CN108492313A - A kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure - Google Patents
A kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure Download PDFInfo
- Publication number
- CN108492313A CN108492313A CN201810114328.4A CN201810114328A CN108492313A CN 108492313 A CN108492313 A CN 108492313A CN 201810114328 A CN201810114328 A CN 201810114328A CN 108492313 A CN108492313 A CN 108492313A
- Authority
- CN
- China
- Prior art keywords
- target
- histogram
- background
- feature
- intelligence
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 230000000007 visual effect Effects 0.000 title claims abstract description 19
- 238000011524 similarity measure Methods 0.000 title claims abstract description 18
- 238000005259 measurement Methods 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims abstract description 10
- 230000003321 amplification Effects 0.000 claims description 14
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 14
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 238000004891 communication Methods 0.000 claims description 4
- XCWPUUGSGHNIDZ-UHFFFAOYSA-N Oxypertine Chemical compound C1=2C=C(OC)C(OC)=CC=2NC(C)=C1CCN(CC1)CCN1C1=CC=CC=C1 XCWPUUGSGHNIDZ-UHFFFAOYSA-N 0.000 claims description 3
- 238000004040 coloring Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 11
- 238000005516 engineering process Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000004438 eyesight Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000001154 acute effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 101150039161 hasB gene Proteins 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000035807 sensation Effects 0.000 description 1
- 230000001225 therapeutic effect Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
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
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration using histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/223—Analysis of motion using block-matching
- G06T7/238—Analysis of motion using block-matching using non-full search, e.g. three-step search
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of dimension self-adaption visual target tracking methods based on middle intelligence similarity measure, choose target area to be tracked including in initial frame, calculate target signature histogram and initial background histogram;True, false, uncertain measure is carried out for target signature attribute, background characteristics like attribute;Intelligence weight vector in foundation;Middle intelligence weight vector is introduced into average drifting strategy to determine present frame target area;For reducing, expanding corresponding true, false, the Uncertainty measured value of dimension calculation, scale more new strategy is determined according to the similar measurement of cosine;Update target background feature histogram.The present invention uses extremely efficient mean shift algorithm, and corresponding middle intelligence measurement calculation amount is small, and weight vector and size estimation complexity are low, efficient, meet real-time modeling method demand;Intelligence collection is theoretical in utilization, and tracked target changing features, target/background characteristics similitude are included in and are considered, and effectively improves the tracking performance when challenges such as track algorithm reply complex background.
Description
Technical field
The present invention relates to technical field of computer vision, in particular to a kind of dimension self-adaptions based on middle intelligence similarity measure
Visual target tracking method.
Background technology
With the promotion of development in science and technology and safety urban construction demand, such as video monitoring, video frequency searching, wisdom traffic,
The computer visions such as automatic Pilot apply the effect ever more important played in we live.However, target following is as it
In a key technology, be still a challenging problem.
In target tracking domain, mean shift algorithm is widely used in visual target tracking.During tracking, mean value drift
It moves algorithm and current goal position is determined at a distance from candidate target region probability density function by minimum tracking target.By
Color histogram feature and efficient target-region locating strategy are used in average drifting track algorithm, efficiency is very high, and energy
More effectively overcome the challenges such as motion blur, target deformation.However, when background area is close with target signature, the algorithm pole
It easily shifts, eventually leads to tracking failure.In view of such problem, some new metric forms or feature are introduced into, such as
Cross-Bin measurements, SIFT feature, textural characteristics, this improves algorithm robustness to a certain extent.In addition, background information
It is introduced into a manner of characteristic weighing, forward-backward algorithm Detection of Stability method is also suggested to reply target scale estimation, through inspection
It surveys, such improved mean shift algorithm has performed more than the increasingly complex track algorithm of most mechanism haveing excellent performance.
