Nk=Nk-1+M。
4. a kind of multiple-target system suitable for flicker noise characterized by comprising
Prediction module, for according to the distribution of the contract of each target of previous moment and existing probability and current time with it is previous when
The time difference at quarter, the form parameter and scale parameter of gamma distribution are generated using didactic method, and then obtain current time
The prediction contract of each target is distributed and prediction existing probability;
Update module utilizes change for existing probability to be distributed and predicted according to the prediction contract of the current time each target
The measurement data at point bayes method Sequential processing current time obtains the contract of novation distribution and more of current time each target
New existing probability;
Generation module, the contract for generating newborn target for the measurement data using current time are distributed, and are the newborn mesh
Specified existing probability is marked, by the contract distribution of the newborn target and the existing probability contract of novation with the current time respectively
Distribution and update existing probability merge, and obtain the contract distribution and existing probability of current time each target;
Extraction module, the target for being less than first threshold for reducing existing probability from the current time each target, and
It is remaining after the reduction using the contract distribution of target remaining after reduction and existing probability as filter recursive input next time
Under target in extract the target that existing probability is greater than second threshold, the contract distribution of the target extracted is as described current
The output at moment, state estimation of the mean value of the contract distribution exported as current target;
Wherein, the prediction module is specifically used for:
Previous moment is indicated with k-1, and k indicates current time, tk-1Indicate the time of previous moment, tkIndicate current time when
Between, the measurement noise at the current time obeys the t distribution of ξ dimension, with S (zk;Hkxk,Rk,rk) indicate the current time measurement
Probability density function, wherein HkxkIndicate the mean value of t distribution, RkIndicate concentration matrix, rkIndicate the freedom degree of t distribution, andThe multivariable contract of previous moment target i is distributed asThe existing probability of target i is ρi,k-1,
Wherein, N indicates Gaussian Profile, and g indicates Gamma distribution, xi,k-1Indicate the state vector of i-th of contract of previous moment distribution,
mi,k-1Indicate the mean value of Gaussian Profile in i-th of contract distribution of previous moment, Pi,k-1Indicate i-th of contract distribution of previous moment
The variance of middle Gaussian Profile,Indicate RkDiagonal entry,And γi,k-1It indicates in i-th of contract distribution of previous moment
The form parameter of Gamma distribution,And ηi,k-1Indicate the scale parameter of Gamma distribution in i-th of contract distribution of previous moment, ξ
For the dimension of state vector, i=1,2 ..., Nk-1, Nk-1For the sum of previous moment target;
According to the time difference of the contract of each target of previous moment distribution and existing probability, current time and previous moment, obtain
The prediction contract of current time each target is distributedThe current time prediction of each target is deposited
It is ρ in probabilityi,k|k-1=Ps,k(tk-tk-1)ρi,k-1;Wherein, i=1,2 ..., Nk-1,When being current
The mean value of Gaussian Profile in i-th of contract distribution is carved,
For the variance of Gaussian Profile in i-th of contract distribution of current time, Sigma point xi,0=mi,k-1, The weight of Sigma pointL=1 ..., ξ,For mesh
Target probability of survival,γi,k|k-1=ργγi,k-1For Gamma distribution in i-th of contract distribution of current time
Form parameter,ηi,k|k-1=ρηηi,k-1For the scale of Gamma distribution in i-th of contract distribution of current time
Parameter, f are nonlinear function, Qk-1For the variance matrix of process noise, the transposition of subscript T representing matrix or vector, T is sampling
Period, δ are known constant, ρα, ρβ, ργ, ρηFor propagation factor, value range be (0,1], rkIt indicates freedom degree, is known normal
Number, ξ are the dimension of state vector, and c is a scale parameter.
