CN105096345B - A kind of method for tracking target and system based on dynamic calculation matrix - Google Patents
A kind of method for tracking target and system based on dynamic calculation matrix Download PDFInfo
- Publication number
- CN105096345B CN105096345B CN201510591575.XA CN201510591575A CN105096345B CN 105096345 B CN105096345 B CN 105096345B CN 201510591575 A CN201510591575 A CN 201510591575A CN 105096345 B CN105096345 B CN 105096345B
- Authority
- CN
- China
- Prior art keywords
- mrow
- msubsup
- msup
- msub
- calculation matrix
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Image Analysis (AREA)
- Medicines Containing Antibodies Or Antigens For Use As Internal Diagnostic Agents (AREA)
Abstract
The invention discloses a kind of method for tracking target based on dynamic calculation matrix and system, it comprises the following steps:S1:By the high dimensional feature boil down to low-dimensional feature of sample, and initialize dynamic calculation matrix;S2:Multiple positive sample set and negative sample set are gathered around target location, grader renewal learning is carried out;S3:It is determined that in the position of present frame target;S4:Dynamic calculation matrix is updated, and return to step S2, until tracking is completed.The present invention extracts the compressive features of target during tracking using dynamic calculation matrix, i.e. during tracking, calculation matrix is updated using the characteristic of the clarification of objective and Naive Bayes Classifier traced into, the fixed form of calculation matrix is improved, the adaptability of the tracking is stronger.The results show, the tracking has good robustness.
Description
Technical field
The present invention relates to computer vision field, more particularly to a kind of method for tracking target based on dynamic calculation matrix and
System.
Background technology
Target following based on video or image sequence is one of hot issue of computer vision, and target detection has
Contact closely, the tracking (tracking-by-detection) based on detection also has benefited from imitating on test problems
Rate and the lifting of the degree of accuracy.Target Tracking Problem needs the difficulty tackled a lot, such as illumination variation, target shape or posture become
The effect all to tracking such as change and complex scene has a great impact.
Tracking problem can substantially be divided into two classes.One class is the tracking based on generation model, this kind of method learning objective
Display model, when rope searches target location, finds and the region that modal distance is nearest or reconstructed error is minimum;Another kind of is to be based on
The tracking of discrimination model, this kind of method using tracking problem as two classes classification problem, using target and background as positive and negative
Two classes, learn the classification boundaries in two class sample characteristics spaces, take full advantage of contextual information.
MIT tracking target uses the detection mesh calibration method in subrange, is a kind of typical based on detection
Tracking.MIT algorithms select Haar-like to select feature and describe target and background, train a shellfish using online boosting methods
This grader of leaf realizes tracking.But fixed relationship of the Haar-like features on space structure also limit the description to target
Ability, goes to select feature in spite of online boosting methods, but training process is cumbersome.CT track algorithms are improved
The space constraint and training process of Haar-like features, i.e., described target and greatly simplified using broad sense Haar-like features
Training process.But calculation matrix just no longer changes after initialization in CT algorithms, the solidification of feature mode is result in, is caused at certain
Track and drift about under a little scenes.Application No. CN201410660331.8, a kind of entitled " video mesh based on compressed sensing
The patent of invention of mark tracking " just employs CT algorithms, but this applies in entirely tracking process all the time according to this matrix
To realize Feature Compression, that is to say, that keep sparse matrix constant during whole tracking, that is, result in consolidating for feature mode
Change, cause to track in some scenarios and drift about.
The content of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of target following based on dynamic calculation matrix
Method and system, in tracking process, improve the fixed form of calculation matrix, i.e., according to the compressive features of the target traced into and Piao
The characteristic of plain Bayes classifier updates calculation matrix, so as to provide a kind of adaptivity stronger method for tracking target.
The purpose of the present invention is achieved through the following technical solutions:A kind of target following based on dynamic calculation matrix
Method, it comprises the following steps:
S1:By the high dimensional feature boil down to low-dimensional feature of sample, and initialize dynamic calculation matrix;
S2:Multiple positive sample set and negative sample set are gathered around target location, grader renewal learning is carried out;
S3:It is determined that in the position of present frame target;
S4:Dynamic calculation matrix is updated, and return to step S2, until tracking is completed.