In terms of scale update, the strategy such as difference scale space filter is also suggested, and achieves certain effect.Although have much with
Track algorithm is suggested, but the challenges such as illumination variation, dimensional variation, motion blur, complex background are faced in object tracking process,
Target following is still an open problem.In numerous algorithms, mean shift algorithm is efficient, and has and show more preferable property
The possibility of energy, the separating capacity that there is an urgent need to have significantly more efficient mechanism to be suggested to boosting algorithm to target and background add
Its strong robustness.Therefore, find out it is a kind of it is abundant fusion seemed particularly using the method for target itself and target background area information
It is important.
Middle intelligence collection theory is proposed by Smarandache, and compared to traditional fuzzy theory, middle intelligence collection theory is in addition to true
(Truth), outside false (Falsity) component is stated, place specially also has been carried out to uncertain (Indeterminacy) component
Reason.Middle intelligence collection theory has shown very big advantage in terms of handling uncertain information.In order to which middle intelligence collection theory is led for engineering
Domain, the concept of intelligence collection is suggested in monodrome, and the value of true, false, uncertain component is limited within the scope of 0 to 1 closed interval.To incite somebody to action
Intelligence collection is used for engineering decision in monodrome, and the similar measurement such as cosine, tangent is suggested.In view of the validity of middle intelligence theory, by
It is widely used in computer vision, Steam Turbine Fault Diagnosis, therapeutic scheme selection etc..For example, for the image point in visual analysis
Problem is cut, middle intelligence similitude clustering method, uncertain filter etc. are introduced image segmentation by some researchers, improve the anti-of algorithm
It makes an uproar performance.During tracking, uncertain information is equally existed.For average drifting track algorithm, color histogram by with
To characterize tracked target.On the one hand, target feature itself can be affected with the variation of target pose or external environment;
In addition, for determining that the tracking box of target location is also the estimated value obtained according to target signature, extracted using the target frame
Target signature is also a uncertain problem.Intelligence theory appropriately introduces uncertain factor to promote track algorithm in how utilizing
Robustness, constitute core of the invention problem.
Invention content
The present invention provides a kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure, purpose exists
It in overcoming defect in the prior art, realizes that simple, computation complexity is low, noiseproof feature is good, can preferably complete background complexity,
Target following in the case of the target scale variation great challenge such as greatly.
To achieve the goals above, the present invention has following constitute:
The dimension self-adaption visual target tracking method based on middle intelligence similarity measure, includes the following steps:
S100:Target area to be tracked is chosen in initial frame, calculates target signature histogram and initial background histogram;
S200:True, false, uncertain measure is carried out for target signature attribute, background characteristics like attribute;
S300:Intelligence weight vector in foundation;
S400:Middle intelligence weight vector is introduced into average drifting strategy to determine present frame target area;
S500:For diminution, expand dimension calculation true accordingly, vacation, Uncertainty measured value, foundation cosine is similar to be measured really
Dimensioning more new strategy;
S600:Update target background feature histogram.
Optionally, in step S100, target signature histogram and initial background histogram is calculated, is included the following steps:
Target signature histogramIt is calculated by following formula:
Wherein,It is histogramOne-component,There is m component, and
Image coordinate for all pixels point in target area with respect to rectangle frame center, b (x) are a mapping letters
Number, by the colouring information of x pixels position be mapped in 1 in the sections m some numerically;
K (x) is kernel function;δ (x) is Kronecker function, and C is normalized parameter, and n is target area pixel sum;
Background area is defined as the equal proportion flared region of target area, and two regional centers are in former target area
The heart.It is assumed that target area is Go, then background area Gb=λ Go-Go, λ is sampling factor.Initial background histogramBy following public
Formula calculates:
Wherein,It is histogramOne-component,There is m component, and
It is correspondingTo fall into the image coordinate at all pixels point relative target rectangle frame center in target background region, k
(x) it is kernel function;δ (x) is Kronecker function, and C ' is normalized parameter, and n ' is target background area pixel point sum.
Optionally, in step S200, true, false, uncertain measurement is carried out, is included the following steps:
For target signature attribute CO, the true T of t momentCO, uncertain ICO, vacation FCOMeasurement is respectively defined as:
WhereinHeaded by frame initialize when target signature histogram in u-th of component,For former frame target signature
Updated feature histogram u components,Wherein λ ∈ (0,1);
For background characteristics like attribute CB, the T of t momentCB、ICB、FCBMeasurement is respectively defined as:
WhereinFor u-th of component of background area feature histogram.