5. multiple-target system as claimed in claim 4, which is characterized in that the update module is specifically used for:
Using the prediction contract distribution of current time each target and predict existing probability as the initial of current time each target
Contract is distributed and is initially present probability, i.e., initial contract distribution is taken asProbability is initially present to be taken asWherein i=1,2 ..., Nk-1,
Sequential processing is successively carried out to the 1st to m-th measurement data using variational Bayesian method;
If the contract distribution of each target and existing probability are respectively before j-th of Measurement and Data ProcessingWithWherein, i=1,2 ..., Nk-1,1
≤j≤M;ByWithIt acquires and is measured with j-th
The existing probability of each target is when updateWherein
The contract of each target is distributed as when acquiring with j-th of measurement updaueWherein,Indicate gamma letter
Number, the mark of tr representing matrix,Indicate mean vector,It indicates
Covariance matrix,Indicate filter gain;Wherein Sigma point The form parameter of Gamma distribution isGamma
The scale parameter of distribution is HkFor observing matrix, RkTo see
Survey noise variance matrix, PD,kFor the detection probability of target, λc,kFor clutter density, I indicates unit matrix, yj,kFor current time
J-th of the measurement data received, subscript T are expressed as the transposition of matrix or vector, and ξ is the dimension of state vector;
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
IfThen the contract of target i is distributed as after j-th of Measurement and Data Processing Its existing probability isWherein
The contract of each target after m-th Measurement and Data Processing is distributed and existing probability is respectivelyWithWherein i=1,2 ..., Nk-1;
Using the contract distribution of each target after m-th Measurement and Data Processing and existing probability as current time each target
Contract of novation distribution, the contract of novation for thus obtaining current time each target is distributed as And the update existing probability of current time each targetWherein i=1 ..., Nk-1,
6. multiple-target system as claimed in claim 4, which is characterized in that the generation module is specifically used for:
If the contract of novation of current time each target is distributed asThe existing probability of each target is ρi,k;Wherein i=1,
2,...,Nk-1, it is distributed as using the contract that the M measurement data at the current time generates current time new life targetAnd the existing probability of specified current time each newborn target isWherein j=1,2 ..., M,For the covariance of Gaussian Profile in the contract distribution of given j-th newborn target,For
The mean value of Gaussian Profile in the contract distribution of j-th of newborn target,By j-th of measurement data y of the current timej,k=
[xj,k yj,k]TIt generates, and WithFor the conjunction of j-th of the current time newborn target
The form parameter of Gamma distribution in same distribution,WithFor gamma in the contract distribution of j-th of the current time newborn target
The scale parameter of distribution;
The contract of the contract of novation distribution of the current time each target and the current time new life target is distributed and is carried out
Merge, the contract for obtaining current time each target is distributed as
The existing probability of the current time each target and the existing probability of the current time new life target are merged, obtained
The existing probability of current time each target isWherein Nk=Nk-1+M。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611090305.1A CN106772353B (en) | 2016-11-29 | 2016-11-29 | A kind of multi-object tracking method and system suitable for flicker noise |
PCT/CN2017/076776 WO2018098926A1 (en) | 2016-11-29 | 2017-03-15 | Multi-target tracking method and system applicable to flicker noise |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611090305.1A CN106772353B (en) | 2016-11-29 | 2016-11-29 | A kind of multi-object tracking method and system suitable for flicker noise |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106772353A CN106772353A (en) | 2017-05-31 |
CN106772353B true CN106772353B (en) | 2019-04-05 |
Family
ID=58915525
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611090305.1A Expired - Fee Related CN106772353B (en) | 2016-11-29 | 2016-11-29 | A kind of multi-object tracking method and system suitable for flicker noise |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN106772353B (en) |
WO (1) | WO2018098926A1 (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107462882B (en) * | 2017-09-08 | 2020-06-02 | 深圳大学 | Multi-maneuvering-target tracking method and system suitable for flicker noise |
CN109474892B (en) * | 2018-11-05 | 2020-07-10 | 浙江工商大学 | Strong robust sensor network target tracking method based on information form |
CN110376582B (en) * | 2019-01-24 | 2022-10-04 | 西安电子科技大学 | Maneuvering target tracking method of self-adaptive GM-PHD |