Step S1 includes following sub-step:
S11:By sampleHigh dimensional featureBoil down to low-dimensional featureI.e.:
V=R (t) x;
In formula, wherein m=(wh)2, n < < m,It is dynamic calculation matrix;
S12:Dynamic calculation matrix R (t) is initialized, that is, the element generated in R (0), formula is:
In formula, s=m/4, the sparse degree of representing matrix;I, j are respectively matrix R (0) ranks mark, ri,j(0) represent
The element of i rows j row.
Step S2 includes following sub-step:
S21:Gather multiple positive samples and multiple negative samples composition positive sample set Z respectively around targetαAnd negative sample
Set Zβ, wherein, α and β are the radius for choosing positive and negative samples, 0 < α < β;And obtain positive sample set ZαWith negative sample collection
Close ZβCompressive features collection VαAnd Vβ;
S22:Using Naive Bayes Classifier H (v), probabilistic model is set up respectively to compressive features v every dimension:
In formula, compressive features v ∈ Vα∪Vβ, compressive features v each element is independently distributed, and positive sample set Zα
With negative sample set ZβObedience is uniformly distributed i.e. p (y=1)=p (y=0);Y ∈ { 0,1 } representative sample label, y=0 represents negative
Sample, y=1 represents positive sample;
S23:Gauss model is set up respectively to compressive features v every dimension:
In formula,It is the average and variance of positive sample model i-th dimension,It is negative sample model i-th dimension
Average and variance;Target and sample pattern are by parameterRepresent;
S24:Grader learning process:Characteristic statisticses value is calculated, iteration updates positive sample model:
In formula, λ is learning rate,WithIt is according to step S23
Average and variance that positive sample modeling statistics is obtained.
Step S3 includes following sub-step:
S31:The collecting sample composition candidate samples set Z around previous frame targetγ, by ZγMiddle sample do feature extraction and
Compression, obtains compressive features v';
S32:Compressive features v' is input to Naive Bayes Classifier H (v), the compressive features v of highest scoring is obtained*:
v*Positive sample confidence level highest feature is categorized into, then this compressive features v*Corresponding sample is target sample,
Then the new position of target is obtained.
Step S4 includes following sub-step:
S41:Using Naive Bayes Classifier H (v) to compressive features v*Each dimensionality analysis:
S42:Define error function:
error(v* i)=- hi(v *);
And represent to need the corresponding row vector of renewal, i.e. the i-th dx dimensional vectors in dynamic calculation matrix using idx:
S43:New row vector is produced using random mannerRow vector R*Element be:
In formula, r1,j *It is row vector R*J-th of element;S=m/4, the sparse degree of representing matrix;
S44:The new row vector R obtained with step S43*The i-th dx dimensional vectors in dynamic calculation matrix are replaced, formula is such as
Under:
In formula, t represents current frame number, i.e., in t frames;
S45:Terminate if target following is completed, otherwise return to step S2.
A kind of Target Tracking System based on dynamic calculation matrix, it includes dynamic calculation matrix initialization module, training
Classifier modules, search object module and dynamic calculation matrix update module;Described dynamic calculation matrix initialization module is used
In the dynamic calculation matrix of initialization;Described training classifier modules are used to be updated according to the compressive features of positive negative sample and classified
Device;Described search object module is used to gather candidate samples around the target of previous frame, and the compression of candidate samples is special
Input grader is levied, the corresponding sample of compressive features of highest scoring is just considered as the target sample of present frame;Described dynamic
The target sample that calculation matrix update module is obtained using the characteristic of grader and by searching for object module measures square to dynamic
Battle array is updated.
Described grader is Naive Bayes Classifier.
