CN105894538A - Target tracking method and target tracking device - Google Patents
Target tracking method and target tracking device Download PDFInfo
- Publication number
- CN105894538A CN105894538A CN201610204077.XA CN201610204077A CN105894538A CN 105894538 A CN105894538 A CN 105894538A CN 201610204077 A CN201610204077 A CN 201610204077A CN 105894538 A CN105894538 A CN 105894538A
- Authority
- CN
- China
- Prior art keywords
- tracking box
- tracking
- frame video
- video image
- image
- 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
Links
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
-
- 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
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Abstract
The invention discloses a target tracking method and a target tracking device, relates to the technical field of visual tracking, and aims to solve the problem that target tracking is of low accuracy. The target tracking method comprises the following steps: getting multiple tracking boxes of different sizes according to the tracking box size of a previous frame of video image and preset tracking box coefficients; respectively extracting the image features in the tracking boxes of different sizes; substituting the image features in the tracking boxes into a correlation filter of the previous frame of video image to get the comprehensive response values of the image features; and getting the maximum comprehensive response value, and determining the tracking box size corresponding to the maximum comprehensive response value as the tracking box size of the current frame of video image. The target tracking method and the target tracking device provided by the invention are used for tracking a target in a video image.
Description
Technical field
The present invention relates to Visual Tracking field, particularly relate to a kind of method for tracking target and device.
Background technology
Target following is fusion image process, pattern recognition, an artificial intelligence and the multiple different technologies such as automatically controls
Comprehensive application technology, be widely used in various field.Target following refer to the moving target in image sequence or
The object that feature is single carries out detecting, identify and following the tracks of, by obtaining parameter or the shapes of target such as the position of target, speed
With the feature such as color, to its further process, thus realize the accurate tracking to moving target thing.
In the research field of target following, numerous scholars propose substantial amounts of tracking, in these trackings,
Tracking box is often used to be tracked following the tracks of target, but during tracking target is tracked by tracking box, due to
Following the tracks of target is motion, and following the tracks of target size and location in video image can change over time, works as tracking
Target size in video image less or bigger time, tracking box of the prior art is easily lost the tracking mesh followed the tracks of
Mark, thus reduce the accuracy of target following.
Summary of the invention
It is an object of the invention to provide a kind of method for tracking target and device, for adapt to target dimensional variation and/
Or rotationally-varying, improve the accuracy of target following.
To achieve these goals, the present invention provides following technical scheme:
On the one hand, the invention provides a kind of method for tracking target, including:
Tracking box size according to previous frame video image and default tracking box coefficient, obtain multiple various sizes of
Tracking box;
Extract the characteristics of image in each different size tracking box described respectively;
Characteristics of image in described tracking box is substituted into the correlation filter of previous frame video image, obtains described image special
The comprehensive response value levied;
Obtain the maximum in described comprehensive response value, and by tracking box size corresponding for maximum described comprehensive response value
It is defined as the tracking box size of current frame video image.
On the other hand, the invention provides a kind of target tracker, including:
Acquisition module, for the tracking box size according to previous frame video image and default tracking box coefficient, obtains
Multiple various sizes of tracking box;
Characteristic extracting module, for extracting the characteristics of image in each various sizes of tracking box respectively;
Correlation filter, for receiving the characteristics of image in tracking box, obtains the comprehensive response value of described characteristics of image;
Tracking box determines module, for obtaining the maximum in described comprehensive response value, and by maximum described comprehensive sound
The tracking box that should be worth correspondence is sized to the tracking box size of current frame video image.
In method for tracking target that the present invention provides and device, according to the tracking box size of previous frame video image with preset
Tracking box coefficient, obtain multiple various sizes of tracking box, obtain characteristics of image in each various sizes of tracking box
Comprehensive response value, is sized to the tracking box chi of current frame video image by tracking box corresponding for maximum comprehensive response value
Very little.When following the tracks of target travel, follow the tracks of the dimensional variation factor of target, the rotationally-varying factor etc. and change, so that following the tracks of
Target size in video image changes, and follows the tracks of target with the tracking box utilizing fixed size, it is impossible to adapt to follow the tracks of mesh
Target dimensional variation and/or rotationally-varying prior art are compared, and the present invention can set and the tracking of previous frame video image
The multiple various sizes of tracking box that frame size is relevant, and choose the maximum tracking box size of comprehensive response value as current video
The tracking box size of picture frame, so that the tracking box following the tracks of target is sized to adapt to the tracking mesh that size changes
Mark, say, that the target following carried out in video image can adapt to follow the tracks of the dimensional variation of target and/or rotationally-varying,
It is accurately tracked by the tracking target that size and location changes, improves the accuracy of target following.
