CN110648355A - Image tracking method, system and related device - Google Patents

Image tracking method, system and related device Download PDF

Info

Publication number
CN110648355A
CN110648355A CN201910935692.1A CN201910935692A CN110648355A CN 110648355 A CN110648355 A CN 110648355A CN 201910935692 A CN201910935692 A CN 201910935692A CN 110648355 A CN110648355 A CN 110648355A
Authority
CN
China
Prior art keywords
particle
particle set
weight
new
transfer
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.)
Pending
Application number
CN201910935692.1A
Other languages
Chinese (zh)
Inventor
陈文林
曹黎俊
勾文清
李翰
喻芳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhongke Intelligence Technology (hunan) Co Ltd
Original Assignee
Zhongke Intelligence Technology (hunan) Co Ltd
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 Zhongke Intelligence Technology (hunan) Co Ltd filed Critical Zhongke Intelligence Technology (hunan) Co Ltd
Priority to CN201910935692.1A priority Critical patent/CN110648355A/en
Publication of CN110648355A publication Critical patent/CN110648355A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

The application provides an image tracking method, which comprises the following steps: inputting image data at a preset moment; generating a new particle set according to the particle set at the previous moment of the preset moment; calculating the weight corresponding to each particle in the new particle set, taking the state parameter corresponding to the particle with the largest weight as the optimal target state, and determining the corresponding pixel block; determining the best matching manifold according to the weight; updating the mean value and the feature vector of the best matching manifold by using the forgetting factor and the pixel block; and updating the weight of the best matching manifold, and determining the change position of the image data according to the updated best matching manifold. The method and the device improve the particle resampling process and reduce the error of the final estimation state. And a new particle set is generated, the problem of particle degradation is solved, and the tracking performance of the change position in the image data is improved. The present application also provides an image tracking system, a computer-readable storage medium, and an image tracking terminal having the above-mentioned advantageous effects.

