CN109064492B - Context-dependent filtering video tracking method based on manifold regularization - Google Patents

Context-dependent filtering video tracking method based on manifold regularization Download PDF

Info

Publication number
CN109064492B
CN109064492B CN201810826449.1A CN201810826449A CN109064492B CN 109064492 B CN109064492 B CN 109064492B CN 201810826449 A CN201810826449 A CN 201810826449A CN 109064492 B CN109064492 B CN 109064492B
Authority
CN
China
Prior art keywords
samples
context
tracking
representing
tracking method
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810826449.1A
Other languages
Chinese (zh)
Other versions
CN109064492A (en
Inventor
宋慧慧
樊佳庆
张开华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Information Science and Technology
Original Assignee
Nanjing University of Information Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Information Science and Technology filed Critical Nanjing University of Information Science and Technology
Priority to CN201810826449.1A priority Critical patent/CN109064492B/en
Publication of CN109064492A publication Critical patent/CN109064492A/en
Application granted granted Critical
Publication of CN109064492B publication Critical patent/CN109064492B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/262—Analysis of motion using transform domain methods, e.g. Fourier domain methods
    • 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
    • G06T2207/20048—Transform domain processing
    • G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a context-dependent filtering video tracking method based on manifold regularization, wherein a basic sample of the relevant filtering tracking method adopts a central area and a peripheral area thereof as positive and negative samples, and a manifold structure is added into a series of circulating samples to perform discrete Fourier transform fast calculation. The context-dependent filtering video tracking method based on manifold regularization effectively utilizes the structural information between the target context areas, constructs the manifold regularization term to constrain the target function, well inhibits the noise under the chaotic background, and obviously improves the precision and the robustness of the tracking algorithm.