Accompanying drawing explanation
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the present invention, this
Bright schematic description and description is used for explaining the present invention, is not intended that inappropriate limitation of the present invention.In the accompanying drawings:
Fig. 1 is the flow chart one of a kind of method for tracking target in the embodiment of the present invention;
Fig. 2 is the method flow diagram two of a kind of target following in the embodiment of the present invention;
Fig. 3 is the method flow diagram three of a kind of target following in the embodiment of the present invention;
Fig. 4 is the method flow diagram four of a kind of target following in the embodiment of the present invention;
Fig. 5 a is the first frame video image and the schematic diagram of tracking box therein in the embodiment of the present invention;
Fig. 5 b is the second frame video image and the schematic diagram of tracking box therein in the embodiment of the present invention;
Fig. 5 c is the 3rd frame video image and the schematic diagram one of tracking box therein in the embodiment of the present invention;
Fig. 5 d is the 3rd frame video image and the schematic diagram two of tracking box therein in the embodiment of the present invention;
Fig. 6 is a kind of method for tracking target flow chart five in the embodiment of the present invention;
Fig. 7 is a kind of method for tracking target flow chart six in the embodiment of the present invention;
Fig. 8 is the structural representation one of a kind of target tracker in the embodiment of the present invention;
Fig. 9 is the structural representation two of a kind of target tracker in the embodiment of the present invention;
Figure 10 is the structural representation three of a kind of target tracker in the embodiment of the present invention;
Figure 11 is the structural representation four of a kind of target tracker in the embodiment of the present invention;
Figure 12 is the structural representation five of a kind of target tracker in the embodiment of the present invention;
Figure 13 is the structural representation six of a kind of target tracker in the embodiment of the present invention.
Detailed description of the invention
In order to further illustrate method for tracking target and the device that the embodiment of the present invention provides, below in conjunction with Figure of description
It is described in detail.
At the method for tracking target that the embodiment of the present invention provides, it is applied to target tracker, example, it is applied to intelligence
Target person tracking in the target vehicle tracking of field of traffic, community or the target following etc. in video record.The present invention
Executive agent include but not limited to, be arranged on smart mobile phone, desktop computer, intelligent television, notebook computer, panel computer etc.
Target tracking module in terminal unit or other independent target trackers.
Referring to Fig. 1, the method for tracking target that the embodiment of the present invention provides includes:
Step 101, obtains a frame video image.Example, obtain the one of the photographic head shooting being carrying out shooting work
Frame video image, or, obtain a certain frame video image etc. in pending one section video.It should be noted that a frame regards
Frequently the tracking box size of image is the tracking box size following the tracks of target in a frame video image.If it should be noted that this frame
Video image is the initial frame video image of pending video image, then jump procedure 108 performs, and otherwise, jump procedure 102 is held
OK.
Step 102, according to tracking box size and the default tracking box coefficient of previous frame video image, obtain multiple not
Tracking box with size.Wherein, the tracking box coefficient preset has multiple, owing to following the tracks of target in two adjacent frame video images
The change of size and location is the least, hence with tracking box size and the default tracking box coefficient of previous frame video image,
For determining that the tracking box size of current frame video image provides the various sizes of tracking box of multiple candidates.Such as, preset with
Track frame coefficient is respectively s1, s2 and s3, then corresponding for tracking box coefficient s1 tracking box a size of k1, and coefficient s2 is corresponding for tracking box
Tracking box a size of k2, tracking box corresponding for tracking box coefficient s3 a size of k3.
Step 103, extracts the characteristics of image in each various sizes of tracking box respectively.Wherein it is possible to utilize different
Extracting method extracts the different types of characteristics of image in tracking box, such as, it is possible to use HOG (Histogram of
Oriented Gradient, histograms of oriented gradients) extracting method obtains the HOG figure in the tracking box of tracking box a size of k1
As feature, and/or LBP (Local Binary Pattern, local binary patterns) extracting method is utilized to obtain tracking box size
For the LBP characteristics of image in the tracking box of k1, the kind of the characteristics of image extracted includes, but are not limited to above two mode.
Step 104, substitutes into the correlation filter of previous frame video image by the characteristics of image in tracking box, obtains image special
The comprehensive response value levied.If it should be noted that the characteristics of image only one in the tracking box extracted, then comprehensive response value is
This kind of characteristics of image substitutes into the response value that the correlation filter of previous frame video image obtains;If the figure in the tracking box extracted
As feature has multiple, then comprehensive response value is that multiple characteristics of image substitutes in the correlation filter of previous frame video image and obtains
The sum of response value, response value is specifically as follows recurrence response value.It should be noted that the characteristics of image extracted in tracking box
Kind is the most, and in follow-up object tracking process, the tracking to following the tracks of target is the most accurate, but the characteristics of image simultaneously extracted
Kind the most, the amount of calculation in object tracking process is the biggest, accordingly, it would be desirable to according to concrete operative scenario and expectation
Effect determines the number of the kind of the characteristics of image extracted.
Step 105, obtains the maximum in comprehensive response value, and by tracking box size corresponding for maximum comprehensive response value
It is defined as the tracking box size of current frame video image.Such as, the tracking box coefficient preset is respectively s1, s2 and s3, then follow the tracks of
Tracking box corresponding for frame coefficient s1 a size of k1, tracking box corresponding for tracking box coefficient s2 a size of k2, tracking box coefficient s3 pair
The tracking box answered a size of k3, if the comprehensive response value that in the tracking box of tracking box a size of k3, characteristics of image is corresponding is maximum, then
Using k3 as the tracking box size of current frame video image.