Optionally, in step S300, intelligence weight vector includes the following steps in foundation:
Weight vector is made of m component, and quantity m is corresponding with target signature histogram component quantity, similar according to cosine
It measures, it is as follows to obtain middle intelligence weights for u-th of component in histogram:
Wherein, wCO、wCBRespectively target signature attribute and the corresponding weights of background characteristics like attribute, wCO,wCB∈[0,
1], two weights sums are 1.
Optionally, in step S400 by middle intelligence weight vector introduce average drifting strategy to determine present frame target area,
Include the following steps:
Using the target location of previous frame as the search starting point of present frame target locationCalculate new target location such as
Under:
Wherein g (x)=- k ' (x),Wherein δ (x) is Kronecker function,
Indicate u-th of component in intelligence weight vector in t moment,Wherein xiFor candidate region
Interior pixel point coordinates, nhFor the total pixel number in candidate region, ChFor normalized parameter, k (x) is kernel function, and h is candidate region
Bandwidth, if being unsatisfactory forIt willWhereinIt indicates the Euclidean distance between two coordinate points, repeats
S400 is executed, until meet end condition, ε herein0To preset judgment threshold.
Optionally, in step S500, for diminution, expand corresponding true, false, the Uncertainty measured value of dimension calculation, foundation
The similar measurement of cosine determines scale more new strategy, includes the following steps:
The case where scale reduces is considered first, based on the analysis to authentic communication and uncertain information, is reduced scale and is corresponded to
Tsa、Isa、FsaMeasurement is respectively defined as:
WhereinThe feature histogram in respective cell domain after being reduced for present frame target area, the zonule is with current
Target area is reference, and regional center is constant, and length and width by reducing λ in proportionsaTimes, λsaFor corresponding zoom factor,It is initial
Goal histogram,The feature histogram of background area is corresponded to for current area domain;
It is measured according to cosine similarity, it is as follows that the ideally corresponding middle intelligence similarity weights of diminution scale can be obtained:
The case where considering scale amplification later obtains amplification scale based on the analysis to authentic communication and uncertain information
Corresponding middle intelligence similarity weight wbscal, wherein amplification coefficient is λba, take λba=λsa, the corresponding T of amplification scaleba、Iba、FbaAmount
Survey is respectively defined as:
Wherein,The feature histogram in big region is corresponded to after amplifying for present frame target area, the big region is with current
Target area is reference, and regional center is constant, and length and width by amplifying λ in proportionbaTimes,For initial target histogram,To work as
Preceding big region corresponds to the feature histogram of background area;
It is measured according to cosine similarity, it is as follows that the ideally corresponding middle intelligence similarity weights of amplification scale can be obtained:
Obtain wsscalAnd wbscalAfterwards, present frame target scale is as follows:
Wherein, s is preset scale factor, and meets s>1;
Determine λnewAfterwards, target following frame is scaled into λnewAgain as new tracking box.
Optionally, in step S600, the method for target background feature histogram is updated, is included the following steps:
Calculate present frame target background feature histogramIf Indicating willIt assigns
Value isWherein ρ is Pasteur's related coefficient, The background that frame moment algorithm uses thus
Feature histogram, ε1To preset judgment threshold.
Optionally, further include following steps after step S600:
S700:Next video frame is chosen, step S200~S600 is then executed, until being disposed all video frame.
The dimension self-adaption visual target tracking method based on middle intelligence similarity measure in the invention is used, is had as follows
Advantageous effect:
(1) present invention uses extremely efficient mean shift algorithm, and corresponding middle intelligence measurement calculation amount is small, weight vector and ruler
Degree estimation complexity is low, efficient, meets real-time modeling method demand;
(2) intelligence collection is theoretical during the present invention utilizes, and tracked target changing features, target/background characteristics similitude are included in
It considers, effectively improves the tracking performance when challenges such as track algorithm reply complex background;
(3) middle intelligence theory is introduced target scale estimation by the present invention, comprehensively utilizes target signature and background characteristics is jointly true
Set the goal dimensional information, and noiseproof feature is high, can estimate compared with the completion target scale of robust, promote Image Tracking Algorithms Performance.
