CN107368785A - The video target tracking method of multinuclear local restriction - Google Patents
The video target tracking method of multinuclear local restriction Download PDFInfo
- Publication number
- CN107368785A CN107368785A CN201710455426.XA CN201710455426A CN107368785A CN 107368785 A CN107368785 A CN 107368785A CN 201710455426 A CN201710455426 A CN 201710455426A CN 107368785 A CN107368785 A CN 107368785A
- Authority
- CN
- China
- Prior art keywords
- sample
- target
- local restriction
- tracking
- multinuclear
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of video target tracking method of multinuclear local restriction, including：Using local restriction linear coding method, the partial structurtes of sample data are incorporated into collaboration method for expressing, represented with obtaining the sample with good classification performance；The nuclear space for representing to expand to multiple features fusion will be cooperateed with using kernel function so that dictionary and rarefaction representation coefficient are strengthened the class discriminating power of target signature；Target following is regarded as two classification problems, using the candidate target of grader highest scoring as tracking target under particle filter framework.Accurate and robust tracking can be carried out to video object using this method.
Description
Technical field
The present invention relates to computer vision field technical field, more particularly to a kind of video object of multinuclear local restriction with
Track method.
Background technology
Visual target tracking is one important research contents of computer vision field, in vision guided navigation, man-machine interaction, intelligence
The fields such as energy traffic, video monitoring are widely applied, and are various follow-up advanced processes, such as target identification, behavioural analysis, are regarded
Frequency image compression encoding and application understand the basis of contour level Video processing and application.Hidden yet with existing in tracking video
The factors such as gear, illumination variation, dimensional variation, mutation, angle change, this causes the video frequency object tracking of accurate robust to turn into one
Very important work.
With the development of compressive sensing theory and sparse coding theory so that rarefaction representation has been applied to video object
Tracking in, its core is the rarefaction representation problem being considered as target under particle filter framework.In rarefaction representation tracking, l1With
Track method has stronger robustness, but because of its demand solution l1Norm minimum problem to solve relatively difficult and time-consuming.In
It is, based on l2The collaboration rarefaction representation of norm is suggested and is applied in target following, although l2Collaboration under norm represents do not have
l1Reconstruction coefficients under norm it is openness strong, but do not have to because mapping matrix can be calculated in advance to update each particle, so as to carry
High operation efficiency；But collaboration method for expressing is substantially a kind of linear method, when nonlinear change (strong light occurs for target
According to change, motion, background are acutely shaken suddenly) when, tracking will be caused to fail.
The content of the invention
The present invention is directed to the deficiencies in the prior art, there is provided a kind of video frequency object tracking side of multinuclear local restriction
Method, scheme raising rendering speed, the quality rendered by EWA enhancings are drawn by designing single pass.
The technical scheme adopted by the invention for realizing the object of the invention is：
A kind of video target tracking method of multinuclear local restriction, comprises the following steps：
(1) partial structurtes of sample data are incorporated into collaboration method for expressing, build the local restriction line of sample characteristics
Property coding；(2) kernel method is utilized, builds the multinuclear local restriction collaboration coding of sample characteristics；(3) support vector machines are based on,
The grader score of sample is embedded into the tracking that video object is realized under particle filter framework；(4) according to target and background sample
This change, dynamic update To Template and background template and grader.SVMs is Support Vector
Machine,SVM。
Preferably, in step (1), represented to obtain the sample with good classification performance, by sample data
Partial structurtes are incorporated into collaboration expression, to build the local restriction of sample characteristics collaboration coding, are specifically：
(a) under framework of sparse representation, test sampleIt is expressed as y=d1x1+d2x2+…+dnxn=Dx, wherein word
Allusion quotationDictionary atom is represented, it cooperates with the mesh represented, and it is cooperateed with
The object function of expression isThe collaboration that sample is obtained by minimizing is expressed asOptimal solutionOnly it is y linear projection, and P=(DTD+λI)-1DTIndependently of y, square is so projected
Battle array P can be previously calculated, and avoid l1Each test sample is both needed to single optimization processing under norm, substantially increases
Arithmetic speed.