It should be noted that in order to the tracking box size making present frame video and graphic is the most accurate, can obtain for the first time
After maximum comprehensive response value, utilize tracking box size corresponding to maximum comprehensive response value and tracking box coefficient, again
Obtain multiple various sizes of tracking box, extract the characteristics of image in newly obtained tracking box, and characteristics of image is brought into relevant
Wave filter, calculates the comprehensive response value of newly obtained characteristics of image, and second time obtains the comprehensive response value of maximum, and will second time
Tracking box size corresponding to the maximum comprehensive response value that obtains is as the tracking box size of current frame video image.Certainly, pin
Same frame video image can also be asked for more times the process of comprehensive response value of maximum, but ask for maximum owing to carrying out
The number of times of comprehensive response value the most, amount of calculation is the biggest, and the real-time of target following is the poorest, therefore, it can according to target with
The requirement of the application scenarios of track method, arranges the number of times of the comprehensive response value asking for maximum.
In the method for tracking target that the present invention provides, according to tracking box size and the default tracking of previous frame video image
Frame coefficient, obtains multiple various sizes of tracking box, obtains the comprehensive of the characteristics of image in each various sizes of tracking box and rings
Should be worth, tracking box corresponding for maximum comprehensive response value is sized to the tracking box size of current frame video image.When with
During track target travel, follow the tracks of the characteristics such as the dimensional variation factor of target, the rotationally-varying factor and change, so that following the tracks of target
Size in video image changes, and follows the tracks of target with the tracking box utilizing fixed size, it is impossible to adapt to follow the tracks of target
Dimensional variation and/or rotationally-varying prior art are compared, and the present invention can set the tracking box chi with previous frame video image
Very little relevant multiple various sizes of tracking box, and choose the maximum tracking box size of comprehensive response value as current video image
The tracking box size of frame, so that the tracking box following the tracks of target is sized to adapt to the tracking target that size changes, also
That is the target following carried out in video image can adapt to follow the tracks of the dimensional variation of target and/or rotationally-varying, accurately
Trace into the tracking target that size and location changes, improve the accuracy of target following.
Refer to Fig. 2, step 106 can also be included before step 102, choose suitable tracking box coefficient, specifically
Content is as follows:
Step 106, chooses the numerical range of tracking box coefficient, chooses multiple tracking box system in numerical range dispersedly
Number.Wherein, the numerical range chosen homodisperse can be chosen multiple tracking box coefficient, it is also possible to the choosing of random dispersion
Take multiple tracking box coefficient.Such as, the numerical range chosen is 0.9~1.1, can homodisperse choose in 0.9~1.1
5 tracking box coefficients, respectively 0.9,0.95,1,1.05 and 1.1;Can also in 0.9~1.1 random dispersion choose 5
Tracking box coefficient, respectively 0.92,0.98,1,1.05,1.08.
Refer to Fig. 3, step 107 can also be included before step 101, choose suitable tracking box coefficient, specifically
Content is as follows:
Step 107, chooses the central value of tracking box coefficient, by central value and chooses default respectively to both sides from central value
The numerical value of number is as tracking box coefficient.Wherein, a central value is first set so that the tracking box coefficient chosen is around central value
Choose, concrete, when choosing other tracking box coefficient from central value respectively to both sides, can choose uniformly, it is possible to
Choose with random.Such as, set central value as 1, uniformly choose two tracking box coefficients from central value is each to both sides, be respectively
0.96,0.98,1.02 and 1.04, say, that the tracking box coefficient chosen is 0.96,0.98,1,1.02 and 1.04;Or,
Two tracking box coefficients are respectively randomly selected to both sides from central value, respectively 0.907,0.952,1.05 and 1.103, namely
Saying, the tracking box coefficient chosen is 0.907,0.952,1,1.05 and 1.103.
It should be noted that the tracking box size of previous frame video image and multiple different tracking box multiplication,
To multiple various sizes of tracking box.Such as, the step 101 in above-described embodiment is specifically as follows: by previous frame video image
Tracking box size respectively with central value and the tracking box multiplication of predetermined number chosen respectively to both sides from central value,
Obtain multiple various sizes of tracking box.Such as: the tracking box coefficient chosen is respectively 0.9,0.95,1,1.05 and 1.1,1 and is
Central value, the tracking box of previous frame video image a size of k1, then the size of the multiple various sizes of tracking box obtained is respectively
For 0.9k1,0.95k1, k1,1.05k1 and 1.1k1.Step 101 in above-described embodiment can also be specifically: is regarded by previous frame
Frequently the tracking box size of image respectively with the multiple tracking box multiplication chosen in numerical range, obtain multiple different size
Tracking box.
Referring to Fig. 4, before step 101, method for tracking target can also include step 108, chooses the first frame and regards
Frequently the tracking target in image, particular content is as follows:
Step 108, receives user and instructs the Object selection of the first frame video image, and instruct according to Object selection, choosing
Surely target is followed the tracks of.Wherein, the first frame video image i.e. initial frame video image, in the first frame video image, need to use
Family selects the tracking target in this section of video, and user can send Object selection by the mode such as button, touch screen and instruct, mesh
Mark follows the tracks of the Object selection instruction of device reception user, selected target of following the tracks of, and after selected tracking target, can obtain following the tracks of target
Tracking box in the first frame video image, tracking box be slightly larger in dimension than tracking target, it is possible to will follow the tracks of aiming circle follow the tracks of
In frame.