The beneficial effects of the invention are as follows:The present invention proposes a kind of method for tracking target based on dynamic calculation matrix.Should
Method extracts the compressive features of target during tracking using dynamic calculation matrix, i.e., during tracking, using tracing into
Clarification of objective and Naive Bayes Classifier characteristic update calculation matrix, improve the fixed form of calculation matrix, should
The adaptability of tracking is stronger.The results show, the tracking has good robustness.
Brief description of the drawings
Fig. 1 is the inventive method flow chart;
Fig. 2 is that feature of present invention is extracted and compression process schematic diagram;
The renewal process schematic diagram of the dynamic calculation matrix of Fig. 3 present invention.
Embodiment
Technical scheme is described in further detail below in conjunction with the accompanying drawings:As shown in figure 1, a kind of based on dynamic measurement
The method for tracking target of matrix, it comprises the following steps:
S1:By the high dimensional feature boil down to low-dimensional feature of sample, and initialize dynamic calculation matrix;
S2:Multiple positive sample set and negative sample set are gathered around target location, grader renewal learning is carried out;
S3:It is determined that in the position of present frame target;
S4:Dynamic calculation matrix is updated, and return to step S2, until tracking is completed.
Step S1 includes following sub-step:
S11:By sampleHigh dimensional featureBoil down to low-dimensional featureI.e.:
V=R (t) x;
In formula, wherein m=(wh)2, n < < m,It is dynamic calculation matrix;
S12:Dynamic calculation matrix R (t) is initialized, that is, the element generated in R (0), formula is:
In formula, s=m/4, the sparse degree of representing matrix;I, j are respectively matrix R (0) ranks mark, ri,j(0) represent
The element of i rows j row.
Step S2 includes following sub-step:
S21:As shown in Fig. 2 the formula in step S12 gathers multiple positive samples and multiple negative respectively around target
Sample composition positive sample set ZαWith negative sample set Zβ, wherein, α and β are the radius for choosing positive and negative samples, 0 < α < β;
And obtain positive sample set ZαWith negative sample set ZβCompressive features collection VαAnd Vβ;
S22:Using Naive Bayes Classifier H (v), probabilistic model is set up respectively to compressive features v every dimension:
In formula, compressive features v ∈ Vα∪Vβ, compressive features v each element is independently distributed, and positive sample set Zα
With negative sample set ZβObedience is uniformly distributed i.e. p (y=1)=p (y=0);Y ∈ { 0,1 } representative sample label, y=0 represents negative
Sample, y=1 represents positive sample;
S23:Gauss model is set up respectively to compressive features v every dimension:
In formula,It is the average and variance of positive sample model i-th dimension,It is negative sample model i-th dimension
Average and variance;Target and sample pattern are by parameterRepresent;
S24:Grader learning process:Characteristic statisticses value is calculated, iteration updates positive sample model:
In formula, λ is learning rate,WithIt is according to step S23
Average and variance that positive sample modeling statistics is obtained.
Step S3 includes following sub-step:
S31:The collecting sample composition candidate samples set Z around previous frame targetγ, by ZγMiddle sample do feature extraction and
Compression, obtains compressive features v', as shown in the formula in step S12;
S32:Compressive features v' is input to Naive Bayes Classifier H (v), the compressive features v of highest scoring is obtained*:
v*Positive sample confidence level highest feature is categorized into, then this compressive features v*Corresponding sample is target sample,
Then the new position of target is obtained.
Step S4 includes following sub-step:
S41:Using Naive Bayes Classifier H (v) to compressive features v*Each dimensionality analysis:
S42:v*In be classified as positive sample may a minimum dimension be that we are concerned about " bad feature ", it is represented should
Dimensional feature can not preferably describe positive negative sample, align sample classification and had the opposite effect for positive sample;Still define error function:
error(v* i)=- hi(v*);
The dimension of above-mentioned " bad feature " corresponding row vector in dynamic calculation matrix is represented with idx, i.e.,:
So, it is determined that the row vector changed, as the i-th dx dimensional vectors are needed in dynamic calculation matrix;
S43:New row vector is produced using random mannerRow vector R*Element be:
In formula, r1,j *It is row vector R*J-th of element;S=m/4, the sparse degree of representing matrix;
S44:The renewal process of dynamic calculation matrix is as shown in figure 3, the new row vector R obtained with step S43*Replace dynamic
I-th dx dimensional vectors in state calculation matrix, formula is as follows:
In formula, t represents current frame number, i.e., in t frames;
S45:Terminate if target following is completed, otherwise return to step S2.
