CN107644199A - A kind of feature based and the rigid-object tracking of Regional Synergetic matching - Google Patents
A kind of feature based and the rigid-object tracking of Regional Synergetic matching Download PDFInfo
- Publication number
- CN107644199A CN107644199A CN201710730447.8A CN201710730447A CN107644199A CN 107644199 A CN107644199 A CN 107644199A CN 201710730447 A CN201710730447 A CN 201710730447A CN 107644199 A CN107644199 A CN 107644199A
- Authority
- CN
- China
- Prior art keywords
- matching
- surf
- region
- consistency
- feature
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to the rigid-object tracking that a kind of feature based and Regional Synergetic match.This method comprises the following steps:1) the selected target region in initial pictures, and detect SURF features in target area;2) in target area, consistency region is built centered on each SURF characteristic points;3) when present image arrives, extract its SURF feature, and carry out with initial pictures cooperateing with matching based on SURF features and consistency region, form matching double points;4) kinematic parameter is calculated according to obtained matching double points, so that it is determined that the target area of present image, realizes target following.The present invention is by studying repeatable rule of the SURF features under complicated change, utilize SURF features and the Scheme Solving kinematic parameter of region template collaboration matching, accurately description and matching can be realized to the local feature of target area, and then ensure robustness, the stability of target following effect.
Description
Technical field
The invention belongs to technical field of computer vision, and in particular to a kind of feature based and the rigid body of Regional Synergetic matching
Method for tracking target.
Background technology
The motion at rigid-object surface any point can represent overall motion so that utilize the spy in target area
Sign is possibly realized to describe target motion.Existing rigid-object tracking, which is directed to extracting in reference picture target area, to be had
There are some features of consistency, and the feature of extraction is quantified and described, such as color characteristic, textural characteristics, Optical-flow Feature.
The part that local feature refers to detect in image-region has consistency, reproducibility and specific feature, Neng Gou
To a certain extent resistance block, yardstick, the complicated change such as rotation, and offer is to the quantitative description of feature.At present, it is special compared to other
Sign, local feature advantage in terms of consistency and specificity is more obvious, it more deep is applied in target following.
When present frame arrives, local feature is extracted to whole region first and is described.And then found ibid by the matching of local feature
The candidate of local feature is corresponding in one frame same target collects.By stochastical sampling consistency algorithm (RANSAC), remove incorrect
Character pair collection, estimate motion transform parameter, realize target following.Fig. 1 gives the tracking block diagram of feature based,
Its main thought is to regard tracking as local feature matching problem.
At present, SURF (Speed-up Robust Feature, accelerate robust features) be characterized in using more and effect compared with
For one of preferable local feature, integral image fast algorithm is introduced primarily into, and height is obtained by performing signed magnitude arithmetic(al) approximation
The response of this second-order differential.SURF algorithm mainly includes feature detection and feature describes two aspects.Feature detection passes through quick
The yardstick and principal direction of each feature are calculated, and draws a circle to approve the constant symmetric neighborhood of dimension rotation centered on test point;Feature
Description carries out Haar feature calculations in the consistency neighborhood, and ultimately forms 64 dimensional feature vectors.Between different images
SURF characteristic matchings are mainly to be realized by the distance between comparative feature vector.
Motion model structure is completed by SURF characteristic matchings.Assuming that x andRepresent respectively between different images
Corresponding SURF characteristic points, then have therebetween following relation:
Wherein, W (x, h) is perspective transformation function, h=(h1,...h8)TIt is kinematic parameter.It is specific to represent as follows:
Wherein x, y represent horizontal stroke, the ordinate of SURF characteristic points.After drawing kinematic parameter, by the target area side of initial frame
Boundary carries out corresponding perspective transform, obtains the target area of present frame.
In video, scene often occur illumination, block, visual angle, one or more changes such as affine, to local feature
Matching cause serious interference.Prior art continue to use with still image identical local feature matching process, can not adapt to
The scene of acute variation occurs, also without the embodiment adaptivity corresponding with the change of scene continuity.
The content of the invention
In video sequence, scene often occurs complicated change, such as yardstick, rotation, illumination, blocks, to rigid-object
Tracking proposes challenge, and the present invention provides a kind of feature based and the rigid-object tracking of Regional Synergetic matching, can be right
The local feature of target area realizes accurately description and matching, and then ensures robustness, the stability of target following effect.