(b) it is substantially a kind of linear method to cooperate with method for expressing, uses non local dictionary atom reformulation candidate target
Coefficient, but the partial structurtes of data often carry more information, such as a sample and the sample around it than global structure
There should be similar coding, thus it is more accurate using the sample reconstruction of local restriction；The partial structurtes of sample data are drawn
Enter to collaboration and represent, its object function isWhereinIt is one
The diagonal matrix of local restriction, the local restriction collaboration that sample is obtained by minimizing are encoded to
Preferably, in step (2), based on kernel method, the multinuclear local restriction collaboration coding of sample characteristics is built, is had
Body is：
(a) nonlinear data is mapped to by High-dimensional Linear nuclear space, images of the sample y in higher dimensional space using kernel method
By Φ=[φ (d1),φ(d2),…,φ(dn)] carrying out linear expression, this increases computation complexity apparently, it is therefore necessary to
Reduce the dimension of feature space.Using accidental projection matrix PTHigh dimensional data is mapped to lower dimensional space, local restriction nuclear coordination
The object function of coding is
Seek it local derviation and can be obtained equal to 0
Wherein [x1,x2,…,xn]TIt is that n maintains number vector x, note
Then the solution of formula (1) can be written as
According to the property of kernel function, P=Φ B are remembered, substitute into formula (2) and formula (3) and obtain respectively
Wherein K=ΦTΦ is core Gram matrixes (positive semi-definite symmetrical matrix).It can pass through core in feature space sample inner product
Function calculates, and for any two sample φ (x), φ (y), there is φ (x)Tφ (y)=(φ (x) φ (y))=k (x, y),
Then
(b) in order to stably track target, target is described using multiple features, the mode for then passing through multi-core integration obtains
Merging kernel function isSo as to obtain the multinuclear local restriction of sample characteristics collaboration coding
The present invention represents target using spatial color histogram and spatial gradient histogram；For each target sample
This, is divided into the subregion of four area equations, and then extracting subregion color property respectively and merging turns into target coloured silk
Color characteristic hc.For Gradient Features, first by core [- 0.5,0,0.5] and [- 0.5,0,0.5]TProcessing is filtered, obtains figure
As gradient；Then same method is taken to obtain four sub-regions histograms of oriented gradients, and it is straight by merging four sub-regions
Square figure obtains spatial gradient histogram hg.If Kc、KgThe respectively nuclear matrix of color property and Gradient Features, Kc、KgIn each member
Element represents the similitude between two histograms respectively；KcIn corresponding element according to Kc(i, j)=BhaCoffTo calculate,
WhereinWithIt is that two color histograms, BhaCoff () are Bhattacharyya coefficient functions；Kg、Kc(·,y)、Kg
The computational methods and K of (, y)cIt is identical.
Preferably, in step (3), based on support vector machines, the grader score of sample is embedded into particle filter
The tracking of video object is realized under ripple framework, is specifically：
(a) a number of target sample and background sample are extracted according to Gaussian Profile around the first frame target area,
Target sample, background sample are also respectively positive sample, negative sample, calculate multinuclear office of the positive and negative samples under current dictionary respectively
Portion's constraint collaboration coding zi, by the coding of positive and negative samples and its corresponding label
Substitute into support vector machines to be trained, by minimizing cost functionCome
Study strategies and methods, the score of grader are calculated as
(b) it will classify and be embedded into the tracking that target is realized under particle filter framework；Particle filter is that Bayes's sequence is important
Sampling techniques, it is frequently used for estimating the posterior density of state variable in dynamical system；It is assumed that moment t dbjective state variable is
st, give target observation collection y1:t={ y1,y2,…,yt, then current state stIt can be determined by maximum a posteriori probability：
The recursive calculation of wherein posterior probability is divided into formula (5) prediction and formula (6) updates two steps
p(st|y1:t)=∫ p (st|st-1)p(st-1|y1:t-1)dst-1 (5)
Wherein p (st|st-1) it is that state transition probability is used for describing dynamic model, p (yt|st) it is that observation likelihood function is used for
Observation model is described；Moved with six affine transformation parameters of image to define consecutive frame target, if st=(υ1,υ2,υ3,υ4,
tx,ty), wherein (υ1,υ2,υ3,υ4) anglec of rotation, yardstick, area ratio, incline direction, (t are represented respectivelyx,ty) represent 2D positions
Parameter, the state transfer of adjacent interframe are then expressed as with Gaussian Profile
p(st|st-1)=N (st；st-1,Σ)
Wherein N () is that gauss of distribution function, Σ are that (its diagonal element is that corresponding sports are joined in s to diagonal covariance matrix
Several variance).Based on the grader learnt, observation model is defined as into p (y | s) ∝ f (z), f (z) is calculated by formula (4)
Classify score, then the candidate target with top score be considered as tracking result.