Description of the drawings
Fig. 1 is the flow chart of the dimension self-adaption visual target tracking method based on middle intelligence similarity measure of the present invention.
Specific implementation mode
In order to more clearly describe the technology contents of the present invention, carried out with reference to specific embodiment further
Description.
Dimension self-adaption visual target tracking method based on middle intelligence similarity measure as shown in Figure 1, includes the following steps:
Step 1:A web camera is set up in monitoring area, and the video data real-time Transmission acquired is extremely counted
Calculation machine terminal.
Step 2:Terminal reads the image data that video camera transmits in real time in an rgb format.
Step 3:Target area to be tracked is chosen manually in initial frame, calculates target signature histogram and initial background is straight
Fang Tu.Target signature histogram is calculated by following formula:
WhereinIt is histogramOne-component, it is assumed thatThere is m component, then hasB (x) is a mapping letter
Number, by the colouring information of x pixels position be mapped in 1 in the sections m some numerically.
K (x) is kernel function;δ (x) is Kronecker function, and C is normalized parameter, and n is target area pixel sum.
Background area is defined as the equal proportion flared region of target area, and two regional centers are in former target area
The heart, it is assumed that target area Go, then background area Gb=λ Go-Go, λ is sampling factor, initial background histogramBy following public
Formula calculates:
Wherein,It is histogramOne-component,There is m component, and
It is correspondingTo fall into the image coordinate at all pixels point relative target rectangle frame center in target background region, k
(x) it is kernel function;δ (x) is Kronecker function, and C ' is normalized parameter, and n ' is target background area pixel point sum.
In the present invention, background area is defined as the equal proportion flared region of target area, and two regional centers are
Former target area center.It is assumed that target area is Go, then background area Gb=λ Go-Go, λ is sampling factor.In the present embodiment, often
A Color Channel is divided into 16 parts, and target signature histogram and background characteristics histogram have 4096 components, background area
Domain sampling factor λ is set as 1.6.
Step 4:Corresponding true, false, Uncertainty measured value is calculated for target signature attribute, background characteristics like attribute.
For target signature attribute CO, true, uncertain, false (T, the I, F) measuring value of t moment is calculated respectively in accordance with following formula:
Wherein,For the updated feature histogram u components of former frame target signature, meet following formula:
Wherein histogram update coefficient lambda ∈ (0,1).
For background characteristics like attribute CB, T, I, F measuring value of t moment are calculated respectively in accordance with following formula:
Wherein,For u-th of component of background area feature histogram.In the present embodiment, histogram update coefficient is set as
0.05。
Step 5:Intelligence weight vector in foundation.Weight vector be made of m component (in this embodiment, middle intelligence weights to
Amount is made of 4096 components), according to cosine similarity amount, the middle intelligence weights of u-th of component meet following formula in histogram:
Wherein wCO、wCBThe corresponding weights of respectively two conditional attributes, wCO,wCB∈ [0,1], two weights sums are 1.
Ideal chose A herein*It is defined as A under any attribute conditions*=<1,0,0>, it is corresponding as a result, in intelligence weights can be with
It is reduced to following formula:
Wherein wCO、wCBThe corresponding weights of respectively two conditional attributes, wCO,wCB∈ [0,1], two weights sums are 1.
The present embodiment wCO,wCBIt is respectively set as 0.4,0.6, however, the present invention is not limited thereto.