The technical solution adopted by the present invention is as follows:
A kind of feature based and the rigid-object tracking of Regional Synergetic matching, it is characterised in that comprise the following steps:
1) the selected target region in initial pictures, and detect SURF features in target area;
2) in target area, consistency region is built centered on each SURF characteristic points;
3) when present image arrives, extract its SURF feature, and carried out with initial pictures based on SURF features and constant
Property region collaboration matching, formed matching double points;
4) kinematic parameter is calculated according to obtained matching double points, so that it is determined that the target area of present image, is realized
Target following.
Further, it is additionally included in line renewal step:Online updating is carried out to SURF features and consistency region, to improve
The adaptivity subsequently tracked.
Further, during step 1) detection SURF features, Hessian matrix determinants are calculated using integral image, then lead to
Selection extreme value is crossed to position SURF characteristic points, and metric space is established by adjusting the size of grid wave filter.
Further, step 3) the collaboration matching based on SURF features and consistency region, is first with SURF
Characteristic matching obtains kinematic parameter initial value, is then matched to obtain precise motion ginseng using the consistency region built
Number, so as to form matching double points.
Further, step 3) finds candidate's corresponding points first with characteristic vector, then describes son according to characteristic vector
Candidate's corresponding points are compared by the matching fraction of matching to being ranked up, then by matching fraction with pre-determined threshold, will be greater than door
The point of limit forms the corresponding points pair of feature based vector description matching, and obtain initial frame using RANSAC algorithms to selecting
I1With t frames ItTarget area between kinematic parameter initial value;If there is no the point pair more than thresholding, then abandon current
Match point, then carry out complementary matching using the consistency region built.
Further, kinematic parameter is calculated using RANSAC algorithms in step 4), realizes the positioning to target.
Further, the method for the online updating is:If correct matching double points are matched by feature based
Arrive, by the use of matching double points as positive sample, update SURF features and consistency region;If correct matching double points are to pass through
What Region Matching obtained, then update consistency region;For the matching double points of mistake, any renewal is not done.
A kind of server, the server include memory and processor, and the memory storage computer program is described
Computer program is configured as by the computing device, and the computer program includes being used to perform in method described above respectively
The instruction of step.
A kind of computer-readable recording medium for storing computer program, when the computer program is computer-executed,
The step of realizing method described above.
The key point of the present invention includes:1) rigid-object tracking problem is solved based on local feature matching;2) in initial frame
Motion model is built to rigid-object between present frame;3) repeatability of the lower SURF features of complicated change;4) build constant
Property region;5) feature based and the collaboration matching scheme in region;6) online updating makes tracking keep adaptivity, ensure that algorithm
Systematicness and completeness.
The present invention proposes a kind of adaptive rigid-object tracking scheme towards complex scene change in video, by right
Repeatable rule of the SURF features under complicated change is studied, and the side of matching is cooperateed with using SURF features and region template
Case solves kinematic parameter, and further renewal is done to SURF features, region template after the completion of tracking, can keep changing complexity
Adaptivity, lift the robustness and stability of tracking effect.
Brief description of the drawings
The tracking block diagram of Fig. 1 feature baseds in the prior art.
The feature based of Fig. 2 present invention and the rigid-object tracking flow chart of Regional Synergetic matching.
The repeatable schematic diagram of the lower SURF features of the complicated changes of Fig. 3.
Fig. 4 consistency area schematics.
Fig. 5 target following schematic diagrames.
Embodiment
Below by specific embodiments and the drawings, the present invention is described in further details.
The present invention proposes a kind of new rigid-object tracking scheme, and is become according to the target between initial frame and present frame
Change and establish contact, fundamentally overcome drifting problem.Because the internal structure of rigid body has Movement consistency, the program utilizes
SURF characteristic matchings determine target area, obtain preliminary motion parameter;It is re-introduced into the matching scheme reply based on region more
Violent object variations, ensure that being best able to the SURF features of adaptation current goal change and its surrounding specific region is fully adopted
With realization becomes more meticulous matching.After completing to the tracking of present frame, SURF features and its surrounding specific region are carried out online more
Newly, the adaptivity subsequently tracked is improved.Tracking is kept the adaptivity to complexity change, reach robust and stabilization
Effect.
The workflow of the present invention is as shown in Figure 2.In initial pictures, selected target region (can be irised out interested manually
Target area), and in target area extract SURF features establish feature description, while each using SURF characteristic points as
Structure region description in the consistency neighborhood at center.When new image arrives, SURF features are extracted first, utilize feature
With obtaining preliminary kinematic parameter;That establishes feature based and region with initial pictures again cooperates with matching, forms final match point
It is right.Kinematic parameter is calculated by RANSAC, realizes the positioning to target.Finally, SURF features and consistency region are carried out
Online updating, it is easy to the processing of subsequent frame.