Illustrating below in conjunction with concrete video image, referring to Fig. 5 a-Fig. 5 c, Fig. 5 a is the first frame of one section of video
Video image, the vehicle in Fig. 5 a is for following the tracks of target, and tracking box K1 is the tracking box in the first frame video image, and tracking box K1
Size be k1;Fig. 5 b is the second frame video image of this section of video, and wherein, the tracking box coefficient preset is respectively 0.95,1 and
1.05, according to tracking box size k1 and three default tracking box coefficients of tracking box K1 of the first frame video image, permissible
Obtaining three tracking box, respectively tracking box K2, tracking box K3 and tracking box K4, the tracking box of tracking box K2 is a size of
0.95k1, the tracking box of tracking box K3 a size of k1, the tracking box of tracking box K4 a size of 1.05k1, by three tracking box K2,
Characteristics of image in K3 and K4 substitutes into the correlation filter in the first frame video image, obtains three comprehensive response values, maximum
What comprehensive response value was corresponding is tracking box K2, then using tracking box size 0.95k1 as the tracking box chi of the second frame video image
Very little;Fig. 5 c is the 3rd frame video image of this section of video, and wherein, the tracking box coefficient preset still is respectively 0.95,1 and
1.05, according to tracking box size 0.95k1 and three default tracking box coefficients of the second frame video image, three can be obtained
Individual tracking box, respectively tracking box K5, tracking box K6 and tracking box K7, the tracking box of tracking box K5 a size of 0.95 × 0.95k1
=0.9025k1, the tracking box of tracking box K6 a size of 0.95k1, the tracking box of tracking box K7 a size of 1.05 × 0.95k1=
0.9975k1, substitutes into the correlation filter in the second frame video image by the characteristics of image in three tracking box K5, K6 and K7,
To three comprehensive response values, what maximum comprehensive response value was corresponding is tracking box K5, then using tracking box size 0.9025k1 as
The tracking box size of the 3rd frame video image.
Need exist for explanation is, it is also possible to according to the variation tendency of the tracking box size in former frame video images, adjust
Whole default tracking box coefficient, this tracking box coefficient be respectively used to adjust subsequent video images tracking box size, such as, according to
Video image shown in Fig. 5 a and Fig. 5 b, it can be determined that follow the tracks of target size in this section of video and taper into, therefore, can
To judge that tracking box size is also being gradually reduced, then refer to the 3rd frame video image that Fig. 5 d, Fig. 5 d is this section of video, its
In, the tracking box coefficient preset is respectively set as 0.9,0.95 and 1, according to tracking box size 0.95k1 of the second frame video image
And three default tracking box coefficients, three tracking box, respectively tracking box K8, tracking box K9 and tracking box can be obtained
K10, the tracking box of tracking box K8 a size of 0.9 × 0.95k1=0.855k1, the tracking box of tracking box K9 a size of 0.95 ×
0.95k1=0.9025k1, the tracking box of tracking box K10 a size of 0.95k1, by the image in three tracking box K8, K9 and K10
Feature substitutes into the correlation filter in the second frame video image, obtains three comprehensive response values, and maximum comprehensive response value is corresponding
Be tracking box K8, then using tracking box size 0.855k1 as the tracking box size of the 3rd frame video image.
Referring to Fig. 6, the step 104 in above-described embodiment specifically can be refined as step 1041, and particular content is as follows:
Step 1041, substitutes into the characteristics of image in tracking box the correlation filter of previous frame video image, utilizes formulaIt is calculated comprehensive response valueWherein,For the correlation filter of previous frame video image, x is
Characteristics of image in tracking box,Fourier transformation for x.In the case of the characteristics of image in tracking box only has one, should
Characteristics of image brings formula intoIn the response value the most comprehensive response response value that obtains.Figure in the tracking box extracted
When having multiple as the kind of feature, each characteristics of image is to there being a correlation filter, by the most of the same race in same tracking box
The characteristics of image of class substitutes into the correlation filter of correspondence respectively, respectively obtains the response value that different types of characteristics of image is corresponding,
Corresponding for characteristics of image different types of in same tracking box response value is added, obtains characteristics of image in same tracking box
Comprehensive response value.Such as, the HOG characteristics of image in the tracking box of tracking box a size of k1 and LBP characteristics of image have been extracted, then
The correlation filter A of HOG characteristics of image substitution previous frame video image obtains response value f1, in the substitution of LBP characteristics of image
The correlation filter B of one frame video image obtains response value f2, by the HOG image in the tracking box of tracking box a size of k1
Response value f1 that feature obtains, response value f2 obtained with LBP characteristics of image is added, and the response value sum obtained is tracking box
The comprehensive response value of the characteristics of image in the tracking box of a size of k1.
It should be noted that formula can be utilizedObtain maximum comprehensive response value, maximum is combined
Close response valueCorresponding tracking box size is as the tracking box size of current frame video image, wherein, max table
Show maximizing, F-1Represent Fourier inversion,Represent that the i-th tracking box coefficient in multiple tracking box coefficient is corresponding
Characteristics of image in tracking box.
Refer to Fig. 7, it is also possible to the correlation filter of current frame video image is updated, in the embodiment of the present invention also
Can include step 109-step 110, particular content is as follows:
Step 109, utilizes the characteristics of image that maximum comprehensive response value is corresponding, the filtering of training current frame video image
Device.The wave filter of current frame video image is for updating the correlation filter of present frame video and graphic.