A kind of Target Tracking System based on dynamic calculation matrix, it includes dynamic calculation matrix initialization module, training
Classifier modules, search object module and dynamic calculation matrix update module;Described dynamic calculation matrix initialization module is used
In the dynamic calculation matrix of initialization;Described training classifier modules are used to be updated according to the compressive features of positive negative sample and classified
Device;Described search object module is used to gather candidate samples around the target of previous frame, and the compression of candidate samples is special
Input grader is levied, the corresponding sample of compressive features of highest scoring is just considered as the target sample of present frame;Described dynamic
The target sample that calculation matrix update module is obtained using the characteristic of grader and by searching for object module measures square to dynamic
Battle array is updated.
Described grader is Naive Bayes Classifier.
Claims (4)
1. a kind of method for tracking target based on dynamic calculation matrix, it is characterised in that:It comprises the following steps:
S1:By the high dimensional feature boil down to low-dimensional feature of sample, and initialize dynamic calculation matrix;
S2:Multiple positive sample set and negative sample set are gathered around target location, grader renewal learning is carried out;
S3:It is determined that in the position of present frame target;
S4:Dynamic calculation matrix is updated, and return to step S2, until tracking is completed;
Step S4 includes following sub-step:
S41:Using Naive Bayes Classifier H (v) to compressive features v*Each dimensionality analysis:
<mrow>
<msub>
<mi>h</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<msup>
<mi>v</mi>
<mo>*</mo>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<msup>
<mi>v</mi>
<mo>*</mo>
</msup>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<msup>
<mi>v</mi>
<mo>*</mo>
</msup>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
S42:Define error function:
error(v* i)=- hi(v*);
And represent to need the corresponding row vector of renewal, i.e. the i-th dx dimensional vectors in dynamic calculation matrix using idx:
<mrow>
<mi>i</mi>
<mi>d</mi>
<mi>x</mi>
<mo>=</mo>
<mi>arg</mi>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mn>1</mn>
<mo>&le;</mo>
<mi>i</mi>
<mo>&le;</mo>
<mi>n</mi>
</mrow>
</munder>
<mi>e</mi>
<mi>r</mi>
<mi>r</mi>
<mi>o</mi>
<mi>r</mi>
<mrow>
<mo>(</mo>
<msub>
<msup>
<mi>v</mi>
<mo>*</mo>
</msup>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
S43:New row vector is produced using random mannerRow vector R*Element be:
In formula, r1,j *It is row vector R*J-th of element;S=m/4, the sparse degree of representing matrix;
S44:The new row vector R obtained with step S43*The i-th dx dimensional vectors in dynamic calculation matrix are replaced, formula is as follows:
<mrow>
<msub>
<mi>r</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>r</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mo>&NotEqual;</mo>
<mi>i</mi>
<mi>d</mi>
<mi>x</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msubsup>
<mi>r</mi>
<mrow>
<mn>1</mn>
<mo>,</mo>
<mi>j</mi>
</mrow>
<mo>*</mo>
</msubsup>
<mo>,</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mi>i</mi>
<mi>d</mi>
<mi>x</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>,</mo>
</mrow>
In formula, t represents current frame number, i.e., in t frames;
S45:Terminate if target following is completed, otherwise return to step S2.
2. a kind of method for tracking target based on dynamic calculation matrix according to claim 1, it is characterised in that:Step S1
Including following sub-step:
S11:By sampleHigh dimensional featureBoil down to low-dimensional featureI.e.:
V=R (t) x;
In formula, wherein m=(wh)2, n=m,It is dynamic calculation matrix;
S12:Dynamic calculation matrix R (t) is initialized, that is, the element generated in R (0), formula is:
In formula, s=m/4, the sparse degree of representing matrix;I, j are respectively matrix R (0) ranks mark, ri,j(0) i rows j is represented
The element of row.