Description

Context-dependent filtering video tracking method based on manifold regularization
Technical Field
The invention relates to a context-dependent filtering video tracking method based on manifold regularization, belongs to the field of image processing, and particularly relates to a video tracking method.
Background
Visual target tracking is a classic computer vision problem that has been studied for many years, but has hitherto been a challenging task, mainly because the target appearance suffers from many disturbing factors such as: fast motion, cluttered background, arbitrary object appearance changes and deformations, etc. A robust appearance model plays an important role in practical tracking, and thus becomes an important research point in recent years.
Recently, discriminant correlation filtering-like methods have demonstrated great success in visual target tracking. These methods learn a correlation filter from a series of cyclic samples, which, due to their special structure, enable fast computation using discrete fourier transforms. However, these loop structures also introduce unwanted boundary effects that result in inaccurate target representations, thus limiting the discriminability of the learned correlation filters. The boundary effect has been partially mitigated by recent elaboration. Another disadvantage of the correlation filtering is that the object shape is described by a bounding box, which introduces clutter background noise in the object, especially when the object suffers from large-scale non-rigid deformations, ultimately resulting in a series of sub-optimal samples. Drift problems eventually arise over time.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide a video tracking method which can effectively inhibit noise under a chaotic background and improve the accuracy and the robustness of a tracking algorithm.
In order to achieve the above object, the present invention provides a context-dependent filtering video tracking method based on manifold regularization, in which a basic sample of the correlation filtering tracking method adopts a central region and a peripheral region thereof as positive and negative samples, and a manifold structure is added to a series of cyclic samples to perform discrete fourier transform fast calculation.
Further, the popular structure is obtained by introducing a linear graph laplacian regularization term into the relevant filtering objective function.
The graph laplacian regularization term is:
Figure GDA0003499599180000011
wherein L is a graph laplacian matrix.
The context-dependent filtering video tracking method based on manifold regularization specifically comprises the following steps:
(1) the method comprises the following steps: we first crop out the central region x0And the area x around itiAs positive and negative samples. Both positive and negative samples can be cyclically shifted to form a series of virtual samples, which are respectively marked as X0And Xi. A cropped central region and a region around the central region, which is referred to as a central region x0Four regions x of upper, lower, left and rightiCentral region x0Determined by the tracking result of the previous frame.
(2) Step two: a graph Laplacian regularization term is then constructed to take into account the similarity between these samples, if two samples xiAnd xjThe closer together in the geometry of the high-dimensional feature, the corresponding projection f (x)i) And f (x)j) And also closer together.
(3) Step three: drawing Laplace regularization term constructed in the previous step
Figure GDA0003499599180000021
Adding the obtained signal into the original context perception correlation filtering tracking to obtain an improved objective function
Figure GDA0003499599180000022
(4) Step four: obtaining a closed solution by using a regularized least squares regression solution method
Figure GDA0003499599180000023
Fast calculation by discrete Fourier transform
Figure GDA0003499599180000024
Namely, the rapid training is completed.
(5) Step five: finally, a response map is rapidly calculated by utilizing discrete Fourier transform, wherein the position of the maximum value is the tracking result. After each frame of tracking is finished, online updating of the model is carried out.
The invention has the beneficial effects that: the context-dependent filtering video tracking method based on manifold regularization effectively utilizes the structural information between the target context areas, constructs the manifold regularization term to constrain the target function, well inhibits the noise under the chaotic background, and obviously improves the precision and the robustness of the tracking algorithm.
Drawings
FIG. 1 is a schematic flow chart of a context-dependent filtering video tracking method based on manifold regularization according to the present invention;
FIG. 2 is a schematic diagram showing the comparison of the video tracking effect of the video tracking method of the present invention with the four methods of RPT, DSST, KCF and Struck in the prior art.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further explained with reference to the accompanying drawings and detailed description.
The invention provides a context-dependent filtering video tracking method based on manifold regularization, which specifically comprises the following operation steps as shown in fig. 1:
(1) the method comprises the following steps: we first crop out the central region x0And the area x around itiAs positive and negative samples. Both positive and negative samples can be cyclically shifted to form a series of virtual samples, which are respectively marked as X0And Xi. A cropped central region and a region around the central region, which is referred to as a central region x0Four regions x of upper, lower, left and rightiCentral region x0Determined by the tracking result of the previous frame.
(2) Step two: a graph Laplacian regularization term is then constructed to take into account the similarity between these samples, if two samples xiAnd xjThe closer together in the geometry of the high-dimensional feature, the corresponding projection f (x)i) And f (x)j) And also closer together. The specific calculation is as follows, the similarity between two samples:
Figure GDA0003499599180000031
where σ is the decay exponent of the gaussian kernel function.
(3) Step three: writing the graph Laplacian regularization term constructed in the previous step into a matrix form
Figure GDA0003499599180000032
Wherein, f (x)i) And f (x)j) Respectively represent samples xiAnd xjCorresponding projection, f ═ f1,f2,...,fm]TTo representA projection function, L being the graph laplacian matrix, calculated from L ═ D-W, where W is the correlation matrix and D is the diagonal matrix, whose diagonal elements
Figure GDA0003499599180000033
f(·)TRepresenting a transpose; wijRepresenting two samples xiAnd xjThe similarity between them.
Adding the obtained signal into an objective function of the original context perception correlation filtering tracking to obtain an improved objective function
Figure GDA0003499599180000034
Wherein the scalar yiRepresents each sample xiCorresponding label, WijRepresenting two samples xiAnd xjSimilarity between them, λ1,λ2Three regular term coefficients are respectively represented, and the three terms are controlled.
Then writing into a matrix form:
Figure GDA0003499599180000035
herein, the
Figure GDA0003499599180000041
The graph laplacian matrix is:
Figure GDA0003499599180000042
(4) step four: obtaining a closed solution by using a regularized least squares regression solution method
Figure GDA0003499599180000043
Fast calculation by discrete Fourier transform
Figure GDA0003499599180000044
Wherein
Figure GDA0003499599180000045
Represents a fast Fourier transform,. indicates a Hadamard product, (. o)*Denotes conjugation, a0Image blocks representing intermediate areas, aiRespective image blocks, y, representing surrounding areas0Denotes a0Labels, W, of all cyclically shifted samples of an image blockijRepresenting two samples xiAnd xjSimilarity between them, λ1,λ2Three regular term coefficients are respectively represented, and the three terms are controlled.
And solving w to finish the quick training.
(5) Step five: finally, since the original detection function is f*(z)＝wTz, fast calculation of response using discrete Fourier transform
Figure GDA0003499599180000051
Wherein
Figure GDA0003499599180000052
Indicating a fast fourier transform, a indicates a hadamard product.
The position of the maximum in the response r is the tracking result. In order to use long-term information in a video sequence, a simple and effective updating strategy is utilized, namely after tracking of each frame is finished, online updating of a model is carried out:
Figure GDA0003499599180000053
where beta is an online update parameter,
Figure GDA0003499599180000054
representing the filter at the current t frame.
Compared with the prior art that the video tracking method is adopted to carry out video tracking effect comparison, the success rate curve on the OTB50 data set is shown in figure 2, the method obtains the highest success rate curve score of 0.615 in the 5 methods, and the effectiveness of the method is illustrated.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims (1)