Preferably, in step (4), in order to realize the video frequency object tracking of robust, dynamically update To Template with
Background template and grader, it is specifically：
(a) on To Template DfRenewal, if α is coefficients of the new tracking result y in target dictionary template, s is y
It is Pasteur's coefficient between sample with the atom corresponding to greatest coefficient in α, siIt is y and To Template DfIn per monatomic Pasteur
Coefficient and smFor its minimum value, two threshold taus are concurrently set1<τ2.If s>τ2, then illustrate that tracking result can be well by target
Template representation；If s<τ1, then explanation tracking target is there occurs strong cosmetic variation, now with y replacements smCorresponding target
Sample；
(b) on background template DbRenewal, in the current frame determine tracking target after, according to height around target area
This distribution selection MBIndividual background sample, the sample in random replacement current background dictionary；
(c) renewal on grader, by the To Template D after current renewalfWith background template DbIn sample, that is, work as
Preceding positive negative sample, brings grader into and is trained and just obtain current grader.
The beneficial effects of the invention are as follows：
The present invention is to use local restriction linear coding method, and the partial structurtes of sample data are incorporated into collaboration expression side
In method, obtain the sample with good classification performance and represent；The core for representing to expand to multiple features fusion will be cooperateed with using kernel function
Space so that dictionary and rarefaction representation coefficient are strengthened the class discriminating power of target signature；Will under particle filter framework
The candidate target of two grader highest scorings can accurately and robustly track video object as tracking target.
Brief description of the drawings
Fig. 1-4 is the video frequency object tracking design sketch and tracking effect comparison diagram of case of the present invention, with other four kinds of sides
Method compares：The IVT methods that Ross in 2008 is proposed, VID methods that L1 methods that Mei in 2011 is proposed, Kwon in 2010 are proposed, 2009
The MIL methods that Babenko is proposed.
Note：In Fig. 1-4, what each rectangle circle was selected is the effect of each method, and rectangle frame corresponding to each method effect divides
Do not represented with label A, B, C, D, E, the method that each label represents respectively is described as follows：
A(IVT)；B(L1)；C(VTD)；D(MIL)；E (present invention)
Fig. 1-Oclcusion2 sequences (note：In the sequence, there is target rotation with blocking change)；From Fig. 1 tracking
As a result as can be seen that the present invention block and rotate when can be accurately tracked by target, IVT, VTD track unsuccessfully (such as #
501), MIL can realize tracking but can not estimate rotationally-varying, l1There is drift phenomenon in tracking.
Fig. 2-DavidIndoor sequences (note：With moving forward and backward for target, target cranial is there occurs rotationally-varying, together
Shi Yin distances distance and there is target size change, and background illumination is also changed therewith)；Fig. 3-Car11 sequences
(note：Background is complicated, objective fuzzy and target size are changed)；IVT algorithms, which are can be seen that, from Fig. 2 and Fig. 3 is better than other
3 kinds of algorithms, there is rotation adaptivity, but dimensional variation adaptability is poor, and the present invention adapts in target rotation with dimensional variation
Others are superior in performance.
Fig. 4 Deer sequences (note：Because there is objective fuzzy phenomenon in the quick motion of target)；From fig. 4, it can be seen that only
The present invention can position target with VTD algorithms, and other algorithms go out the phenomenon (such as #40) of active target, while VTD tracking
There is the situation of skew.The present invention can realize the target following of accurate robust under complex environment.
Embodiment
Under technical scheme is described in further detail by specific embodiment and with reference to accompanying drawing.