Although SURF features keep the adaptivity to complexity change to a certain extent, it is not meant to all
The consistency that SURF features can be consistent.When complicated change occurs, a part of SURF features can not be again detected,
The work such as follow-up feature description and matching can not more be carried out.By taking Fig. 3 as an example, a part of feature repeats for illumination variation:
Some other features are more sensitive for affine change.Reason is that each SURF features are described with a circle.
If left semicircle and right semi-circle are with distinct contrast, after illumination changes, the contrast of this left and right is still present, and is now characterized in
It is reproducible:If the grey scale change around the center of circle is more sharp, the grey scale change of outer peripheral areas is flat, when the affine change of generation
Afterwards, the grey scale change of interior exterior domain is still kept, and feature can detect again.
Specific implementation is described as follows:
Step 1:SURF feature extractions
SURF feature extractions calculate Hessian matrix determinants using integral image, then are positioned by choosing extreme value.Tool
Body, to point x=(x, y) place on image I, yardstick s Hessian matrix Hs (x, s) are expressed as:
With LxxExemplified by (x, s), Gaussian function second dervative is represented in x=(x, y) place and figure I convolution, specific use side
Lattice wave filter (box filter) DxxCome approximate.By introducing associated weight w, the balance to Hessian matrix determinants is realized:
det(Happrox)=DxxDyy-(wDxy)2 (4)
Wherein, det (Happrox) it is two-dimensional matrix determinant, Dyy-And DxyThe second order Gauss difference of different directions is represented respectively
Template.
For SURF feature detections, original image size need not be changed by establishing metric space, but be filtered by adjusting grid
The size of ripple device is realized, convolutional calculation is being carried out with original image.The approximate representation of grid wave filter and integral image are combined
Computational efficiency is lifted, calculates filter template size normalization det (Happrox)。
The layer (octave) formed by different size grid wave filters is exactly the expression to metric space.Point of interest is determined
Position is that non-maxima suppression plan is performed in the image centered on candidate point and 3 × 3 × 3 neighborhoods including metric space
Slightly, using the corresponding points with maximum or minimum value as characteristic point, while yardstick s is obtained.
Step 2:Consistency region is built
In initial frame, consistency region P is built for each SURF characteristic points γ in target areaγ, definition is with γ
For the border circular areas that the center of circle, 2.5s are radius, s is the yardstick of SURF features, as shown in Figure 4.Wherein 2.5s is in the present embodiment
Value, other values can also be used as radius.
Step 3:The collaboration matching in feature based and region
Assuming that initial frame target area SURF feature point sets are combined into B={ b1,b2,...,bR, character pair vector description
{U1,U2,...,UR}.When t frames arrive, SURF feature detections are carried out first, obtain set of characteristic points Υt={ γt,1,
γt,2,...,γt,Q, the sub- V of character pair vector descriptiont={ Vt,1,Vt,2,...,Vt,Q}.Then characteristic vector V is utilizedt,rLook for
To feature brCandidate's corresponding points ψt,r, describe the sub reliability matched according to characteristic vector afterwards and (match fraction, refer to two
Euclidean distance between characteristic vector) by candidate's corresponding points to being ranked up, then will matching fraction UrVt,r(ψt,r) with pre- gating
Limit λ (the minimum believable matching fraction set) is compared, and the point that will be greater than thresholding to form subclass Ψ to electingt=
{ψt,1,ψt,2,...,ψt,M, the corresponding points pair of feature based vector description matching are formed, are obtained initially using RANSAC algorithms
Frame I1With t frames ItTarget area between kinematic parameter initial valueIn the present embodiment, the scope for matching fraction is 0-1,
The value of pre-determined threshold takes 0.5.
Otherwise (i.e. be not present more than thresholding point to), then abandon current matching point ψt,r, then it is constant using having built
Property region carry out complementary matching.The purpose matched using consistency region is still to solve for initial frame I1With t frames ItTarget
Precise motion parameter h between regiont,1, and then the positioning that becomes more meticulous to current goal is realized, relation therebetween is as follows:
I1(x)≈It(W(x,ht,1)) (6)
Wherein, W is perspective transformation function.
Specifically, characteristic matching is carried out to the target area of present frame and initial frame, statistics uses consistency Region Matching
Point Φ={ φ1,φ2,...,φN, by the consistency region P for minimizing each point φφGray scale difference quadratic sum ask
Solution:
This programme come solution formula (7), passes through the h being calculated using inverse composition methodt,1, further find Φ=
{φ1,φ2,...,φNIn the corresponding point set K of present framet={ κt,1,κt,2,...,κt,N}。ht,1During iterative
Initial value is the kinematic parameter of former frame
During RANSAC, initial frame target area SURF set of characteristic points B={ b1,b2,...,bRCorresponding point set
Close by Ψt={ ψt,1,...,ψt,MAnd Kt={ κt,1,κt,2,...,κt,NComposition, and R=M+N.