Description

Image tracking method, system and related device
Technical Field
The present disclosure relates to the field of image tracking, and in particular, to an image tracking method, system and related apparatus.
Background
The reason that existing SIR particle filters are widely used is that the weight distribution function is independent of the observation and the sampling of particles based on such a distribution function is relatively simple. However, the weight distribution function does not utilize the current observation information, so the sampling efficiency is not high, and in addition, resampling is needed in each iteration, so the particles lose some diversity, and the image tracking performance is low.
Disclosure of Invention
An object of the present application is to provide an image tracking method, system, computer-readable storage medium, and image tracking terminal capable of improving image tracking performance.
In order to solve the technical problem, the present application provides an image tracking method, which has the following specific technical scheme:
inputting image data at a preset moment;
generating a new particle set according to the particle set at the previous moment of the preset moment;
calculating the weight corresponding to each particle in the new particle set, and taking the state parameter corresponding to the particle with the maximum weight as the optimal target state;
determining the best matching manifold according to the weight;
updating the mean and the feature vector of the best matching manifold by using a forgetting factor and the pixel block;
and updating the weight of the best matching manifold by using the updated mean value and the feature vector, and determining the change position of the image data according to the best matching manifold after the weight is updated.
Generating a new particle set according to the particle set at the previous moment of the preset moment comprises:
and generating a new particle set by using an SSIR filter according to the particle set at the previous moment of the preset moment.
Generating a new particle set by using an SSIR filter according to the particle set at the previous moment of the preset moment comprises the following steps:
selecting a preset number of particles with the maximum weight from the particle set at the previous moment of the preset moment;
performing particle resampling on the particles with the maximum weight of the preset number to obtain a resampled particle set;
calculating the target state of the resampling particle set by using a Burg algorithm, and obtaining a substitute particle set;
transferring the resampled particle set by using a first state transfer function to obtain a first transfer particle set, and transferring the substitute particle set by using a second state transfer function to obtain a second transfer particle set;
and combining the first transfer particle set and the second transfer particle set to obtain the new particle set.
Wherein, resampling the particles with the maximum weight of the preset number of particles comprises:
and utilizing an accumulative distribution function to resample the particles with the preset number of the maximum weights.
The present application further provides an image tracking system, comprising:
the image input module is used for inputting image data at a preset moment;
the new particle generation module is used for generating a new particle set according to the particle set at the previous moment of the preset moment;
a weight calculation module, configured to calculate a weight corresponding to each particle in the new particle set, and determine a pixel block corresponding to the optimal target state by using a state parameter corresponding to the particle with the largest weight as the optimal target state;
the matching manifold determining module is used for determining the best matching manifold according to the weight;
the data updating module is used for updating the mean value and the feature vector of the optimal matching manifold by using a forgetting factor and the pixel block;
and the image tracking module is used for updating the weight of the best matching manifold by using the updated mean value and the updated feature vector and determining the change position of the image data according to the best matching manifold after the weight is updated.
The new particle generation module is specifically a module for generating a new particle set by using an SSIR filter according to the particle set at the previous time of the preset time.
Wherein the new particle generation module comprises:
a particle selection unit, configured to select a preset number of particles with the largest weight from a particle set at a time before the preset time;
the resampling unit is used for performing particle resampling on the particles with the maximum weight of the preset number to obtain a resampled particle set;
the state calculation selection unit is used for calculating the target state of the resampling particle set by using a Burg algorithm and obtaining a substitute particle set;
a particle transfer unit, configured to transfer the resampled particle set by using a first state transfer function to obtain a first transfer particle set, and transfer the substitute particle set by using a second state transfer function to obtain a second transfer particle set;
and a new particle generation unit, configured to merge the first transfer particle set and the second transfer particle set to obtain the new particle set.
The resampling unit is specifically a unit for performing particle resampling on the preset number of particles with the maximum weight by using an accumulative distribution function.
The present application also provides a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the image tracking method as described above.
The application also provides an image tracking terminal, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor realizes the steps of the image tracking method when calling the computer program in the memory.