1. A video tracking method based on context-dependent filtering of manifold regularization is characterized in that: the method for tracking the context-dependent filtering video comprises the following specific steps:
(1) the method comprises the following steps: first, the center area x is cut out0As a positive sample, a region x around the positive sample is cut outiAs a negative sample; the positive samples are cyclically shifted out of a series of virtual samples denoted X0The negative samples are circularly shifted to obtain a series of virtual samples as Xi；
(2) Step two: constructing a graph Laplace regularization term to consider the similarity among the samples;
(3) step three: drawing Laplace regularization term constructed in the previous step
Figure FDA0003499599170000011
Adding the improved objective function into context perception correlation filtering tracking to obtain an improved objective function
Figure FDA0003499599170000012
Wherein, f (x)i) And f (x)j) Respectively represent samples xiAnd xjCorresponding projections, f denotes the projection function, L is the graph laplacian matrix, calculated from L ═ D-W, where W is the correlation matrixD is a diagonal matrix whose diagonal elements
Figure FDA0003499599170000013
f(·)TRepresenting a transpose; wijRepresenting two samples xiAnd xjSimilarity between them;
wherein the scalar yiRepresents each sample xiCorresponding labels, λ1，λ2Respectively representing three regular term coefficients;
(4) step four: obtaining a closed solution by using a regularized least squares regression solution method
Figure FDA0003499599170000021
Fast calculation by discrete Fourier transform
Figure FDA0003499599170000022
Namely, completing the quick training;
wherein
Figure FDA0003499599170000023
Represents a fast Fourier transform,. indicates a Hadamard product, (. o)*Denotes conjugation, a0Image blocks representing intermediate areas, aiRespective image blocks, y, representing surrounding areas0Denotes a0Labels of all cyclic shifted samples of the image block;
(5) step five: finally, fast calculating a response graph by utilizing discrete Fourier transform, wherein the position of the maximum value is the tracking result; after each frame of tracking is finished, online updating of the model is carried out.
CN201810826449.1A 2018-07-25 2018-07-25 Context-dependent filtering video tracking method based on manifold regularization Active CN109064492B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810826449.1A CN109064492B (en) 2018-07-25 2018-07-25 Context-dependent filtering video tracking method based on manifold regularization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810826449.1A CN109064492B (en) 2018-07-25 2018-07-25 Context-dependent filtering video tracking method based on manifold regularization

Publications (2)

Publication Number Publication Date
CN109064492A CN109064492A (en) 2018-12-21
CN109064492B true CN109064492B (en) 2022-04-01

Family

ID=64835474

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810826449.1A Active CN109064492B (en) 2018-07-25 2018-07-25 Context-dependent filtering video tracking method based on manifold regularization

Country Status (1)

Country Link
CN (1) CN109064492B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106097381A (en) * 2016-05-27 2016-11-09 北京理工大学 A kind of method for tracking target differentiating Non-negative Matrix Factorization based on manifold
CN107067410A (en) * 2016-05-27 2017-08-18 北京理工大学 A kind of manifold regularization correlation filtering method for tracking target based on augmented sample
CN107610049A (en) * 2017-08-21 2018-01-19 华侨大学 Image super-resolution method based on sparse Regularization Technique and weighting guiding filtering

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106097381A (en) * 2016-05-27 2016-11-09 北京理工大学 A kind of method for tracking target differentiating Non-negative Matrix Factorization based on manifold
CN107067410A (en) * 2016-05-27 2017-08-18 北京理工大学 A kind of manifold regularization correlation filtering method for tracking target based on augmented sample
CN107610049A (en) * 2017-08-21 2018-01-19 华侨大学 Image super-resolution method based on sparse Regularization Technique and weighting guiding filtering

Also Published As

Publication number Publication date
CN109064492A (en) 2018-12-21

Similar Documents

Publication Publication Date Title
CN108734723B (en) Relevant filtering target tracking method based on adaptive weight joint learning
CN107358623B (en) Relevant filtering tracking method based on significance detection and robustness scale estimation
CN111179307A (en) Visual target tracking method for full-volume integral and regression twin network structure
CN110599500B (en) Tumor region segmentation method and system of liver CT image based on cascaded full convolution network
CN111260688A (en) Twin double-path target tracking method
CN108171133B (en) Dynamic gesture recognition method based on characteristic covariance matrix
CN110120065B (en) Target tracking method and system based on hierarchical convolution characteristics and scale self-adaptive kernel correlation filtering
CN111738344A (en) Rapid target detection method based on multi-scale fusion
CN107067410B (en) Manifold regularization related filtering target tracking method based on augmented samples
CN112232134B (en) Human body posture estimation method based on hourglass network and attention mechanism
CN111640138B (en) Target tracking method, device, equipment and storage medium
Li et al. Pitaya detection in orchards using the MobileNet-YOLO model
CN105976397B (en) A kind of method for tracking target
CN110889865A (en) Video target tracking method based on local weighted sparse feature selection
Zhang et al. Sparse learning-based correlation filter for robust tracking
CN109685830B (en) Target tracking method, device and equipment and computer storage medium
CN102592135B (en) Visual tracking method of subspace fusing target space distribution and time sequence distribution characteristics
Li et al. Robust object tracking with discrete graph-based multiple experts
CN111724411A (en) Multi-feature fusion tracking method based on hedging algorithm
CN112489088A (en) Twin network visual tracking method based on memory unit
CN113361329B (en) Robust single-target tracking method based on example feature perception
CN109064492B (en) Context-dependent filtering video tracking method based on manifold regularization
Bourennane et al. An enhanced visual object tracking approach based on combined features of neural networks, wavelet transforms, and histogram of oriented gradients
CN113470074B (en) Self-adaptive space-time regularization target tracking method based on block discrimination
CN114863132A (en) Method, system, equipment and storage medium for modeling and capturing image spatial domain information

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
GR01 Patent grant
GR01 Patent grant