Reference picture 1-4, a kind of video target tracking method of multinuclear local restriction, methods described include：
(1) partial structurtes of sample data are incorporated into collaboration method for expressing, build the local restriction line of sample characteristics
Property coding.It is specific as follows：(a) under framework of sparse representation, test sample y=d1x1+d2x2+…+dnxn=Dx, it, which is cooperateed with, represents
Mesh its collaboration represent object function beThe association of sample is obtained by minimizing
It is same to be expressed as(b) partial structurtes of sample data are incorporated into collaboration to represent, its object function isThe local restriction collaboration that sample is obtained by minimizing is encoded to
(2) kernel method is utilized, builds the multinuclear local restriction collaboration coding of sample characteristics.It is specific as follows：(a) sample y exists
Image in higher dimensional space is by Φ=[φ (d1),φ(d2),…,φ(dn)] carry out linear expression, using accidental projection matrix PTAgain
High dimensional data is mapped to lower dimensional space, the object function (formula (1)) of local restriction nuclear coordination coding is obtained, seeks it local derviation simultaneously
Equal to 0 and the property according to kernel function, note P=Φ B obtain object function
(b) fusion kernel function is obtained by way of multi-core integration isUsing spatial color histogram and
Spatial gradient histogram represents target：For each target sample, the subregion of four area equations is divided into, so
Extracting subregion color property respectively afterwards and merging turns into targeted color feature hc.For Gradient Features, first by core [-
] and [- 0.5,0,0.5] 0.5,0,0.5TProcessing is filtered, obtains image gradient；Then same method is taken to obtain four
Subregion histograms of oriented gradients, and obtain spatial gradient histogram h by merging four sub-regions histogramsg.According toTo calculate KcMiddle element, Kg、Kc(·,y)、KgThe computational methods and K of (, y)cIt is identical.So as to
Obtain the multinuclear local restriction collaboration coding of sample characteristics
(3) SVM is based on, the grader score of sample is embedded into the tracking that video object is realized under particle filter framework.
It is specific as follows：(a) extract target sample (positive sample) according to Gaussian Profile around the first frame target area and background sample is (negative
Sample), multinuclear local restriction collaboration coding z of the positive negative sample under current dictionary is calculated respectivelyi, by the coding of positive negative sample and
Its corresponding labelSubstitute into SVM to be trained, by minimizing cost functionCarry out Study strategies and methods, the score f (z) of grader is obtained by formula (4)；(b) base
In the grader learnt, the observation model under particle filter framework is defined as p (y | s) ∝ f (z), obtains top score
Candidate target, as tracking result.
(4) change according to target and background sample, dynamic update To Template and background template and grader.Specifically
It is as follows：(a) To Template DfRenewal.If s>τ2, then illustrate that tracking result can be represented by To Template well, retain Df；
If s<τ1, then explanation tracking target is there occurs strong cosmetic variation, now with y replacements smCorresponding target sample.(b)
Background template DbRenewal, in the current frame determine tracking target after, around target area according to Gaussian Profile select MBThe individual back of the body
Scape sample, the sample in random replacement current background dictionary.(c) renewal of grader, by the To Template D after current renewalf
With background template DbIn sample, bring grader into and be trained and just obtain current grader.
Reference picture 1-4 gives the contrast of tracking effect of the present invention and other four kinds of method tracking effects, it can be seen that
Generation target motion blur, dimensional variation and quickly move and block, illumination variation when, tracking effect of the invention is better than it
His four kinds of methods, realize accurate and robust target following.
Claims (5)
1. a kind of video target tracking method of multinuclear local restriction, it is characterised in that the tracking comprises the steps：
(1) partial structurtes of sample data are incorporated into collaboration method for expressing, build the local restriction uniform enconding of sample characteristics；
(2) kernel method is utilized, builds the multinuclear local restriction collaboration coding of sample characteristics；(3) SVMs is based on, by point of sample
Class device score is embedded into the tracking that video object is realized under particle filter framework；(4) change according to target and background sample, move
State updates To Template and background template and grader.