The application provides an image tracking method, which comprises the following steps: inputting image data at a preset moment; generating a new particle set according to the particle set at the previous moment of the preset moment; calculating the weight corresponding to each particle in the new particle set, and taking the state parameter corresponding to the particle with the maximum weight as the optimal target state; determining the best matching manifold according to the weight; updating the mean value and the feature vector of the best matching manifold by using a forgetting factor; and updating the weight of the best matching manifold by using the updated mean value and the feature vector, and determining the change position of the image data according to the best matching manifold after the weight is updated.
According to the method and the device, the time continuity characteristic of the target state is utilized, the particle resampling process is improved, and the error of the final estimation state is reduced. And a new particle set is generated on the basis of the SIR filter, so that the problem of particle degradation is solved, and the tracking performance of a change position in image data is improved. The application also provides an image tracking system, a computer readable storage medium and an image tracking terminal, which have the beneficial effects and are not repeated herein.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart of an image tracking method according to an embodiment of the present application;
fig. 2 is a variation graph of a forgetting factor provided in an embodiment of the present application;
fig. 3 is a schematic structural diagram of an image tracking system according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Referring to fig. 1, fig. 1 is a flowchart of an image tracking method according to an embodiment of the present application, where the tracking method includes:
s101: inputting image data at a preset moment;
the preset time is not limited to this, and may be any time.
S102: generating a new particle set according to the particle set at the previous moment of the preset moment;
and generating a new particle set by using an SSIR filter according to the particle set at the previous moment of the preset moment.
The method comprises the following specific steps:
s1021: selecting a preset number of particles with the maximum weight from the particle set at the previous moment of the preset moment;
s1022: performing particle resampling on the particles with the maximum weight of the preset number to obtain a resampled particle set;
s1023: calculating the target state of the resampling particle set by using a Burg algorithm, and obtaining a substitute particle set;
s1024: transferring the resampled particle set by using a first state transfer function to obtain a first transfer particle set, and transferring the substitute particle set by using a second state transfer function to obtain a second transfer particle set;
s1025: and combining the first transfer particle set and the second transfer particle set to obtain the new particle set.
The following description is made using the associated pseudo code of algorithm 1:
input-set of particles at previous time t-1
Figure BDA0002221517630000051
1 from particlesSelecting N from the set1The particles with the maximum weight;
for this N1Resampling the particles to generate a particle set
Figure BDA0002221517630000052
3 computing X using the Burg algorithmpredAnd producing a collection of particles
Figure BDA0002221517630000053
4: utilization of
Figure BDA0002221517630000054
Pair chit-1,1Transferring to obtain new particle set Lambdat,1；
5 utilization ofPair chit-1,2Transferring to obtain new particle set Lambdat,2；
6:Output:
Figure BDA0002221517630000056
Specifically, the particle resampling may be performed on the preset number of particles with the maximum weight by using an accumulative distribution function, which is shown in algorithm 2 below.
The specific process of the algorithm 2 is as follows:
1、Input：
Figure BDA0002221517630000057
2. initialize CDF (cumulative distribution function): c. C1＝0；
3、for k＝2；k≤Ns；k++do
4. Construction of CDF:
Figure BDA0002221517630000058
5、end for
6. starting from the bottom of the CDF: k is 1
7. Selecting a random point: u. of1～U[0,1/Ns]
8、for j＝1…Ns do
9、uj＝uj-1+(j-1)*1/Ns
10、whileuj＞ckdo
11、k＝k+1；
12、end while
13. Generating a sample:
14. weighting:
Figure BDA0002221517630000062
15. recording the mother sample index: k is a radical ofj＝k
16、end for
17、Output:
Figure BDA0002221517630000063
S103: calculating a weight corresponding to each particle in the new particle set, taking a state parameter corresponding to the particle with the largest weight as an optimal target state, and determining a pixel block corresponding to the optimal target state;
in particular, a formula may be utilized
Figure BDA0002221517630000064
And calculating the weight corresponding to each particle in the new particle set. Wherein the apparent likelihood p (o)t|Xt) Indicating the target state is XtWhen observed as OtThe probability of (c).
Optimal target state
Figure BDA0002221517630000065
The corresponding original image pixel block is marked as
Figure BDA0002221517630000066
S104: determining the best matching manifold according to the weight;
the observations do not typically change abruptly, so it is highly likely that the observations of the target will fall on the same local manifold at both earlier and later times. This is the time continuity of the apparent manifold model. This property can be used to design a better search strategy to find the best matching local apparent manifold, and the specific process and pseudo code are as follows:
1, arranging all local manifolds according to the order of the weight.
2:found＝false；
3:for i＝1；i≤N；i++do
If distance di≤k1σdi then
5:found＝true
6:Mi*＝Mi
7:Break
8:End if
9:End for
10:If found＝＝false then
11 local manifold model MNRemoving from the appearance model;
construction of a New M Using recent Observation dataN；
13:Mi*＝MN；
14:end if
15:Output:Mi*；
S105: updating the mean and the feature vector of the best matching manifold by using a forgetting factor and the pixel block;
after the optimal state of the target is determined, the next step is to update the appearance model of the target to adapt to the change of the appearance data. Although the distribution of the object appearance data is described by using a group of local manifolds, the new observation at each moment is only located in one local manifold in the low-dimensional manifold space, so that the distribution of the object appearance data is used
Figure BDA0002221517630000071
Matching each local manifold and comparing to find the best matched local manifold Mi*. If the best match can be found, then use
Figure BDA0002221517630000072