Step 4:Target following
Target following schematic diagram is as shown in figure 5, circle represents the corresponding SURF features established by SURF characteristic matchings
Point;Square frame represents the corresponding consistency region established by consistency region.Initial frame I1With t frames ItTarget area it
Between final kinematic parameter ht,1Based on above-mentioned corresponding points to being calculated, the target area of present frame is finally determined.
Step 5:Online updating
, it is necessary to be updated to SURF features, consistency region, more new technological process is as shown in table 1 after completion target following.Such as
The correct matching double points of fruit match to obtain by feature based, by the use of matching double points as positive sample, renewal SURF features,
Consistency region;If correct matching double points are obtained by Region Matching, renewal consistency region;For mistake
Matching double points, any renewal is not done.
The online updating flow of table 1
The above embodiments are merely illustrative of the technical solutions of the present invention rather than is limited, the ordinary skill of this area
Technical scheme can be modified by personnel or equivalent substitution, without departing from the spirit and scope of the present invention, this
The protection domain of invention should be to be defined described in claims.
Claims (9)
1. a kind of feature based and the rigid-object tracking of Regional Synergetic matching, it is characterised in that comprise the following steps:
1) the selected target region in initial pictures, and detect SURF features in target area;
2) in target area, consistency region is built centered on each SURF characteristic points;
3) when present image arrives, its SURF feature is extracted, and carries out with initial pictures being based on SURF features and consistency area
The collaboration matching in domain, forms matching double points;
4) kinematic parameter is calculated according to obtained matching double points, so that it is determined that the target area of present image, realizes target
Tracking.
2. the method as described in claim 1, it is characterised in that be additionally included in line renewal step:To SURF features and consistency
Region carries out online updating, to improve the adaptivity subsequently tracked.
3. method as claimed in claim 1 or 2, it is characterised in that during step 1) detection SURF features, utilize integral image meter
Hessian matrix determinants are calculated, then SURF characteristic points are positioned by choosing extreme value, and by adjusting the size of grid wave filter
To establish metric space.
4. method as claimed in claim 1 or 2, it is characterised in that step 3) is described to be based on SURF features and consistency region
Collaboration matching, be to obtain kinematic parameter initial value first with SURF characteristic matchings, then utilize the consistency area built
Domain is matched to obtain precise motion parameter, so as to form matching double points.
5. method as claimed in claim 4, it is characterised in that step 3) finds candidate's corresponding points first with characteristic vector,
Then it is according to the matching fraction of characteristic vector description son matching that candidate's corresponding points are same default to being ranked up, then by matching fraction
Thresholding is compared, and be will be greater than the point of thresholding to selecting, is formed the corresponding points pair of feature based vector description matching, and utilize
RANSAC algorithms obtain initial frame I1With t frames ItTarget area between kinematic parameter initial value;If there is no more than door
The point pair of limit, then current matching point is abandoned, then complementary matching is carried out using the consistency region built.
6. method as claimed in claim 1 or 2, it is characterised in that motion ginseng is calculated using RANSAC algorithms in step 4)
Number, realizes the positioning to target.
7. method as claimed in claim 2, it is characterised in that the method for the online updating is:If correct match point
To matching to obtain by feature based, by the use of matching double points as positive sample, SURF features and consistency region are updated;Such as
The correct matching double points of fruit are obtained by Region Matching, then update consistency region;For the matching double points of mistake, do not do
Any renewal.
8. a kind of server, it is characterised in that the server includes memory and processor, the memory storage computer
Program, the computer program are configured as by the computing device, and the computer program includes will for perform claim
Ask the instruction of each step in any claim methods described in 1 to 7.