Step 6:Weight vector is introduced into average drifting strategy to determine present frame target area.With the target of previous frame
Search starting point of the position as present frame target locationCalculate new target location
Wherein,If being unsatisfactory forIt willRepeat this
Step, until meet end condition, ε herein0It should be set as smaller value, avoid influencing tracking accuracy.The present embodiment ε0It is set as
0.1。
Step 7:For reducing, expanding corresponding true, false, the Uncertainty measured value of dimension calculation, according to the similar measurement of cosine
Determine scale more new strategy.The case where scale reduces is considered first, reduces corresponding T, I, F measuring value of scale according to following formula
It calculates:
WhereinThe feature histogram in respective cell domain after being reduced for present frame target area, equally with current goal
Region is reference, and regional center is constant, and length and width by reducing λ in proportionsaTimes, λsaFor corresponding zoom factor.For initial target
Histogram.The feature histogram of background area is corresponded to for current region.It is easy to find, TsaFor target area reduce after with
The similarity of initial target feature, IsaFor target area reduce after correspond to background area andSimilarity.Ideal herein
Select A*It is also defined as A under any attribute conditions*=<1,0,0>, measured according to cosine similarity, can be obtained and selected with ideal
The similarity selectedAccording to same strategy, the corresponding middle intelligence similarity weights of amplification scale are calculated
wbscal, wherein amplification coefficient is λba, take λsa=λba.Specifically, the corresponding T of amplification scaleba、Iba、FbaMeasurement is defined respectively
For:
Wherein,The feature histogram in big region is corresponded to after amplifying for present frame target area, the big region is with current
Target area is reference, and regional center is constant, and length and width by amplifying λ in proportionbaTimes,For initial target histogram,To work as
Preceding big region corresponds to the feature histogram of background area.
Obtain wsscalAnd wbscalAfterwards, present frame target scaleS is scale factor,
And meet s>1.The setting of the scale factor be in order to avoid rescaling is excessively frequent, or adjustment when by noise jamming.Really
Determine λnewAfterwards, target following frame is amplified into λnewAgain as new tracking box.In the present embodiment, the λ with amplification coefficient is reducedbsaWith
λbaIt can be set to that 0.3, scale factor s is set to 1.1, λ0It must be set to a smaller value, it is acute to avoid target frame
Strong variation, herein λ0It is set as 0.04, however, the present invention is not limited thereto.
Step 8:Update target background feature histogram.Calculate present frame target background feature histogramIf Wherein ρ is Pasteur's related coefficient, the b background characteristics histograms that frame moment algorithm uses thus
Figure.ε1Renewal rate is determined to a certain extent, should take the value of a compromise.The present embodiment ε1It is set as 0.5, but the present invention is not
It is limited to this.
Step 9:After having handled current video frame, next new video frame is chosen, then executes step 4 to step
Eight, until all video frame are disposed.
Compared with prior art, use the dimension self-adaption sensation target based on middle intelligence similarity measure in the invention with
Track method, has the advantages that:
(1) present invention uses extremely efficient mean shift algorithm, and corresponding middle intelligence measurement calculation amount is small, weight vector and ruler
Degree estimation complexity is low, efficient, meets real-time modeling method demand;
(2) intelligence collection is theoretical during the present invention utilizes, and tracked target changing features, target/background characteristics similitude are included in
It considers, effectively improves the tracking performance when challenges such as track algorithm reply complex background;
(3) middle intelligence theory is introduced target scale estimation by the present invention, comprehensively utilizes target signature and background characteristics is jointly true
Set the goal dimensional information, and noiseproof feature is high, can estimate compared with the completion target scale of robust, promote Image Tracking Algorithms Performance.
In this description, the present invention is described with reference to its specific embodiment.But it is clear that can still make
Various modifications and alterations are without departing from the spirit and scope of the invention.Therefore, the description and the appended drawings should be considered as illustrative
And not restrictive.
Claims (8)
1. a kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure, which is characterized in that including walking as follows
Suddenly:
S100:Target area to be tracked is chosen in initial frame, calculates target signature histogram and initial background histogram;
S200:True, false, uncertain measure is carried out for target signature attribute, background characteristics like attribute;
S300:Intelligence weight vector in foundation;
S400:Middle intelligence weight vector is introduced into average drifting strategy to determine present frame target area;
S500:For reducing, expanding corresponding true, false, the Uncertainty measured value of dimension calculation, ruler is determined according to the similar measurement of cosine
Spend more new strategy;
S600:Update target background feature histogram.