To Mi*Updates are made while other local manifolds are not changed. A suitable matching criterion is defined:
di≤k1σdi
wherein k is1Is a constant parameter, σdiIs a distance diThe mean square error of (c).
Finding matching local manifold Mi*Then, the weight updating mode of each local manifold is as follows:
ωi,t＝(1-ρω)ωi,t-1+ρωBi,t
where ρ isωIs the learning rate, for matched Mi*Corresponding to Bi,tOther assignments are 0.
Mean square error σdiAlso through a constant rate pdUpdating;
Figure BDA0002221517630000073
wherein d isi*Is an observation
Figure BDA0002221517630000074
With matched local appearance model Mi*The distance between them.
In order to make the constructed appearance model reflect the observation change of the target in the recent time more accurately and rapidly, the author defines a constant forgetting factor f to reduce the proportion of the historical observation in the appearance model. The specific gravities of the mean center and the feature vector observed in the model at each time point exponentially decline, and the number of frames in which the model can actually reflect changes in the apparent data of the target is approximately 1/(1-f). But doing so takes a longer time for the model to converge. A varying learning rate of 1/t can be used, allowing rapid convergence in parameter estimation without requiring many iterations. In order to enable the mean center and feature vectors of the constructed model to quickly converge into a practical range, a varying forgetting factor is also used:
Figure BDA0002221517630000081
wherein f and beta are constant parameters, and f ≦ 1 and beta ≦ 1 are satisfied. Variable ciRepresenting a partial manifold Mi*Number of matches to observations. Fig. 2 is a variation graph of a forgetting factor provided in the embodiment of the present application, and fig. 2 shows a forgetting factor ftThe value of (d) increases as the number of matches increases and approaches a constant f.
S106: and updating the weight of the best matching manifold by using the updated mean value and the feature vector, and determining the change position of the image data according to the best matching manifold after the weight is updated.
For the particle filter, there are two main problems to be solved, one is how to sample the particles, i.e. how to define the weight distribution function, and the other is how to update the weights. Sir (sampling inportand cd filtering) particle filter set weight fraction:
q(Xt|X1:t-1,Ot)＝p(Xt|Xt-1)
and the weight of the particle is obtained by normalizing the apparent likelihood function p (ot | Xt). The reason that SIR particle filters are widely used is that the weight distribution function is independent of the observation and the sampling of particles based on this distribution function is relatively simple. However, the weight distribution function does not utilize the current observation information, so the sampling efficiency is not high, and in addition, resampling is needed in each iteration, so the particles lose some diversity. Resampling here means that the particles with small weight are removed from the particle set, and the particles with large weight are left. The following describes a specific implementation process of the SIR particle filtering algorithm:
input device
Figure BDA0002221517630000082
1:for k＝1；k≤Ns；k++do
2, sampling:
Figure BDA0002221517630000083
and 3, calculating a weight:
Figure BDA0002221517630000084
4:end for
5, normalizing the weight:
resampling using algorithm 2:
Figure BDA0002221517630000086
output of
In some cases, for example, the appearance of the target changes sharply, the appearance model at this time may not update and accurately describe the appearance of the target in time, and the particle weight calculated according to the appearance model may also have a large error, so that the distribution of the particle set may not meet the expected effect, and the newly generated particle set may be distributed at a position far from the true state. It is desirable to reduce the proportion of particles in this portion and to increase the proportion of particles closer to the real state.
For this reason, it is considered to improve the process of particle resampling by utilizing the time-continuous characteristic of the target state, i.e., the target state is continuously changed without abrupt change. At any time t, the current state of the target is estimated by using a regression model and the historical state information of the target, the estimation and the true value are not greatly different at most of the time, and the error of the final estimation state can be reduced by sampling the area around the estimation state.
The target is in a short timeCan be considered to be linear, so a linear autoregressive filter (AR filter) can be used to estimate the motion state at time t. Suppose a given set of historical data zt-n,zt-n+1...zt-1The target state at time t can be estimated by:
Figure BDA0002221517630000091
wherein a isiIs the coefficient, n is the order of the filter, the residual epsilontRepresenting gaussian noise.
There are many ways to calculate linear autoregressive filter coefficients, the Burg algorithm being one of the better ways to solve for filter coefficients by minimizing forward and backward prediction errors while satisfying the Levingson-Durbin recursion condition based on modern power spectrum estimation. For short length of history data, the Burg algorithm can obtain stable AR model and high frequency resolution, and the calculation cost is small. The Burg algorithm is used to calculate the coefficients of the AR model.
Assume a set of particles at time (t-1)
Figure BDA0002221517630000092
Corresponding particle weight
Figure BDA0002221517630000093
By normalizationIs obtained at this point NsSelecting N1 particles with the largest weight from the sample particles, then resampling the particles by using an algorithm 2, and recording a particle set obtained by resampling as
Figure BDA0002221517630000095
And the remaining N2(N2＝Ns-N1) One particle is discarded and another set of N is used2Particle substitution, is described as
Figure BDA0002221517630000096
Where X ispred＝{xpred,ypred,hpred,wpredDenotes the target state at time t estimated using the AR filter. To avoid particle degradation problems, N is usually chosen2Greater than N1E.g. N1＝5·N2。
XpredThe most interesting of the four states of (1) is the horizontal and vertical displacement of the object, so the object box height h at time t is assumedpredAnd width wpredX is calculated using only the AR filter as the value of the previous time (t-1)predAnd ypred：