9. a kind of computer-readable recording medium for storing computer program, it is characterised in that the computer program is calculated
When machine performs, the step of realizing any claim methods described in claim 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710730447.8A CN107644199A (en) | 2017-08-23 | 2017-08-23 | A kind of feature based and the rigid-object tracking of Regional Synergetic matching |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710730447.8A CN107644199A (en) | 2017-08-23 | 2017-08-23 | A kind of feature based and the rigid-object tracking of Regional Synergetic matching |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107644199A true CN107644199A (en) | 2018-01-30 |
Family
ID=61110261
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710730447.8A Pending CN107644199A (en) | 2017-08-23 | 2017-08-23 | A kind of feature based and the rigid-object tracking of Regional Synergetic matching |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107644199A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108830885A (en) * | 2018-05-31 | 2018-11-16 | 北京空间飞行器总体设计部 | One kind being based on the relevant detection false alarm rejection method of multidirections difference residual energy |
CN110516528A (en) * | 2019-07-08 | 2019-11-29 | 杭州电子科技大学 | A kind of moving-target detection and tracking method based under movement background |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778831A (en) * | 2016-11-28 | 2017-05-31 | 国家计算机网络与信息安全管理中心 | The online tagsort of rigid-object and tracking based on gauss hybrid models |
CN106897721A (en) * | 2017-01-23 | 2017-06-27 | 国家计算机网络与信息安全管理中心 | The rigid-object tracking that a kind of local feature is combined with bag of words |
CN106934395A (en) * | 2017-02-15 | 2017-07-07 | 国家计算机网络与信息安全管理中心 | The rigid-object tracking that a kind of use SURF features and color characteristic are blended |
-
2017
- 2017-08-23 CN CN201710730447.8A patent/CN107644199A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778831A (en) * | 2016-11-28 | 2017-05-31 | 国家计算机网络与信息安全管理中心 | The online tagsort of rigid-object and tracking based on gauss hybrid models |
CN106897721A (en) * | 2017-01-23 | 2017-06-27 | 国家计算机网络与信息安全管理中心 | The rigid-object tracking that a kind of local feature is combined with bag of words |
CN106934395A (en) * | 2017-02-15 | 2017-07-07 | 国家计算机网络与信息安全管理中心 | The rigid-object tracking that a kind of use SURF features and color characteristic are blended |
Non-Patent Citations (1)
Title |
---|
权巍等: "SURF与RANSAC算法结合的图像跟踪方法", 《计算机仿真》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108830885A (en) * | 2018-05-31 | 2018-11-16 | 北京空间飞行器总体设计部 | One kind being based on the relevant detection false alarm rejection method of multidirections difference residual energy |
CN108830885B (en) * | 2018-05-31 | 2021-12-07 | 北京空间飞行器总体设计部 | Detection false alarm suppression method based on multi-directional differential residual energy correlation |
CN110516528A (en) * | 2019-07-08 | 2019-11-29 | 杭州电子科技大学 | A kind of moving-target detection and tracking method based under movement background |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Gu et al. | Model-based referenceless quality metric of 3D synthesized images using local image description | |
CN103310453B (en) | A kind of fast image registration method based on subimage Corner Feature | |
CN103325112B (en) | Moving target method for quick in dynamic scene | |
CN109684925B (en) | Depth image-based human face living body detection method and device | |
CN105844630B (en) | A kind of image super-resolution fusion denoising method of binocular vision | |
CN107945111B (en) | Image stitching method based on SURF (speeded up robust features) feature extraction and CS-LBP (local binary Pattern) descriptor | |
CN104134200B (en) | Mobile scene image splicing method based on improved weighted fusion | |
CN104715487B (en) | A kind of sub-pixel edge detection method based on Zernike pseudo-matrix | |
Lo et al. | Joint trilateral filtering for depth map super-resolution | |
CN107516316A (en) | It is a kind of that the method that focus mechanism is split to static human image is introduced in FCN | |
CN109242959B (en) | Three-dimensional scene reconstruction method and system | |
CN107506795A (en) | A kind of local gray level histogram feature towards images match describes sub- method for building up and image matching method | |
CN105335952B (en) | Matching power flow computational methods and device and parallax value calculating method and equipment | |
CN106023187A (en) | Image registration method based on SIFT feature and angle relative distance | |
CN110298344A (en) | A kind of positioning of instrument knob and detection method based on machine vision | |
CN115908415B (en) | Edge-based defect detection method, device, equipment and storage medium | |
CN107452028A (en) | A kind of method and device for determining target image positional information | |
CN103020898A (en) | Sequence iris image super-resolution reconstruction method | |
CN107194310A (en) | The rigid-object tracking matched based on scene change classifications and online local feature | |
CN111310508A (en) | Two-dimensional code identification method | |
Kim et al. | Sredgenet: Edge enhanced single image super resolution using dense edge detection network and feature merge network | |
CN115471682A (en) | Image matching method based on SIFT fusion ResNet50 | |
Liu et al. | Asflow: Unsupervised optical flow learning with adaptive pyramid sampling | |
CN117557565B (en) | Detection method and device for lithium battery pole piece | |
CN107644199A (en) | A kind of feature based and the rigid-object tracking of Regional Synergetic matching |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180130 |