2. the dimension self-adaption visual target tracking method according to claim 1 based on middle intelligence similarity measure, feature
It is, in step S100, calculates target signature histogram and initial background histogram, include the following steps:
Target signature histogramIt is calculated by following formula:
Wherein,It is histogramOne-component,There is m component, and
Image coordinate for all pixels point in target area with respect to rectangle frame center, b (x) is a mapping function, will
The colouring information of x pixels position be mapped in 1 in the sections m some numerically;
K (x) is kernel function;δ (x) is Kronecker function, and C is normalized parameter, and n is target area pixel sum;
Background area is defined as the equal proportion flared region of target area, and two regional centers are former target area center,
It is assumed that target area is Go, then background area Gb=λ Go-Go, λ is sampling factor, initial background histogramPass through following formula
It calculates:
Wherein,It is histogramOne-component,There is m component, and
It is correspondingTo fall into the image coordinate at all pixels point relative target rectangle frame center in target background region, k (x) is
Kernel function;δ (x) is Kronecker function, and C ' is normalized parameter, and n ' is target background area pixel point sum.
3. the dimension self-adaption visual target tracking method according to claim 2 based on middle intelligence similarity measure, feature
It is, in step S200, carries out true, false, uncertain measurement, include the following steps:
For target signature attribute CO, the true T of t momentCO, uncertain ICO, vacation FCOMeasurement is respectively defined as:
WhereinHeaded by frame initialize when target signature histogram in u-th of component,It is updated for former frame target signature
Feature histogram u components afterwards,Wherein λ ∈ (0,1);
For background characteristics like attribute CB, the T of t momentCB、ICB、FCBMeasurement is respectively defined as:
WhereinFor u-th of component of background area feature histogram.
4. the dimension self-adaption visual target tracking method according to claim 3 based on middle intelligence similarity measure, feature
It is, in step S300, intelligence weight vector includes the following steps in foundation:
Weight vector is made of m component, and quantity m is corresponding with target signature histogram component quantity, foundation cosine similarity amount,
It is as follows to obtain middle intelligence weights for u-th of component in histogram:
Wherein, wCO、wCBRespectively target signature attribute and the corresponding weights of background characteristics like attribute, wCO,wCB∈ [0,1], two
A weights sum is 1.
5. the dimension self-adaption visual target tracking method according to claim 4 based on middle intelligence similarity measure, feature
It is, middle intelligence weight vector is introduced into average drifting strategy to determine present frame target area in step S400, including walk as follows
Suddenly:
Using the target location of previous frame as the search starting point of present frame target locationIt is as follows to calculate new target location:
Wherein g (x)=- k ' (x),Wherein δ (x) is Kronecker function,It indicates
U-th of component in t moment in intelligence weight vector,Wherein xiFor in candidate region
Pixel point coordinates, nhFor the total pixel number in candidate region, ChFor normalized parameter, k (x) is kernel function, and h is candidate region bandwidth,
If being unsatisfactory forIt willWhereinIt indicates the Euclidean distance between two coordinate points, repeats
S400, until meet end condition, ε herein0To preset judgment threshold.
6. the dimension self-adaption visual target tracking method according to claim 5 based on middle intelligence similarity measure, feature
It is, in step S500, for reducing, expanding corresponding true, false, the Uncertainty measured value of dimension calculation, according to cosine analog quantity
It surveys and determines scale more new strategy, include the following steps:
The case where scale reduces is considered first, based on the analysis to authentic communication and uncertain information, reduces the corresponding T of scalesa、
Isa、FsaMeasurement is respectively defined as:
WhereinThe feature histogram in respective cell domain after being reduced for present frame target area, the zonule is with current goal
Region is reference, and regional center is constant, and length and width by reducing λ in proportionsaTimes, λsaFor corresponding zoom factor,For initial target
Histogram,The feature histogram of background area is corresponded to for current area domain;
It is measured according to cosine similarity, it is as follows that the ideally corresponding middle intelligence similarity weights of diminution scale can be obtained:
The case where considering scale amplification later is obtained amplification scale and is corresponded to based on the analysis to authentic communication and uncertain information
Middle intelligence similarity weight wbscal, wherein amplification coefficient is λba, take λba=λsa, the corresponding T of amplification scaleba、Iba、FbaIt measures and divides
It is not defined as:
Wherein,The feature histogram in big region is corresponded to after amplifying for present frame target area, the big region is with current goal
Region is reference, and regional center is constant, and length and width by amplifying λ in proportionbaTimes,For initial target histogram,It is current big
Region corresponds to the feature histogram of background area;
It is measured according to cosine similarity, it is as follows that the ideally corresponding middle intelligence similarity weights of amplification scale can be obtained:
Obtain wsscalAnd wbscalAfterwards, present frame target scale is as follows:
Wherein, s is preset scale factor, and meets s>1;
Determine λnewAfterwards, target following frame is scaled into λnewAgain as new tracking box.