CN109800721B (en) * | 2019-01-24 | 2020-10-23 | 深圳大学 | Multi-target tracking method and system |
CN110309950B (en) * | 2019-05-16 | 2022-12-30 | 中国人民解放军海军工程大学 | Method and device for optimizing initial storage scheme in modular storage mode |
CN110390684A (en) * | 2019-07-16 | 2019-10-29 | 深圳大学 | Multi-object tracking method and system under a kind of flicker noise |
CN110503071B (en) * | 2019-08-29 | 2021-11-23 | 电子科技大学 | Multi-target tracking method based on variational Bayesian label multi-Bernoulli superposition model |
CN111325776A (en) * | 2020-02-19 | 2020-06-23 | 江苏理工学院 | PHD multi-target tracking method based on variational Bayesian T distribution Kalman filtering |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB201004232D0 (en) * | 2010-03-15 | 2010-04-28 | Bae Systems Plc | Target tracking |
CN103424742B (en) * | 2013-07-24 | 2015-05-13 | 深圳大学 | Target tracking method and target tracking system of sequential processing measurement data |
CN103778320A (en) * | 2013-12-30 | 2014-05-07 | 杭州电子科技大学 | Multi-sensor quantitative fusion target tracking method based on variational Bayesian |
CN103902829B (en) * | 2014-04-11 | 2017-02-15 | 深圳大学 | Target tracking method and system transmitting edge distribution and existence probability |
CN104318059B (en) * | 2014-09-24 | 2018-02-02 | 深圳大学 | Method for tracking target and tracking system for non-linear Gaussian Systems |
CN104867163A (en) * | 2015-05-28 | 2015-08-26 | 深圳大学 | Marginal distribution passing measurement-driven target tracking method and tracking system thereof |
CN105719312B (en) * | 2016-01-19 | 2018-07-27 | 深圳大学 | Multi-object tracking method based on sequential Bayesian filter and tracking system |
-
2016
- 2016-11-29 CN CN201611090305.1A patent/CN106772353B/en not_active Expired - Fee Related
-
2017
- 2017-03-15 WO PCT/CN2017/076776 patent/WO2018098926A1/en active Application Filing
Also Published As
Publication number | Publication date |
---|---|
WO2018098926A1 (en) | 2018-06-07 |
CN106772353A (en) | 2017-05-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106772353B (en) | A kind of multi-object tracking method and system suitable for flicker noise | |
CN105719312B (en) | Multi-object tracking method based on sequential Bayesian filter and tracking system | |
CN101944234B (en) | Multi-object tracking method and device driven by characteristic trace | |
CN104318059B (en) | Method for tracking target and tracking system for non-linear Gaussian Systems | |
CN107462882B (en) | Multi-maneuvering-target tracking method and system suitable for flicker noise | |
Ristic et al. | Calibration of multi-target tracking algorithms using non-cooperative targets | |
WO2017124299A1 (en) | Multi-target tracking method and tracking system based on sequential bayesian filtering | |
CN105761276B (en) | Based on the iteration RANSAC GM-PHD multi-object tracking methods that adaptively newborn target strength is estimated | |
CN106405538A (en) | Multi-target tracking method and tracking system suitable for clutter environment | |
CN110501671A (en) | A kind of method for tracking target and device based on measurement distribution | |
Pang et al. | Anomaly detection based on data stream monitoring and prediction with improved Gaussian process regression algorithm | |
Lee et al. | Estimating bedrock and surface layer boundaries and confidence intervals in ice sheet radar imagery using MCMC | |
Ma et al. | Improved adaptive Kalman filter with unknown process noise covariance | |
CN104867163A (en) | Marginal distribution passing measurement-driven target tracking method and tracking system thereof | |
CN106168943A (en) | A kind of method for following the tracks of turning machine moving-target and system thereof | |
Karakuş et al. | Ship wake detection in X-band SAR images using sparse GMC regularization | |
Chavali et al. | Hierarchical particle filtering for multi-modal data fusion with application to multiple-target tracking | |
CN106054167A (en) | Intensity filter-based multi-extended target tracking method | |
Gehly et al. | An AEGIS-CPHD filter to maintain custody of GEO space objects with limited tracking data | |
Ilisei et al. | A technique for the automatic estimation of ice thickness and bedrock properties from radar sounder data acquired at Antarctica | |
Prokopenko | Statistical synthesis of robust signal detection algorithms under conditions of aprioristic uncertainty | |
Soysal et al. | Data fusion in a multistatic radar network using covariance intersection and particle filtering | |
Rezaie et al. | Shrinked (1− α) ensemble Kalman filter and α Gaussian mixture filter | |
CN104063615A (en) | Target tracking method and tracking system based on variable coefficient alpha-beta filter | |
CN102509288A (en) | SAR (Synthetic Aperture Radar) image change detection method based on Memetic kernel clustering |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190405 Termination date: 20191129 |
|
CF01 | Termination of patent right due to non-payment of annual fee |