3. a kind of method for tracking target based on dynamic calculation matrix according to claim 1, it is characterised in that:Step S2
Including following sub-step:
S21:Gather multiple positive samples and multiple negative samples composition positive sample set Z respectively around targetαWith negative sample set Zβ, wherein, α and β are the radius for choosing positive and negative samples, 0 < α < β;And obtain positive sample set ZαWith negative sample set Zβ
Compressive features collection VαAnd Vβ;
S22:Using Naive Bayes Classifier H (v), probabilistic model is set up respectively to compressive features v every dimension:
<mrow>
<mi>H</mi>
<mrow>
<mo>(</mo>
<mi>v</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msubsup>
<mi>&Pi;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>=</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msubsup>
<mi>&Pi;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<mi>y</mi>
<mo>=</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
In formula, compressive features v ∈ VαUVβ, compressive features v each element is independently distributed, and positive sample set ZαWith negative sample
This set ZβObedience is uniformly distributed i.e. p (y=1)=p (y=0);Y ∈ { 0,1 } representative sample label, y=0 represents negative sample, y
=1 represents positive sample;
S23:Gauss model is set up respectively to compressive features v every dimension:
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
<mo>:</mo>
<mi>N</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&mu;</mi>
<mi>i</mi>
<mn>0</mn>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>&sigma;</mi>
<mi>i</mi>
<mn>0</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
<mrow>
<mi>p</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>v</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mi>y</mi>
<mo>=</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>:</mo>
<mi>N</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&mu;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>&sigma;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
In formula, It is the average and variance of positive sample model i-th dimension, It is the average and variance of negative sample model i-th dimension;
Target and sample pattern are by parameterRepresent;
S24:Grader learning process:Characteristic statisticses value is calculated, iteration updates positive sample model:
<mrow>
<msubsup>
<mi>&mu;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>&LeftArrow;</mo>
<msubsup>
<mi>&lambda;&mu;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>+</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>&lambda;</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>&mu;</mi>
<mn>1</mn>
</msup>
<mo>;</mo>
</mrow>
<mrow>
<msubsup>
<mi>&sigma;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>&LeftArrow;</mo>
<msqrt>
<mrow>
<mi>&lambda;</mi>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&sigma;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>&lambda;</mi>
<mo>)</mo>
</mrow>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>&sigma;</mi>
<mn>1</mn>
</msup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mi>&lambda;</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>&lambda;</mi>
<mo>)</mo>
</mrow>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>&mu;</mi>
<mi>i</mi>
<mn>1</mn>
</msubsup>
<mo>-</mo>
<msup>
<mi>&mu;</mi>
<mn>1</mn>
</msup>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<mo>;</mo>
</mrow>
In formula, λ is learning rate,WithIt is the positive sample according to step S23
Average and variance that modeling statistics is obtained.