7. the dimension self-adaption visual target tracking method according to claim 6 based on middle intelligence similarity measure, feature
It is, in step S600, updates the method for target background feature histogram, include the following steps:
Calculate present frame target background feature histogramIf Indicating willIt is assigned a value ofWherein ρ is Pasteur's related coefficient, The background characteristics that frame moment algorithm uses thus
Histogram, ε1To preset judgment threshold.
8. the dimension self-adaption visual target tracking method according to claim 1 based on middle intelligence similarity measure, feature
It is, further includes following steps after step S600:
S700:Next video frame is chosen, step S200~S600 is then executed, until being disposed all video frame.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810114328.4A CN108492313A (en) | 2018-02-05 | 2018-02-05 | A kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810114328.4A CN108492313A (en) | 2018-02-05 | 2018-02-05 | A kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108492313A true CN108492313A (en) | 2018-09-04 |
Family
ID=63344307
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810114328.4A Withdrawn CN108492313A (en) | 2018-02-05 | 2018-02-05 | A kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108492313A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109711445A (en) * | 2018-12-18 | 2019-05-03 | 绍兴文理学院 | The similar method of weighting of intelligence in the super-pixel of target following classifier on-line training sample |
CN109741345A (en) * | 2018-12-29 | 2019-05-10 | 绍兴文理学院 | Strengthen the middle intelligence partitioning parameters automatically selecting method of specific region class objective attribute target attribute |
CN110059413A (en) * | 2019-04-19 | 2019-07-26 | 中国航空无线电电子研究所 | A kind of method for diagnosing faults |
CN110378195A (en) * | 2019-05-27 | 2019-10-25 | 浙江工业大学 | Multi-object tracking method based on histogram caching method |
CN110458045A (en) * | 2019-07-22 | 2019-11-15 | 浙江大华技术股份有限公司 | Acquisition methods, image processing method and the device of response probability histogram |
CN110826472A (en) * | 2019-11-01 | 2020-02-21 | 新疆大学 | Image detection method and device |
CN112508962A (en) * | 2020-09-27 | 2021-03-16 | 绍兴文理学院 | Target image region subsequence separation method based on time correlation image sequence |
CN113344963A (en) * | 2021-05-27 | 2021-09-03 | 绍兴市北大信息技术科创中心 | Seed point self-adaptive target tracking system based on image segmentation |
CN113379806A (en) * | 2021-08-13 | 2021-09-10 | 南昌工程学院 | Target tracking method and system based on learnable sparse conversion attention mechanism |
-
2018
- 2018-02-05 CN CN201810114328.4A patent/CN108492313A/en not_active Withdrawn
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109711445A (en) * | 2018-12-18 | 2019-05-03 | 绍兴文理学院 | The similar method of weighting of intelligence in the super-pixel of target following classifier on-line training sample |
CN109711445B (en) * | 2018-12-18 | 2020-10-16 | 绍兴文理学院 | Super-pixel medium-intelligence similarity weighting method for target tracking classifier on-line training sample |
CN109741345B (en) * | 2018-12-29 | 2020-09-15 | 绍兴文理学院 | Automatic selection method of intelligent segmentation parameters for strengthening target attributes of specific area classes |
CN109741345A (en) * | 2018-12-29 | 2019-05-10 | 绍兴文理学院 | Strengthen the middle intelligence partitioning parameters automatically selecting method of specific region class objective attribute target attribute |
CN110059413A (en) * | 2019-04-19 | 2019-07-26 | 中国航空无线电电子研究所 | A kind of method for diagnosing faults |
CN110059413B (en) * | 2019-04-19 | 2022-11-15 | 中国航空无线电电子研究所 | Fault diagnosis method |