Step 110, utilizes the filtering of current frame video image after the correlation filter of previous frame video image and training
Device, the correlation filter of more newly obtained current frame video image.Next frame picture frame is carried out repeat during target following above-mentioned
The method of target following, but correlation filter be update after the correlation filter of current frame video image.Concrete, according to public affairs
FormulaThe correlation filter of more newly obtained current frame video imageWherein,Regard for previous frame
Frequently the correlation filter used in image,For the wave filter of the current frame video image after training, θ is the first balance parameter, 0
≤ θ≤1, it should be noted that general 1-θ > θ, such as, θ can be with value for 0.012.
It should be noted that the wave filter of the current frame video image in the embodiment of the present invention isIts
In,For the wave filter of current frame video image, x is the characteristics of image in tracking box,For the Fourier transformation of x,For x's
Conjugation after Fourier transformation,Returning label for gaussian-shape, λ is the second balance parameter, 0≤λ≤1.It should be noted that ⊙
For dot product symbol, represent the element multiplication of correspondence position among two matrixes, thus simplify in the tracking goal approach of the present invention
Calculating in Fourier.
With a concrete data instance, above-mentioned calculating process will be described below: the image that the comprehensive response value of maximum is corresponding is special
Levyλ=0.0002, substitutes into formula
Training obtains the wave filter of current frame video imageUtilize the wave filter of current frame video imageWith previous frame video image
Correlation filterObtain the correlation filter of current frame video imageIn the tracking determining next frame video image
During frame size, the tracking box coefficient preset is respectively 0.95,1 and 1.05, then three the various sizes of tracking obtained
Characteristics of image in frame is respectively Characteristics of image in various sizes of for above three tracking box is substituted into more newly obtained
Current frame video image wave filter in, and obtain the comprehensive response value of characteristics of image in the tracking box of each size respectively
ForThe comprehensive response value that can obtain maximum isThen willCorresponding tracking box size is as the tracking box chi of next frame video image
Very little.Certainly, being merely illustrative of, the concrete numerical value not representing the embodiment of the present invention is confined to this herein.
Refer to Fig. 8, the embodiment of the present invention additionally provides a kind of target tracker 20, including:
Acquisition module 21, for the tracking box size according to previous frame video image and default tracking box coefficient,
To multiple various sizes of tracking box;
Characteristic extracting module 22, for extracting the characteristics of image in each various sizes of tracking box respectively;
Correlation filter 23, for receiving the characteristics of image in tracking box, obtains the comprehensive response value of characteristics of image;
Tracking box determines module 24, for obtaining the maximum in comprehensive response value, and by maximum comprehensive response value pair
The tracking box answered is sized to the tracking box size of current frame video image.
In the target tracker 20 that the present invention provides, acquisition module 21 is according to the tracking box size of previous frame video image
With default tracking box coefficient, obtain multiple various sizes of tracking box, by characteristic extracting module 22 and correlation filter 23
Obtaining the comprehensive response value of characteristics of image in each various sizes of tracking box, tracking box determines that module 24 is by maximum comprehensive
Tracking box corresponding to response value is sized to the tracking box size of current frame video image.When following the tracks of target travel, follow the tracks of
The characteristics such as the dimensional variation factor of target, the rotationally-varying factor change, so that following the tracks of big in video image of target
Little change, follow the tracks of target with the tracking box utilizing fixed size, it is impossible to adapt to follow the tracks of dimensional variation and/or the rotation of target
The prior art of change is compared, and the present invention can set the multiple different chis that the tracking box size from previous frame video image is relevant
Very little tracking box, and choose the comprehensive response value maximum tracking box size tracking box size as current video image frame, from
And make the tracking box following the tracks of target be sized to adapt to the tracking target that size changes, say, that in video image
The target following carried out can adapt to follow the tracks of the dimensional variation of target and/or rotationally-varying, is accurately tracked by size and location
The tracking target changed, improves the accuracy of target following.
Refer to Fig. 9, when the kind of the characteristics of image in the tracking box extracted is equal to or more than two, the tracking box of extraction
In the kind of characteristics of image is equal with the number of correlation filter 23 and one_to_one corresponding.
Referring to Figure 10, target tracker 20 also includes the first coefficient setting module 25, the first coefficient setting module 25
For choosing the numerical range of tracking box coefficient, numerical range is chosen multiple tracking box coefficient dispersedly.
Or, referring to Figure 11, target tracker 20 also includes that the second coefficient setting module 26, the second coefficient set mould
Block 26, for Selection Center value, chooses the numerical value of predetermined number as tracking box using central value and from central value to both sides respectively
Coefficient.
Further, acquisition module 21 specifically for by the tracking box size of previous frame video image respectively with central value with
And the tracking box multiplication of the predetermined number chosen respectively to both sides from central value, obtain multiple various sizes of tracking box,
Central value is 1.
Referring to Figure 12, target tracker 20 also includes target chosen module 27, is used for receiving user to the first frame figure
The Object selection instruction of picture, and instruct according to Object selection, selected tracking target.
Concrete, correlation filter 23 is additionally operable to receive the characteristics of image in tracking box, utilizes formula
It is calculated comprehensive response valueWherein,For the correlation filter of previous frame video image, x is the figure in tracking box
As feature,Fourier transformation for x.