4. a kind of method for tracking target based on dynamic calculation matrix according to claim 1, it is characterised in that:Step S3
Including following sub-step:
S31:The collecting sample composition candidate samples set Z around previous frame targetγ, by ZγMiddle sample does feature extraction and pressure
Contracting, obtains compressive features v';
S32:Compressive features v' is input to Naive Bayes Classifier H (v), the compressive features v of highest scoring is obtained*:
<mrow>
<msup>
<mi>v</mi>
<mo>*</mo>
</msup>
<mo>=</mo>
<mi>arg</mi>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mi>v</mi>
<mo>&Element;</mo>
<msup>
<mi>V</mi>
<mi>&gamma;</mi>
</msup>
</mrow>
</munder>
<mi>H</mi>
<mrow>
<mo>(</mo>
<mi>v</mi>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
v*Positive sample confidence level highest feature is categorized into, then this compressive features v*Corresponding sample is target sample, then
Obtain the new position of target.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510591575.XA CN105096345B (en) | 2015-09-15 | 2015-09-15 | A kind of method for tracking target and system based on dynamic calculation matrix |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510591575.XA CN105096345B (en) | 2015-09-15 | 2015-09-15 | A kind of method for tracking target and system based on dynamic calculation matrix |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105096345A CN105096345A (en) | 2015-11-25 |
CN105096345B true CN105096345B (en) | 2017-11-07 |
Family
ID=54576683
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510591575.XA Active CN105096345B (en) | 2015-09-15 | 2015-09-15 | A kind of method for tracking target and system based on dynamic calculation matrix |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105096345B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105976397B (en) * | 2016-04-28 | 2019-03-26 | 西安电子科技大学 | A kind of method for tracking target |
CN107066922B (en) * | 2016-12-30 | 2021-05-07 | 西安天和防务技术股份有限公司 | Target tracking method for monitoring homeland resources |
CN107122722A (en) * | 2017-04-19 | 2017-09-01 | 大连理工大学 | A kind of self-adapting compressing track algorithm based on multiple features |
CN107977982B (en) * | 2017-11-30 | 2021-11-02 | 云南大学 | Video target tracking method based on compressed regularization block difference |
CN110418137B (en) * | 2019-07-31 | 2021-05-25 | 东华大学 | Cross subset guided residual block set measurement rate regulation and control method |
CN111127514B (en) * | 2019-12-13 | 2024-03-22 | 华南智能机器人创新研究院 | Method and device for tracking target by robot |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103632382B (en) * | 2013-12-19 | 2016-06-22 | 中国矿业大学(北京) | A kind of real-time multiscale target tracking based on compressed sensing |
CN104392467A (en) * | 2014-11-18 | 2015-03-04 | 西北工业大学 | Video target tracking method based on compressive sensing |
CN104616302A (en) * | 2015-02-04 | 2015-05-13 | 四川中科腾信科技有限公司 | Real-time object identification method |
-
2015
- 2015-09-15 CN CN201510591575.XA patent/CN105096345B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN105096345A (en) | 2015-11-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105096345B (en) | A kind of method for tracking target and system based on dynamic calculation matrix | |
Wang et al. | Student-teacher feature pyramid matching for anomaly detection | |
CN104850865B (en) | A kind of Real Time Compression tracking of multiple features transfer learning | |
Gao et al. | Dynamic zoom-in network for fast object detection in large images | |
CN104599292B (en) | A kind of anti-noise moving object detection algorithm decomposed based on low-rank matrix | |
CN103258214B (en) | Based on the Classifying Method in Remote Sensing Image of image block Active Learning | |
CN103345643B (en) | A kind of Classifying Method in Remote Sensing Image | |
CN106204638A (en) | A kind of based on dimension self-adaption with the method for tracking target of taking photo by plane blocking process | |
CN111079602A (en) | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint | |
CN102034096B (en) | Video event recognition method based on top-down motion attention mechanism | |
CN107679465A (en) | A kind of pedestrian's weight identification data generation and extending method based on generation network | |
CN107423747B (en) | A kind of conspicuousness object detection method based on depth convolutional network | |
CN109063649B (en) | Pedestrian re-identification method based on twin pedestrian alignment residual error network | |
CN108734210A (en) | A kind of method for checking object based on cross-module state multi-scale feature fusion | |
CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
CN107169994A (en) | Correlation filtering tracking based on multi-feature fusion | |
CN105976397B (en) | A kind of method for tracking target | |
CN105701482A (en) | Face recognition algorithm configuration based on unbalance tag information fusion | |
CN106780552A (en) | Anti-shelter target tracking based on regional area joint tracing detection study | |
CN105574545B (en) | The semantic cutting method of street environment image various visual angles and device | |
CN103065158A (en) | Action identification method of independent subspace analysis (ISA) model based on relative gradient | |
CN106651917A (en) | Image target tracking algorithm based on neural network | |
CN103268484A (en) | Design method of classifier for high-precision face recognitio | |
CN106127766B (en) | Method for tracking target based on Space Coupling relationship and historical models | |
CN110414616A (en) | A kind of remote sensing images dictionary learning classification method using spatial relationship |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |