CN110084834A - A kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction - Google Patents

A kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction Download PDF

Info

Publication number
CN110084834A
CN110084834A CN201910349128.1A CN201910349128A CN110084834A CN 110084834 A CN110084834 A CN 110084834A CN 201910349128 A CN201910349128 A CN 201910349128A CN 110084834 A CN110084834 A CN 110084834A
Authority
CN
China
Prior art keywords
tensor
feature
value decomposition
singular value
characteristic
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.)
Granted
Application number
CN201910349128.1A
Other languages
Chinese (zh)
Other versions
CN110084834B (en
Inventor
傅衡成
周武能
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.)
Donghua University
National Dong Hwa University
Original Assignee
Donghua University
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 Donghua University filed Critical Donghua University
Priority to CN201910349128.1A priority Critical patent/CN110084834B/en
Publication of CN110084834A publication Critical patent/CN110084834A/en
Application granted granted Critical
Publication of CN110084834B publication Critical patent/CN110084834B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • 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]
    • 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
    • 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
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

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

Abstract

The invention discloses a kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction, comprising: extracts various features to every one-frame video data and constructs tensor structure；Singular value decomposition is carried out to the tensor of building；To the feature training correlation filter after dimensionality reduction, target is tracked.The present invention can effectively reduce feature quantity, accelerate tracking velocity and preferably remain the structural information of feature compared with the modes such as traditional principal component analysis Feature Dimension Reduction based on vector；The robustness that tensor singular value decomposition there is invariance enhancing tracker to rotate to target the rotation of feature.

Description

A kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction
Technical field
The present invention relates to a kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction, belong to video mesh Mark tracking technique field.
Background technique
Target following is significant to the development in the fields such as robot, unmanned plane, automatic Pilot, Navigation And Guidance.Example Such as, in human-computer interaction process, camera constantly tracks human body behavior, and makes machine by a series of analysis processing People reaches to human body attitude, movement, the understanding of gesture, so that the friendly exchange of people and machine be better achieved；In unmanned plane mesh During mark tracking, the visual information of target is constantly obtained, and sends ground control station to, by algorithm to sequence of video images Analyzed, obtain tracking target real-time position information, with guarantee tracking target be in real time unmanned plane field range it It is interior.
Object tracking process be related to illumination variation, dimensional variation, plane internal rotation, plane external rotation, block, deformation, Motion blur, the quickly series of challenges such as movement, background spot, low resolution, in recent years " correlation filtering " class method for tracking target Not only tracking velocity is very fast, and tracking accuracy also shows well, but being continuously increased with various features, correlation filter Tracking velocity decline is serious.
In recent years, characteristics of image used in correlation filtering is continuously increased, such as color name feature, histogram of gradients feature And the depth characteristic of depth convolutional neural networks, these features have biggish promotion to tracking accuracy, but make correlation The tracking velocity decline of filter is quickly.
Summary of the invention
The object of the present invention is to provide the damages that one kind can promote precision and speed using more features The method for tracking target of mistake.
In order to achieve the above object, the technical solution of the present invention is to provide a kind of special based on quick tensor singular value decomposition Levy the method for tracking target of dimensionality reduction, which is characterized in that carry out target following, packet using tensor singular value decomposition Feature Dimension Reduction feature Include following steps:
(1) the gradient orientation histogram feature HOG, color name feature CN, pre-training for extracting t frame tracking result window are good Depth convolution feature CNN；
(2) feature that step (1) is extracted is discharged into the dropping cut slice of tensor, forms 4 mutually independent three ranks tensors, It is denoted as L respectivelyi, i=1,2,3,4；
(3) as unit of the dropping cut slice of each tensor, average characteristics, the average spy of ith feature tensor are calculated separately Sign is denoted as Mi, then have:
In formula, NiFor the dropping cut slice quantity of ith feature tensor；Li(j: :) indicate three rank tensor LiJ-th of water Flat slice；
(4) make the dropping cut slice of each characteristic tensor that corresponding average characteristics be individually subtracted, and be denoted as Ai；
(5) Fast Fourier Transform (FFT) is utilized, temporal signatures tensor is transformed into frequency domain, and to each feature of frequency domain form Each dropping cut slice of tensor carries out traditional Singular Value Decomposition Using, and intercepts to the column of left singular matrix, wherein The number of reservation is that the preceding k dimension of column is also equal to the dimension of characteristic tensor after dimensionality reduction；To the singular value decomposition of each side slice After completion, the left singular matrix of reservation is formed left singularity characteristics vector according still further to original putting in order, finally by fast Left singularity characteristics vector is transformed into time domain, the left singularity characteristics tensor of each forms of time and space from frequency domain by fast Fourier inversion It is denoted as respectively
(6) tensor product fortune is carried out with the characteristic tensor for subtracting average characteristics using the left singularity characteristics tensor of forms of time and space Calculation obtains the characteristic tensor after dimensionality reduction, and accordingly obtains plus before to each of characteristic tensor after each dimensionality reduction front slice Average characteristics, obtain characteristic tensor Fi:
In formula, tprod () indicates tensor product；tran(Ui) indicate characteristic tensor U transposition
(7) to each characteristic tensor FiTransposition is carried out, front slice is discharged into side slice, then to each spy Levy tensor FiSide lateral plate all train and a filter and be denoted asAnd filter before is updated, the public affairs of update Formula is shown below:
In formula,Indicate that the filter of i-th kind of feature of t frame, η are the learning rate of filter；
(8) gradient orientation histogram feature HOG, color name feature CN and the depth convolution of the candidate region of t+1 are extracted Feature CNN, the sequence and step (2) of discharge are consistent, each feature of the projection operator and t+1 frame that are obtained with t frame Tensor carries out tensor product, obtains characteristic tensor after dimensionality reduction；
(9) filter obtained with t frame carries out convolution algorithm to the characteristic tensor after dimensionality reduction, each characteristic tensor Side is sliced to obtain a confidence map, these confidence maps are added, and finally obtains a response diagram, by the maximum position of response diagram As target in the position of t+1 frame；