Referring to Figure 13, described target tracker 20 also includes training module 28 and more new module 29.
Training module 28, is used for the characteristics of image utilizing maximum comprehensive response value corresponding, trains current frame video image
Wave filter.Concrete, more new module is specifically for according to formulaMore newly obtained present frame video
The correlation filter of imageWherein,For the correlation filter of previous frame video image,Regard for the present frame after training
Frequently the wave filter of image, θ is the first balance parameter, 0≤θ≤1.The wave filter of current frame video image is
Wherein,For the wave filter of current frame video image, x is the characteristics of image in tracking box,For the Fourier transformation of x,For x
Fourier transformation after conjugation,Returning label for gaussian-shape, λ is the second balance parameter, 0≤λ≤1.
More new module 29, the current frame video image after the correlation filter utilizing previous frame video image and training
Wave filter, the correlation filter of more newly obtained current frame video image.
Each embodiment in this specification all uses the mode gone forward one by one to describe, identical similar portion between each embodiment
Dividing and see mutually, what each embodiment stressed is the difference with other embodiments.For target following
For the embodiment of device, owing to it is substantially similar to the embodiment of method of target following, so describing fairly simple, phase
The part of the embodiment seeing the method for target following in place of pass illustrates.
In the description of above-mentioned embodiment, specific features, structure, material or feature can be at any one or many
Individual embodiment or example combine in an appropriate manner.
The above, the only detailed description of the invention of the present invention, but protection scope of the present invention is not limited thereto, and any
Those familiar with the art, in the technical scope that the invention discloses, can readily occur in change or replace, should contain
Cover within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with described scope of the claims.
Claims (15)
1. a method for tracking target, it is characterised in that including:
Tracking box size according to previous frame video image and default tracking box coefficient, obtain multiple various sizes of tracking
Frame;
Extract the characteristics of image in each different size tracking box described respectively;
Characteristics of image in described tracking box is substituted into the correlation filter of previous frame video image, obtains described characteristics of image
Comprehensive response value;
Obtain the maximum in described comprehensive response value, and tracking box size corresponding for maximum described comprehensive response value is determined
Tracking box size for current frame video image.
Method for tracking target the most according to claim 1, it is characterised in that in the tracking box according to previous frame video image
Size and default tracking box coefficient, before obtaining multiple various sizes of tracking box, also include:
Choose the numerical range of described tracking box coefficient, described numerical range is chosen multiple described tracking box system dispersedly
Number.
Method for tracking target the most according to claim 1, it is characterised in that in the tracking box according to previous frame video image
Size and default tracking box coefficient, before obtaining multiple various sizes of tracking box, also include:
Selection Center value, chooses the numerical value of predetermined number as institute using described central value and from described central value respectively to both sides
State tracking box coefficient.
Method for tracking target the most according to claim 3, it is characterised in that described central value is 1;According to previous frame video
The tracking box size of image and default tracking box coefficient, obtain multiple various sizes of tracking box, including:
The tracking box size of previous frame video image is selected to both sides respectively with described central value and from described central value respectively
The tracking box multiplication of the predetermined number taken, obtains multiple various sizes of tracking box.
Method for tracking target the most according to claim 1, it is characterised in that also include:
Receive user the Object selection of the first frame video image is instructed, and instruct according to described Object selection, selected described with
Track target.
Method for tracking target the most according to claim 1, it is characterised in that the characteristics of image in tracking box is substituted into upper one
The correlation filter of frame video image, obtains the comprehensive response value of described characteristics of image, including:
Characteristics of image in described tracking box is substituted into the correlation filter of previous frame video image, utilizes formulaIt is calculated comprehensive response valueWherein,For the correlation filter of previous frame video image, x is
Characteristics of image in described tracking box,Fourier transformation for x.
Method for tracking target the most according to claim 1, it is characterised in that maximum described comprehensive response value is corresponding
Tracking box be sized to current frame video image tracking box size after, also include:
Utilize the characteristics of image that maximum described comprehensive response value is corresponding, the wave filter of training current frame video image;
Utilize the wave filter of current frame video image after the correlation filter of previous frame video image and training, more newly obtained work as
The correlation filter of front frame video image.
Method for tracking target the most according to claim 7, it is characterised in that utilize the correlation filtering of previous frame video image
The wave filter of the current frame video image after device and training, the correlation filter of more newly obtained current frame video image, including:
According to formulaThe correlation filter of more newly obtained current frame video imageWherein,
For the correlation filter of previous frame video image,For the wave filter of the current frame video image after training, θ is the first balance ginseng
Amount, 0≤θ≤1.
Method for tracking target the most according to claim 8, it is characterised in that the wave filter of described current frame video image isWherein,For the wave filter of described current frame video image, x is the characteristics of image in described tracking box,
For the Fourier transformation of x,For the conjugation after the Fourier transformation of x,Returning label for gaussian-shape, λ is the second balance parameter, 0
≤λ≤1。
10. a target tracker, it is characterised in that including:
Acquisition module, for the tracking box size according to previous frame video image and default tracking box coefficient, obtains multiple
Various sizes of tracking box;
Characteristic extracting module, for extracting the characteristics of image in each various sizes of tracking box respectively;
Correlation filter, for receiving the characteristics of image in tracking box, obtains the comprehensive response value of described characteristics of image;
Tracking box determines module, for obtaining the maximum in described comprehensive response value, and by maximum described comprehensive response value
Corresponding tracking box is sized to the tracking box size of current frame video image.