2. the video target tracking method of multinuclear local restriction as claimed in claim 1, it is characterised in that：In step (1),
Represented to obtain the sample with good classification performance, the partial structurtes of sample data are incorporated into collaboration expression, carry out structure
The local restriction collaboration coding of sample characteristics is built, is specifically：(a) under framework of sparse representation, test sampleIt is expressed as
Y=d1x1+d2x2+…+dnxn=Dx, wherein dictionaryIts cooperate with represent object function beThe collaboration that sample is obtained by minimizing is expressed as
(b) partial structurtes of sample data are incorporated into collaboration to represent, its object function is
WhereinIt is the diagonal matrix of a local restriction, the local restriction that sample is obtained by minimizing is assisted
It is same to be encoded to
3. the video target tracking method of multinuclear local restriction as claimed in claim 1, it is characterised in that：In step (2),
Based on kernel method, the multinuclear local restriction collaboration coding of sample characteristics is built, is specifically：(a) kernel method is used by non-linear number
According to High-dimensional Linear nuclear space is mapped to, images of the sample y in higher dimensional space is by Φ=[φ (d1),φ(d2),…,φ(dn)]
Carry out linear expression, using accidental projection matrix PTHigh dimensional data is mapped to lower dimensional space again, local restriction nuclear coordination coding
Object function is
Local derviation and the property according to kernel function are asked it,
Note P=Φ B obtain object function and areWherein K=ΦTΦ
It is core Gram matrixes；(b) in order to stably track target, target is described using multiple features, then passes through the side of multi-core integration
Formula obtains merging kernel functionSo as to obtain the multinuclear local restriction of sample characteristics collaboration coding
4. the video target tracking method of multinuclear local restriction as claimed in claim 1, it is characterised in that：In step (3),
Based on SVMs, the grader score of sample is embedded into the tracking that video object is realized under particle filter framework, specifically
It is：(a) a number of target sample and background sample, target sample are extracted according to Gaussian Profile around the first frame target area
Originally, background sample is also respectively positive sample, negative sample, calculates multinuclear local restriction of the positive and negative samples under current dictionary respectively
Collaboration coding zi, by the coding of positive and negative samples and its corresponding labelSubstitute into branch
Hold vector machine to be trained, by minimizing cost functionCarry out learning classification
Device, the score of grader are calculated as(b) based on the grader learnt, by the sight under particle filter framework
Survey model definition is p (y | s) ∝ f (z), and the candidate target with top score is tracking result.
5. the video target tracking method of multinuclear local restriction as claimed in claim 1, it is characterised in that：In step (4),
In order to realize the video frequency object tracking of robust, To Template and background template and grader are dynamicallyd update, is specifically：(a)
On To Template DfRenewal, if tracking result y can well by To Template represent if retain Df, otherwise obtain y and DfIn
The minimum dictionary atom of Pasteur's coefficient, and replaced it by y；(b) on background template DbRenewal, in the current frame determine tracking
After target, M is selected according to Gaussian Profile around target areaBIndividual background sample, the sample in random replacement current background dictionary
This；(c) renewal on grader, by the To Template D after current renewalfWith background template DbIn sample, bring grader into
It is trained and just obtains current grader.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710455426.XA CN107368785A (en) | 2017-06-16 | 2017-06-16 | The video target tracking method of multinuclear local restriction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710455426.XA CN107368785A (en) | 2017-06-16 | 2017-06-16 | The video target tracking method of multinuclear local restriction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107368785A true CN107368785A (en) | 2017-11-21 |
Family
ID=60306515
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710455426.XA Pending CN107368785A (en) | 2017-06-16 | 2017-06-16 | The video target tracking method of multinuclear local restriction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107368785A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108719A (en) * | 2018-01-05 | 2018-06-01 | 重庆邮电大学 | A kind of Weighted Kernel is sparse and cooperates with the Hyperspectral Image Classification method for representing coefficient |
| CN109348410A (en) * | 2018-11-16 | 2019-02-15 | 电子科技大学 | Indoor orientation method based on global and local joint constraint transfer learning |