(10) last frame is judged whether it is, if not last frame, then enable t=t+1, return step (1), if finally One frame then stops tracking.
Preferably, in step (1), the gradient orientation histogram feature HOG includes 31 layers, the color name feature CN packet Containing 11 layers, the Layer1 of the depth convolution feature CNN includes 512 layers comprising 96 layers, Layer5.
Preferably, in step (5), the Singular Value Decomposition Using of frequency domain form is shown below, and carries out more to projection operator Newly, more new formula are as follows:
In formula, Pi tIndicate that the projection operator of t frame ith feature tensor, α indicate the learning rate of projection operator.
The present invention can effectively reduce feature quantity, accelerate tracking velocity, with traditional principal component based on vector point The modes such as analysis Feature Dimension Reduction are compared, and the structural information of feature is preferably remained；Tensor singular value decomposition has the rotation of feature The robustness for thering is invariance enhancing tracker to rotate target.
Detailed description of the invention
Fig. 1 is the flow algorithm that the present invention is implemented；
Fig. 2 is the example for the tensor property dimensionality reduction that the present invention is implemented.
Specific embodiment
Present invention will be further explained below with reference to specific examples.It should be understood that these embodiments are merely to illustrate the present invention Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, those skilled in the art Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited Range.
The present invention provides a kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction, including it is following Step:
(1) the gradient orientation histogram feature HOG, color name feature CN, pre-training for extracting t frame tracking result window are good Depth convolution feature CNN.Wherein gradient orientation histogram feature HOG includes 31 layers, and color name feature CN includes 11 layers, depth volume The Layer1 of product feature CNN includes 512 layers comprising 96 layers, Layer5.
(2) these features are discharged into the dropping cut slice of tensor, 4 mutually independent three ranks tensors is formed, is denoted as respectively Li, i=1,2,3,4.
(3) as unit of the dropping cut slice of each tensor, average characteristics are calculated separately.
In formula, NiIt is 31 for the dropping cut slice quantity of ith feature tensor, such as the dropping cut slice quantity of HOG characteristic tensor； Li(j: :) indicate three rank tensor LiJ-th of dropping cut slice, MiFor the average characteristics of i-th of tensor.
(4) make the dropping cut slice of each characteristic tensor that corresponding average characteristics be individually subtracted, and be denoted as Ai, wherein i=1,2, 3,4.
(5) Fast Fourier Transform (FFT) is utilized, temporal signatures tensor is transformed into frequency domain, and to each feature of frequency domain form Each dropping cut slice of tensor carries out traditional Singular Value Decomposition Using, and intercepts to the column of left singular matrix, wherein The number of reservation is that the preceding k dimension of column is also equal to the dimension of characteristic tensor after dimensionality reduction.To the singular value decomposition of each side slice After completion, the left singular matrix of reservation is formed tensor according still further to original putting in order, it is anti-finally by fast Fourier Transformation, is transformed into time domain from frequency domain for left singularity characteristics vector, the left singularity characteristics tensor of each forms of time and space (also known as projects Operator) it is denoted as respectivelyWherein the Singular Value Decomposition Using of frequency domain form is shown below, and is updated to projection operator, More new formula are as follows:
In formula, Pi tIndicate the projection operator of t frame ith feature tensor；The learning rate of α expression projection operator.
(6) tensor product operation is carried out with the characteristic tensor for subtracting average characteristics using the unusual tensor in a left side of forms of time and space to obtain Characteristic tensor after to dimensionality reduction, and each of characteristic tensor after each dimensionality reduction front slice is accordingly obtained plus before flat Equal feature.It is shown below:
In formula, tprod () indicates tensor product, tran (Ui) indicate tensor UiTransposition, FiIndicate obtained feature Amount.
(7) transposition is carried out to each characteristic tensor, front slice is discharged into side slice, then to each tensor Side lateral plate all train and a filter and be denoted asAnd filter before is updated, the formula of update such as following formula It is shown:
In formula,Indicate that the filter of i-th kind of feature of t frame, η are the learning rate of filter.
(8) gradient orientation histogram feature HOG, color name feature CN and the depth convolution of the candidate region of t+1 are extracted Feature CNN, the sequence and step (2) of discharge are consistent, each feature of the projection operator and t+1 frame that are obtained with t frame Tensor carries out tensor product, obtains characteristic tensor after dimensionality reduction.
(9) filter obtained with t frame carries out convolution algorithm to the characteristic tensor after dimensionality reduction, each characteristic tensor Side is sliced to obtain a confidence map, these confidence maps are added, and finally obtains a response diagram, by the maximum position of response diagram As target in the position of t+1 frame.
(10) last frame is judged whether it is, if not last frame, then enable t=t+1, return step (1), if finally One frame then stops tracking.