11. devices according to claim 10, it is characterised in that the device of described target following also includes that the first coefficient sets
Module, described first coefficient setting module, for choosing the numerical range of described tracking box coefficient, divides in described numerical range
Choose multiple described tracking box coefficient scatteredly.
12. devices according to claim 10, it is characterised in that the device of described target following also includes that the second coefficient sets
Module, described second coefficient setting module is used for Selection Center value, divides to both sides by described central value and from described central value
Do not choose the numerical value of predetermined number as described tracking box coefficient.
13. devices according to claim 10, it is characterised in that described correlation filter is additionally operable to receive in described tracking box
Characteristics of image, utilize formulaIt is calculated comprehensive response valueWherein,For previous frame video
The correlation filter of image, x is the characteristics of image in described tracking box,Fourier transformation for x.
14. devices according to claim 10, it is characterised in that described target tracker also includes training module and renewal
Module;
Described training module, is used for the characteristics of image utilizing maximum described comprehensive response value corresponding, trains present frame video figure
The wave filter of picture;
Described more new module, the current frame video image after the correlation filter utilizing previous frame video image and training
Wave filter, the correlation filter of more newly obtained current frame video image.
15. according to device described in claim 14, it is characterised in that described more new module is specifically for according to formulaThe correlation filter of more newly obtained current frame video imageWherein,For previous frame video
The correlation filter of image,For the wave filter of the current frame video image after training, θ is the first balance parameter, 0≤θ≤1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610204077.XA CN105894538A (en) | 2016-04-01 | 2016-04-01 | Target tracking method and target tracking device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610204077.XA CN105894538A (en) | 2016-04-01 | 2016-04-01 | Target tracking method and target tracking device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105894538A true CN105894538A (en) | 2016-08-24 |
Family
ID=57013325
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610204077.XA Pending CN105894538A (en) | 2016-04-01 | 2016-04-01 | Target tracking method and target tracking device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105894538A (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106530327A (en) * | 2016-11-07 | 2017-03-22 | 北京航空航天大学 | Quick real-time discrimination type tracing method based on multi-local-feature learning |
CN107644217A (en) * | 2017-09-29 | 2018-01-30 | 中国科学技术大学 | Method for tracking target based on convolutional neural networks and correlation filter |
CN107943077A (en) * | 2017-11-24 | 2018-04-20 | 歌尔股份有限公司 | A kind of method for tracing, device and the unmanned plane of unmanned plane drop target |
CN108053425A (en) * | 2017-12-25 | 2018-05-18 | 北京航空航天大学 | A kind of high speed correlation filtering method for tracking target based on multi-channel feature |
CN108154522A (en) * | 2016-12-05 | 2018-06-12 | 北京深鉴科技有限公司 | Target tracking system |
CN108171719A (en) * | 2017-12-25 | 2018-06-15 | 北京奇虎科技有限公司 | Video penetration management method and device based on the segmentation of adaptive tracing frame |
CN108229456A (en) * | 2017-11-22 | 2018-06-29 | 深圳市商汤科技有限公司 | Method for tracking target and device, electronic equipment, computer storage media |
CN108280808A (en) * | 2017-12-15 | 2018-07-13 | 西安电子科技大学 | The method for tracking target of correlation filter is exported based on structuring |
CN108475072A (en) * | 2017-04-28 | 2018-08-31 | 深圳市大疆创新科技有限公司 | A kind of tracking and controlling method, device and aircraft |
CN108711165A (en) * | 2018-05-21 | 2018-10-26 | 西安电子科技大学 | Filter training method and method for tracking target |
CN108983956A (en) * | 2017-11-30 | 2018-12-11 | 成都通甲优博科技有限责任公司 | Body feeling interaction method and device |
CN109712167A (en) * | 2017-10-25 | 2019-05-03 | 财团法人资讯工业策进会 | Target detecting and method for tracing and system |
CN112132864A (en) * | 2020-09-21 | 2020-12-25 | 大连遨游智能科技有限公司 | Robot following method based on vision and following robot |
WO2021057309A1 (en) * | 2019-09-27 | 2021-04-01 | 深圳云天励飞技术股份有限公司 | Tracked target determination method and related device |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104574445A (en) * | 2015-01-23 | 2015-04-29 | 北京航空航天大学 | Target tracking method and device |
-
2016
- 2016-04-01 CN CN201610204077.XA patent/CN105894538A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104574445A (en) * | 2015-01-23 | 2015-04-29 | 北京航空航天大学 | Target tracking method and device |
Non-Patent Citations (5)
Title |
---|
YANG LI 等: "A Scale Adaptive Kernel Correlation Filter Tracker with Feature Integration", 《13TH EUROPEAN CONFERENCE ON COMPUTER VISION》 * |
张三喜: "《弹道特征参数摄像测量》", 31 March 2014 * |
张雷 等: "基于相关滤波器的视觉目标跟踪方法", 《光电子 激光》 * |