| CN112578675A (en) * | 2021-02-25 | 2021-03-30 | 中国人民解放军国防科技大学 | High-dynamic vision control system and task allocation and multi-core implementation method thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440645A (en) * | 2013-08-16 | 2013-12-11 | 东南大学 | Target tracking algorithm based on self-adaptive particle filter and sparse representation |
| CN103544483A (en) * | 2013-10-25 | 2014-01-29 | 合肥工业大学 | United target tracking method based on local sparse representation and system thereof |
| CN104766343A (en) * | 2015-03-27 | 2015-07-08 | 电子科技大学 | Vision target tracking method based on sparse representation |
| KR101542206B1 (en) * | 2014-04-24 | 2015-08-12 | (주)디브이알씨앤씨 | Method and system for tracking with extraction object using coarse to fine techniques |
-
2017
- 2017-06-16 CN CN201710455426.XA patent/CN107368785A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103440645A (en) * | 2013-08-16 | 2013-12-11 | 东南大学 | Target tracking algorithm based on self-adaptive particle filter and sparse representation |
| CN103544483A (en) * | 2013-10-25 | 2014-01-29 | 合肥工业大学 | United target tracking method based on local sparse representation and system thereof |
| KR101542206B1 (en) * | 2014-04-24 | 2015-08-12 | (주)디브이알씨앤씨 | Method and system for tracking with extraction object using coarse to fine techniques |
| CN104766343A (en) * | 2015-03-27 | 2015-07-08 | 电子科技大学 | Vision target tracking method based on sparse representation |
Non-Patent Citations (3)
| Title |
|---|
| LAI WEI等: "Kernel locality-constrained collaborative representation based discriminant analysis", 《 KNOWLEDGE-BASED SYSTEMS 》 * |
| LINGFENG WANG等: "Visual Tracking Via Kernel Sparse Representation With Multikernel Fusion", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
| 林冬冬: "基于稀疏表达和机器学习的目标跟踪的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108719A (en) * | 2018-01-05 | 2018-06-01 | 重庆邮电大学 | A kind of Weighted Kernel is sparse and cooperates with the Hyperspectral Image Classification method for representing coefficient |
| CN109348410A (en) * | 2018-11-16 | 2019-02-15 | 电子科技大学 | Indoor orientation method based on global and local joint constraint transfer learning |
| CN112578675A (en) * | 2021-02-25 | 2021-03-30 | 中国人民解放军国防科技大学 | High-dynamic vision control system and task allocation and multi-core implementation method thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bhagat et al. | Indian sign language gesture recognition using image processing and deep learning | |
| Zheng et al. | A novel background subtraction algorithm based on parallel vision and Bayesian GANs | |
| CN109800689B (en) | Target tracking method based on space-time feature fusion learning | |
| Liu et al. | Denet: A universal network for counting crowd with varying densities and scales | |
| CN104933417B (en) | A kind of Activity recognition method based on sparse space-time characteristic | |
| Yang et al. | Small object augmentation of urban scenes for real-time semantic segmentation | |
| CN106599836B (en) | Multi-face tracking method and tracking system | |
| CN107609460A (en) | A kind of Human bodys' response method for merging space-time dual-network stream and attention mechanism | |
| Zhang et al. | Image object detection and semantic segmentation based on convolutional neural network | |
| CN104835182A (en) | Method for realizing dynamic object real-time tracking by using camera | |
| Kok et al. | GrCS: Granular computing-based crowd segmentation | |
| Zhou et al. | Research on pedestrian detection technology based on the SVM classifier trained by HOG and LTP features | |
| CN113657414B (en) | Object identification method | |
| CN115661246A (en) | Attitude estimation method based on self-supervision learning | |
| CN107368785A (en) | The video target tracking method of multinuclear local restriction | |
| CN113312973A (en) | Method and system for extracting features of gesture recognition key points | |
| Pham et al. | Pencilnet: Zero-shot sim-to-real transfer learning for robust gate perception in autonomous drone racing | |
| Rani et al. | An effectual classical dance pose estimation and classification system employing convolution neural network–long shortterm memory (CNN-LSTM) network for video sequences | |
| Ren et al. | A new multi-scale pedestrian detection algorithm in traffic environment | |
| Cao et al. | RetinaMOT: rethinking anchor-free YOLOv5 for online multiple object tracking | |
| Zheng et al. | Self-supervised monocular depth estimation based on combining convolution and multilayer perceptron | |
| Wang et al. | Visual object tracking based on light-field imaging in the presence of similar distractors | |
| Chen et al. | MICPL: Motion-Inspired Cross-Pattern Learning for Small-Object Detection in Satellite Videos | |
| Chun et al. | Human action recognition based on improved motion history image and deep convolutional neural networks | |
| Hua et al. | Scn: dilated silhouette convolutional network for video action recognition |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171121 |
|
| WD01 | Invention patent application deemed withdrawn after publication |