CN110378195A (en) * | 2019-05-27 | 2019-10-25 | 浙江工业大学 | Multi-object tracking method based on histogram caching method |
CN110378195B (en) * | 2019-05-27 | 2021-10-15 | 浙江工业大学 | Multi-target tracking method based on histogram cache method |
CN110458045A (en) * | 2019-07-22 | 2019-11-15 | 浙江大华技术股份有限公司 | Acquisition methods, image processing method and the device of response probability histogram |
CN110826472A (en) * | 2019-11-01 | 2020-02-21 | 新疆大学 | Image detection method and device |
CN110826472B (en) * | 2019-11-01 | 2023-06-27 | 新疆大学 | Image detection method and device |
CN112508962A (en) * | 2020-09-27 | 2021-03-16 | 绍兴文理学院 | Target image region subsequence separation method based on time correlation image sequence |
CN113344963A (en) * | 2021-05-27 | 2021-09-03 | 绍兴市北大信息技术科创中心 | Seed point self-adaptive target tracking system based on image segmentation |
CN113379806A (en) * | 2021-08-13 | 2021-09-10 | 南昌工程学院 | Target tracking method and system based on learnable sparse conversion attention mechanism |
CN113379806B (en) * | 2021-08-13 | 2021-11-09 | 南昌工程学院 | Target tracking method and system based on learnable sparse conversion attention mechanism |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108492313A (en) | A kind of dimension self-adaption visual target tracking method based on middle intelligence similarity measure | |
US9824294B2 (en) | Saliency information acquisition device and saliency information acquisition method | |
CN110097093A (en) | A kind of heterologous accurate matching of image method | |
CN111369597B (en) | Particle filter target tracking method based on multi-feature fusion | |
CN111882586B (en) | Multi-actor target tracking method oriented to theater environment | |
CN113361542B (en) | Local feature extraction method based on deep learning | |
CN113095263B (en) | Training method and device for pedestrian re-recognition model under shielding and pedestrian re-recognition method and device under shielding | |
Yuan et al. | Learning to count buildings in diverse aerial scenes | |
CN104881029A (en) | Mobile robot navigation method based on one point RANSAC and FAST algorithm | |
CN101872415A (en) | Video copying detection method being suitable for IPTV | |
CN101630407A (en) | Method for positioning forged region based on two view geometry and image division | |
CN111241987B (en) | Multi-target model visual tracking method based on cost-sensitive three-branch decision | |
CN105205807A (en) | Remote sensing image change detection method based on sparse automatic code machine | |
CN109344758B (en) | Face recognition method based on improved local binary pattern | |
CN111160292B (en) | Human eye detection method | |
JP7243372B2 (en) | Object tracking device and object tracking method | |
US20230386023A1 (en) | Method for detecting medical images, electronic device, and storage medium | |
Li et al. | Remote-Sensing-Based Change Detection Using Change Vector Analysis in Posterior Probability Space: A Context-Sensitive Bayesian Network Approach | |
CN106021610A (en) | Video fingerprint extracting method based on salient region | |
CN108241837B (en) | Method and device for detecting remnants | |
CN112200831B (en) | Dynamic template-based dense connection twin neural network target tracking method | |
JP4285644B2 (en) | Object identification method, apparatus and program | |
CN113095384A (en) | Remote sensing image matching method based on context characteristics of straight line segments | |
CN111860174A (en) | Method for detecting shielding face by fusing RepGT-RepBox function | |
CN109711445A (en) | The similar method of weighting of intelligence in the super-pixel of target following classifier on-line training sample |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180904 |