Figure BDA0002221517630000101
Wherein xt-iAnd yt-iIs the historical displacement data of the target in the horizontal and vertical directions, n is the order of the filter,
selecting n as 2, a in the experimentiAnd biAre the coefficients of the filter, calculated by the Burg algorithm and historical data of the first 5 frames.
New sample particle at time t
Figure BDA0002221517630000102
By means of a state transfer function p (X)t|Xt-1) And (6) sampling to obtain.
Based on the set of particles χt-1,1The new particles obtained by sampling are as follows:
Figure BDA0002221517630000103
wherein N represents a Gaussian distribution, Σ1Is a diagonal covariance matrix that is,
Figure BDA0002221517630000104
are all constant parameters.
Set of new particles is marked as
Figure BDA0002221517630000106
Based on the set of particles χt-1,2The new particles obtained by sampling are as follows:
Figure BDA0002221517630000107
wherein N represents a Gaussian distribution, Σ2Is a diagonal covariance matrix that is,
Figure BDA0002221517630000108
Figure BDA0002221517630000109
is initially equal to
Figure BDA00022215176300001010
Same, but with a constant parameter pσUpdating them online:
σx2,t＝(1-ρσ)σx2,t-1+ρσ|xt-xt-1|
σy2,t＝(1-ρσ)σy2,t-1+ρσ|yt-yt-1|
the motion state of a general object is slowly changing most of the time, so generally σx2And σy2With a smaller value. Based on xt-12The new particle set obtained by sampling is recorded as
Finally all the new particle sets generated are recorded as
Figure BDA00022215176300001012
According to the method and the device, the time continuity characteristic of the target state is utilized, the particle resampling process is improved, and the error of the final estimation state is reduced. And a new particle set is generated on the basis of the SIR filter, so that the problem of particle degradation is solved, and the tracking performance of a change position in image data is improved.
In the following, an image tracking system provided by an embodiment of the present application is described, and the image tracking system described below and the image tracking method described above may be referred to correspondingly.
Referring to fig. 3, fig. 3 is a schematic structural diagram of an image tracking system provided in an embodiment of the present application, where the system includes:
an image input module 100, configured to input image data at a preset time;
a new particle generating module 200, configured to generate a new particle set according to the particle set at the previous time of the preset time;
a weight calculation module 300, configured to calculate a weight corresponding to each particle in the new particle set, and determine a pixel block corresponding to the optimal target state by using a state parameter corresponding to the particle with the largest weight as the optimal target state;
a matching manifold determining module 400, configured to determine an optimal matching manifold according to the weight;
a data updating module 500, configured to update the mean and the feature vector of the best matching manifold by using a forgetting factor and the pixel block;
an image tracking module 600, configured to update the weight of the best matching manifold by using the updated mean and the updated feature vector, and determine a change position of the image data according to the best matching manifold updated by the weight.
As a preferred embodiment, the new particle generating module 200 may specifically be a module configured to generate a new particle set according to a particle set at a time previous to the preset time by using an SSIR filter.
Further, the new particle generation module 200 may include:
a particle selection unit, configured to select a preset number of particles with the largest weight from a particle set at a time before the preset time;
the resampling unit is used for performing particle resampling on the particles with the maximum weight of the preset number to obtain a resampled particle set;
the state calculation selection unit is used for calculating the target state of the resampling particle set by using a Burg algorithm and obtaining a substitute particle set;
a particle transfer unit, configured to transfer the resampled particle set by using a first state transfer function to obtain a first transfer particle set, and transfer the substitute particle set by using a second state transfer function to obtain a second transfer particle set;
and a new particle generation unit, configured to merge the first transfer particle set and the second transfer particle set to obtain the new particle set.
Preferably, the resampling unit is specifically a unit configured to perform particle resampling on the preset number of particles with the maximum weight by using an accumulative distribution function.
The present application also provides a computer readable storage medium having stored thereon a computer program which, when executed, may implement the steps provided by the above-described embodiments. The storage medium may include: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The application also provides an image tracking terminal, which may include a memory and a processor, wherein the memory stores a computer program, and the processor may implement the steps provided by the above embodiments when calling the computer program in the memory. Of course, the image tracking terminal may further include various network interfaces, power supplies and other components.
The embodiments are described in a progressive manner in the specification, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system provided by the embodiment, the description is relatively simple because the system corresponds to the method provided by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present application are explained herein using specific examples, which are provided only to help understand the method and the core idea of the present application. It should be noted that, for those skilled in the art, it is possible to make several improvements and modifications to the present application without departing from the principle of the present application, and such improvements and modifications also fall within the scope of the claims of the present application.
It is further noted that, in the present specification, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