朱虹: "《数字图像技术与应用》", 31 May 2011 * |
高志清: "《AutoCAD建筑设计培训教程》", 31 October 2004 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106530327A (en) * | 2016-11-07 | 2017-03-22 | 北京航空航天大学 | Quick real-time discrimination type tracing method based on multi-local-feature learning |
CN108154522A (en) * | 2016-12-05 | 2018-06-12 | 北京深鉴科技有限公司 | Target tracking system |
CN108154522B (en) * | 2016-12-05 | 2022-07-12 | 赛灵思电子科技(北京)有限公司 | Target tracking system |
US11587355B2 (en) | 2017-04-28 | 2023-02-21 | SZ DJI Technology Co., Ltd. | Tracking control method, device, and aircraft |
CN108475072A (en) * | 2017-04-28 | 2018-08-31 | 深圳市大疆创新科技有限公司 | A kind of tracking and controlling method, device and aircraft |
CN107644217A (en) * | 2017-09-29 | 2018-01-30 | 中国科学技术大学 | Method for tracking target based on convolutional neural networks and correlation filter |
CN107644217B (en) * | 2017-09-29 | 2020-06-26 | 中国科学技术大学 | Target tracking method based on convolutional neural network and related filter |
CN109712167A (en) * | 2017-10-25 | 2019-05-03 | 财团法人资讯工业策进会 | Target detecting and method for tracing and system |
CN108229456A (en) * | 2017-11-22 | 2018-06-29 | 深圳市商汤科技有限公司 | Method for tracking target and device, electronic equipment, computer storage media |
CN107943077A (en) * | 2017-11-24 | 2018-04-20 | 歌尔股份有限公司 | A kind of method for tracing, device and the unmanned plane of unmanned plane drop target |
CN108983956B (en) * | 2017-11-30 | 2021-07-06 | 成都通甲优博科技有限责任公司 | Somatosensory interaction method and device |
CN108983956A (en) * | 2017-11-30 | 2018-12-11 | 成都通甲优博科技有限责任公司 | Body feeling interaction method and device |
CN108280808A (en) * | 2017-12-15 | 2018-07-13 | 西安电子科技大学 | The method for tracking target of correlation filter is exported based on structuring |
CN108171719B (en) * | 2017-12-25 | 2021-07-23 | 北京奇虎科技有限公司 | Video crossing processing method and device based on self-adaptive tracking frame segmentation |
CN108171719A (en) * | 2017-12-25 | 2018-06-15 | 北京奇虎科技有限公司 | Video penetration management method and device based on the segmentation of adaptive tracing frame |
CN108053425A (en) * | 2017-12-25 | 2018-05-18 | 北京航空航天大学 | A kind of high speed correlation filtering method for tracking target based on multi-channel feature |
CN108711165A (en) * | 2018-05-21 | 2018-10-26 | 西安电子科技大学 | Filter training method and method for tracking target |
WO2021057309A1 (en) * | 2019-09-27 | 2021-04-01 | 深圳云天励飞技术股份有限公司 | Tracked target determination method and related device |
CN112132864A (en) * | 2020-09-21 | 2020-12-25 | 大连遨游智能科技有限公司 | Robot following method based on vision and following robot |
CN112132864B (en) * | 2020-09-21 | 2024-04-09 | 大连遨游智能科技有限公司 | Vision-based robot following method and following robot |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105894538A (en) | Target tracking method and target tracking device | |
CN108986140B (en) | Target scale self-adaptive tracking method based on correlation filtering and color detection | |
CN108734723B (en) | Relevant filtering target tracking method based on adaptive weight joint learning | |
CN107292318B (en) | Image significance object detection method based on center dark channel prior information | |
CN109598234B (en) | Key point detection method and device | |
CN103260081B (en) | A kind of video image zooming processing method and processing device | |
CN109343700B (en) | Eye movement control calibration data acquisition method and device | |
CN110765882B (en) | Video tag determination method, device, server and storage medium | |
CN106097385B (en) | A kind of method and apparatus of target following | |
CN110345610B (en) | Control method and device of air conditioner and air conditioning equipment | |
CN106097383A (en) | A kind of method for tracking target for occlusion issue and equipment | |
Li et al. | End-to-end feature integration for correlation filter tracking with channel attention | |
CN110472577A (en) | Video tracing method when a kind of long based on adaptive correlation filtering | |
CN108805902A (en) | A kind of space-time contextual target tracking of adaptive scale | |
CN114495169A (en) | Training data processing method, device and equipment for human body posture recognition | |
CN106934339A (en) | A kind of target following, the extracting method of tracking target distinguishing feature and device | |
CN113538509B (en) | Visual tracking method and device based on adaptive correlation filtering feature fusion learning | |
CN105701840A (en) | System for real-time tracking of multiple objects in video and implementation method | |
CN113269808B (en) | Video small target tracking method and device | |
Zelinsky | Specifying the components of attention in a visual search task | |
CN112099330B (en) | Holographic human body reconstruction method based on external camera and wearable display control equipment | |
CN106033550B (en) | Method for tracking target and device | |
WO2020237674A1 (en) | Target tracking method and apparatus, and unmanned aerial vehicle | |
CN112966546A (en) | Embedded attitude estimation method based on unmanned aerial vehicle scout image | |
CN111027589A (en) | Multi-division target detection algorithm evaluation system and method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into 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: 20160824 |