Claims (3)

1. a kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction, which is characterized in that odd using tensor Different value characteristics of decomposition dimensionality reduction feature carries out target following, includes the following steps:
(1) the good depth of the gradient orientation histogram feature HOG, color name feature CN, pre-training of t frame tracking result window is extracted Convolution feature CNN；
(2) feature that step (1) is extracted is discharged into the dropping cut slice of tensor, forms 4 mutually independent three ranks tensors, respectively It is denoted as Li, i=1,2,3,4；
(3) as unit of the dropping cut slice of each tensor, average characteristics, the average characteristics note of ith feature tensor are calculated separately For Mi, then have:
In formula, NiFor the dropping cut slice quantity of ith feature tensor；Li(j: :) indicate three rank tensor LiJ-th of horizontal cutting Piece；
(4) make the dropping cut slice of each characteristic tensor that corresponding average characteristics be individually subtracted, and be denoted as Ai；
(5) Fast Fourier Transform (FFT) is utilized, temporal signatures tensor is transformed into frequency domain, and to each characteristic tensor of frequency domain form Each dropping cut slice carry out traditional Singular Value Decomposition Using, and the column of left singular matrix are intercepted, wherein retain Number be that the preceding k dimension of column is also equal to the dimension of characteristic tensor after dimensionality reduction；The singular value decomposition of each side slice is completed Later, the left singular matrix of reservation is formed left singularity characteristics vector according still further to original putting in order, finally by quick Fu In leaf inverse transformation, left singularity characteristics vector is transformed into time domain from frequency domain, the left singularity characteristics tensor difference of each forms of time and space It is denoted as
(6) tensor product operation is carried out with the characteristic tensor for subtracting average characteristics using the left singularity characteristics tensor of forms of time and space to obtain Characteristic tensor after to dimensionality reduction, and each of characteristic tensor after each dimensionality reduction front slice is accordingly obtained plus before flat Equal feature obtains characteristic tensor Fi:
In formula, tprod () indicates tensor product；tran(Ui) indicate characteristic tensor U transposition
(7) to each characteristic tensor FiTransposition is carried out, front slice is discharged into side slice, then to each feature Measure FiSide lateral plate all train and a filter and be denoted asAnd filter before is updated, the formula of update is such as Shown in following formula:
In formula,Indicate that the filter of i-th kind of feature of t frame, η are the learning rate of filter；
(8) gradient orientation histogram feature HOG, color name feature CN and the depth convolution feature of the candidate region of t+1 are extracted CNN, the sequence and step (2) of discharge are consistent, each characteristic tensor of the projection operator and t+1 frame that are obtained with t frame Tensor product is carried out, characteristic tensor after dimensionality reduction is obtained；
(9) filter obtained with t frame carries out convolution algorithm to the characteristic tensor after dimensionality reduction, the side of each characteristic tensor Slice obtain a confidence map, these confidence maps are added, a response diagram is finally obtained, using the maximum position of response diagram as Target is in the position of t+1 frame；
(10) judge whether it is last frame, if not last frame, then enable t=t+1, return step (1), if last Frame then stops tracking.
2. the as described in claim a kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction, feature It is, in step (1), the gradient orientation histogram feature HOG includes 31 layers, and the color name feature CN includes 11 layers, institute State depth convolution feature CNN Layer1 include 96 layers, Layer5 include 512 layers.
3. the as described in claim a kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction, feature It is, in step (5), the Singular Value Decomposition Using of frequency domain form is shown below, and is updated to projection operator, more new formula Are as follows:
In formula, Pi tIndicate that the projection operator of t frame ith feature tensor, α indicate the learning rate of projection operator.
CN201910349128.1A 2019-04-28 2019-04-28 Target tracking method based on rapid tensor singular value decomposition feature dimension reduction Active CN110084834B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910349128.1A CN110084834B (en) 2019-04-28 2019-04-28 Target tracking method based on rapid tensor singular value decomposition feature dimension reduction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910349128.1A CN110084834B (en) 2019-04-28 2019-04-28 Target tracking method based on rapid tensor singular value decomposition feature dimension reduction

Publications (2)

Publication Number Publication Date
CN110084834A true CN110084834A (en) 2019-08-02
CN110084834B CN110084834B (en) 2021-04-06

Family

ID=67417289

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910349128.1A Active CN110084834B (en) 2019-04-28 2019-04-28 Target tracking method based on rapid tensor singular value decomposition feature dimension reduction

Country Status (1)

Country Link
CN (1) CN110084834B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110766723A (en) * 2019-10-22 2020-02-07 湖南大学 Unmanned aerial vehicle target tracking method and system based on color histogram similarity
CN116202760A (en) * 2023-05-05 2023-06-02 赛腾机电科技（常州）有限公司 Singular value decomposition method and system for third-order tensor for mechanical fault diagnosis
CN117058886A (en) * 2023-10-12 2023-11-14 安徽宇疆科技有限公司 Beidou space-time data model based on third-order tensor and traffic flow analysis method

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0920120A3 (en) * 1997-11-27 2001-08-29 Vectron Elektronik GmbH Method and apparatus for controlling the movement path of the workpiece supporting head of an orbital vibration welding system
CN102592135A (en) * 2011-12-16 2012-07-18 温州大学 Visual tracking method of subspace fusing target space distribution and time sequence distribution characteristics
CN104021395A (en) * 2014-06-20 2014-09-03 华侨大学 Target tracing algorithm based on high-order partial least square method
CN104751169A (en) * 2015-01-10 2015-07-01 哈尔滨工业大学（威海） Method for classifying rail failures of high-speed rail
CN107093189A (en) * 2017-04-18 2017-08-25 山东大学 Method for tracking target and system based on adaptive color feature and space-time context
CN108305297A (en) * 2017-12-22 2018-07-20 上海交通大学 A kind of image processing method based on multidimensional tensor dictionary learning algorithm
CN108665481A (en) * 2018-03-27 2018-10-16 西安电子科技大学 Multilayer depth characteristic fusion it is adaptive resist block infrared object tracking method
CN108764249A (en) * 2018-04-23 2018-11-06 云南民族大学 A kind of invariable rotary multi-source image method for describing local characteristic, system and device
US10213274B1 (en) * 2018-03-12 2019-02-26 King Saud University Method of tracking and navigation for a dental instrument
CN109447073A (en) * 2018-11-08 2019-03-08 电子科技大学 A kind of method for detecting infrared puniness target based on tensor Robust Principal Component Analysis