Claims (10)

1. An image tracking method, comprising:
inputting image data at a preset moment;
generating a new particle set according to the particle set at the previous moment of the preset moment;
calculating a weight corresponding to each particle in the new particle set, taking a state parameter corresponding to the particle with the largest weight as an optimal target state, and determining a pixel block corresponding to the optimal target state;
determining the best matching manifold according to the weight;
updating the mean and the feature vector of the best matching manifold by using a forgetting factor and the pixel block;
and updating the weight of the best matching manifold by using the updated mean value and the feature vector, and determining the change position of the image data according to the best matching manifold after the weight is updated.
2. The image tracking method according to claim 1, wherein generating a new particle set from the particle set at the previous time of the preset time comprises:
and generating a new particle set by using an SSIR filter according to the particle set at the previous moment of the preset moment.
3. The image tracking method of claim 2, wherein generating a new set of particles from the set of particles at a time previous to the preset time and using an SSIR filter comprises:
selecting a preset number of particles with the maximum weight from the particle set at the previous moment of the preset moment;
performing particle resampling on the particles with the maximum weight of the preset number to obtain a resampled particle set;
calculating the target state of the resampling particle set by using a Burg algorithm, and obtaining a substitute particle set;
transferring the resampled particle set by using a first state transfer function to obtain a first transfer particle set, and transferring the substitute particle set by using a second state transfer function to obtain a second transfer particle set;
and combining the first transfer particle set and the second transfer particle set to obtain the new particle set.
4. The image tracking method of claim 3, wherein the performing the particle resampling on the preset number of the particles with the largest weight comprises:
and utilizing an accumulative distribution function to resample the particles with the preset number of the maximum weights.
5. An image tracking system, comprising:
the image input module is used for inputting image data at a preset moment;
the new particle generation module is used for generating a new particle set according to the particle set at the previous moment of the preset moment;
a weight calculation module, configured to calculate a weight corresponding to each particle in the new particle set, and determine a pixel block corresponding to the optimal target state by using a state parameter corresponding to the particle with the largest weight as the optimal target state;
the matching manifold determining module is used for determining the best matching manifold according to the weight;
the data updating module is used for updating the mean value and the feature vector of the optimal matching manifold by using a forgetting factor and the pixel block;
and the image tracking module is used for updating the weight of the best matching manifold by using the updated mean value and the updated feature vector and determining the change position of the image data according to the best matching manifold after the weight is updated.
6. The image tracking system of claim 5, wherein the new particle generation module is specifically a module for generating a new particle set according to the particle set at the previous time of the preset time and by using an SSIR filter.
7. The image tracking system of claim 6, wherein the new particle generation module comprises:
a particle selection unit, configured to select a preset number of particles with the largest weight from a particle set at a time before the preset time;
the resampling unit is used for performing particle resampling on the particles with the maximum weight of the preset number to obtain a resampled particle set;
the state calculation selection unit is used for calculating the target state of the resampling particle set by using a Burg algorithm and obtaining a substitute particle set;
a particle transfer unit, configured to transfer the resampled particle set by using a first state transfer function to obtain a first transfer particle set, and transfer the substitute particle set by using a second state transfer function to obtain a second transfer particle set;
and a new particle generation unit, configured to merge the first transfer particle set and the second transfer particle set to obtain the new particle set.
8. The image tracking system of claim 7, wherein the resampling unit is specifically a unit configured to perform particle resampling on the preset number of particles with the largest weight by using a cumulative distribution function.
9. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the image tracking method according to any one of claims 1 to 4.
10. An image tracking terminal, characterized in that it comprises a memory in which a computer program is stored and a processor which, when it is called up in said memory, implements the steps of the image tracking method according to any one of claims 1 to 4.
CN201910935692.1A 2019-09-29 2019-09-29 Image tracking method, system and related device Pending CN110648355A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910935692.1A CN110648355A (en) 2019-09-29 2019-09-29 Image tracking method, system and related device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910935692.1A CN110648355A (en) 2019-09-29 2019-09-29 Image tracking method, system and related device

Publications (1)

Publication Number Publication Date
CN110648355A true CN110648355A (en) 2020-01-03

Family

ID=68993259

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910935692.1A Pending CN110648355A (en) 2019-09-29 2019-09-29 Image tracking method, system and related device

Country Status (1)

Country Link
CN (1) CN110648355A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220051044A1 (en) * 2020-08-14 2022-02-17 Fujitsu Limited Image processing apparatus and computer-readable storage medium for storing screen processing program
CN115529606A (en) * 2021-06-25 2022-12-27 中国移动通信集团吉林有限公司 Parameter updating method and system and electronic equipment
CN116541668A (en) * 2023-07-06 2023-08-04 杭州光粒科技有限公司 Swimming stroke number determining method, device, equipment and storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102982555A (en) * 2012-11-01 2013-03-20 江苏科技大学 Guidance infrared small target tracking method based on self-adaption manifold particle filters
CN106355628A (en) * 2015-07-16 2017-01-25 中国石油化工股份有限公司 Image-text knowledge point marking method and device and image-text mark correcting method and system
CN108009571A (en) * 2017-11-16 2018-05-08 苏州大学 A kind of semi-supervised data classification method of new direct-push and system
CN109376787A (en) * 2018-10-31 2019-02-22 聚时科技(上海)有限公司 Manifold learning network and computer visual image collection classification method based on it
CN109978920A (en) * 2019-03-28 2019-07-05 中科智感科技(湖南)有限公司 A kind of method for tracking target and target following equipment

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102982555A (en) * 2012-11-01 2013-03-20 江苏科技大学 Guidance infrared small target tracking method based on self-adaption manifold particle filters
CN106355628A (en) * 2015-07-16 2017-01-25 中国石油化工股份有限公司 Image-text knowledge point marking method and device and image-text mark correcting method and system
CN108009571A (en) * 2017-11-16 2018-05-08 苏州大学 A kind of semi-supervised data classification method of new direct-push and system
CN109376787A (en) * 2018-10-31 2019-02-22 聚时科技(上海)有限公司 Manifold learning network and computer visual image collection classification method based on it
CN109978920A (en) * 2019-03-28 2019-07-05 中科智感科技(湖南)有限公司 A kind of method for tracking target and target following equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
JIANWEI DING ET AL.: "Tracking by local structural manifold learning in a new SSIR particle filter", 《NEUROCOMPUTING》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220051044A1 (en) * 2020-08-14 2022-02-17 Fujitsu Limited Image processing apparatus and computer-readable storage medium for storing screen processing program
US11682188B2 (en) * 2020-08-14 2023-06-20 Fujitsu Limited Image processing apparatus and computer-readable storage medium for storing screen processing program
CN115529606A (en) * 2021-06-25 2022-12-27 中国移动通信集团吉林有限公司 Parameter updating method and system and electronic equipment
CN116541668A (en) * 2023-07-06 2023-08-04 杭州光粒科技有限公司 Swimming stroke number determining method, device, equipment and storage medium
CN116541668B (en) * 2023-07-06 2023-09-22 杭州光粒科技有限公司 Swimming stroke number determining method, device, equipment and storage medium

Similar Documents

Publication Publication Date Title
Bottou et al. Optimization methods for large-scale machine learning
Yang et al. Feed-forward neural network training using sparse representation
CN110648355A (en) Image tracking method, system and related device
CN102214302A (en) Recognition device, recognition method, and program
CN110413878B (en) User-commodity preference prediction device and method based on adaptive elastic network
CN111063398A (en) Molecular discovery method based on graph Bayesian optimization
CN113342474B (en) Method, equipment and storage medium for predicting customer flow and training model
CN113821983A (en) Engineering design optimization method and device based on proxy model and electronic equipment
CN109117881A (en) A kind of multiple view cluster machine learning method with missing core
CN114186518A (en) Integrated circuit yield estimation method and memory
CN116796639A (en) Short-term power load prediction method, device and equipment
EP4032035A1 (en) Efficient computational inference
US20210125095A1 (en) Analysis method and analysis system
CN118298906A (en) Protein and small molecule docking method, device, electronic equipment and storage medium
CN117193008A (en) Small sample robust imitation learning training method oriented to high-dimensional disturbance environment, electronic equipment and storage medium
CN110275895B (en) Filling equipment, device and method for missing traffic data
CN107995027B (en) Improved quantum particle swarm optimization algorithm and method applied to predicting network flow
CN116543259A (en) Deep classification network noise label modeling and correcting method, system and storage medium
Chhablani et al. Multiplicative weight updates for extensive form games
Chang et al. A survey of some simulation-based algorithms for Markov decision processes
CN111027709B (en) Information recommendation method and device, server and storage medium
CN110675424A (en) Method, system and related device for tracking target object in image
CN114912518A (en) Reinforced learning grouping method, device and medium based on user group typical characteristics
CN113947030A (en) Equipment demand prediction method based on gradient descent gray Markov model
JP7438544B2 (en) Neural network processing device, computer program, neural network manufacturing method, neural network data manufacturing method, neural network utilization device, and neural network downsizing method

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200103