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0920120A3 (en) * 1997-11-27 2001-08-29 Vectron Elektronik GmbH Method and apparatus for controlling the movement path of the workpiece supporting head of an orbital vibration welding system
CN102592135A (en) * 2011-12-16 2012-07-18 温州大学 Visual tracking method of subspace fusing target space distribution and time sequence distribution characteristics
CN104021395A (en) * 2014-06-20 2014-09-03 华侨大学 Target tracing algorithm based on high-order partial least square method
CN104751169A (en) * 2015-01-10 2015-07-01 哈尔滨工业大学（威海） Method for classifying rail failures of high-speed rail
CN107093189A (en) * 2017-04-18 2017-08-25 山东大学 Method for tracking target and system based on adaptive color feature and space-time context
CN108305297A (en) * 2017-12-22 2018-07-20 上海交通大学 A kind of image processing method based on multidimensional tensor dictionary learning algorithm
US10213274B1 (en) * 2018-03-12 2019-02-26 King Saud University Method of tracking and navigation for a dental instrument
CN108665481A (en) * 2018-03-27 2018-10-16 西安电子科技大学 Multilayer depth characteristic fusion it is adaptive resist block infrared object tracking method
CN108764249A (en) * 2018-04-23 2018-11-06 云南民族大学 A kind of invariable rotary multi-source image method for describing local characteristic, system and device
CN109447073A (en) * 2018-11-08 2019-03-08 电子科技大学 A kind of method for detecting infrared puniness target based on tensor Robust Principal Component Analysis

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
CHRISTOPHER W ET AL: "《Computation Estimation of Scene Structure Through Texture Gradient Cues》", 《ELECTRONIC IMAGING》 *
亚森江.木沙等: "《张量核范数回归的目标跟踪》", 《中国图象图形学报》 *
张亚红: "《 基于显著性检测和压缩感知的视觉跟踪》", 《中国优秀硕士学位论文全文数据库信息科技辑》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110766723A (en) * 2019-10-22 2020-02-07 湖南大学 Unmanned aerial vehicle target tracking method and system based on color histogram similarity
CN116202760A (en) * 2023-05-05 2023-06-02 赛腾机电科技（常州）有限公司 Singular value decomposition method and system for third-order tensor for mechanical fault diagnosis
CN116202760B (en) * 2023-05-05 2023-08-18 赛腾机电科技（常州）有限公司 Singular value decomposition method and system for third-order tensor for mechanical fault diagnosis
CN117058886A (en) * 2023-10-12 2023-11-14 安徽宇疆科技有限公司 Beidou space-time data model based on third-order tensor and traffic flow analysis method

Also Published As

Publication number Publication date
CN110084834B (en) 2021-04-06

Similar Documents

Publication Publication Date Title
CN108010067B (en) A kind of visual target tracking method based on combination determination strategy
Liu et al. Visual tracking in complex scenes: A location fusion mechanism based on the combination of multiple visual cognition flows
CN108053419B (en) Multi-scale target tracking method based on background suppression and foreground anti-interference
CN108986140B (en) Target scale self-adaptive tracking method based on correlation filtering and color detection
CN107016689A (en) A kind of correlation filtering of dimension self-adaption liquidates method for tracking target
JP4677046B2 (en) Fast human pose estimation using appearance and motion via multidimensional boost regression
CN111311647B (en) Global-local and Kalman filtering-based target tracking method and device
CN110084836A (en) Method for tracking target based on the response fusion of depth convolution Dividing Characteristics
CN113920170B (en) Pedestrian track prediction method, system and storage medium combining scene context and pedestrian social relationship
CN109146912A (en) A kind of visual target tracking method based on Objective analysis
CN111724411B (en) Multi-feature fusion tracking method based on opposite-impact algorithm
CN102156995A (en) Video movement foreground dividing method in moving camera
CN110084834A (en) A kind of method for tracking target based on quick tensor singular value decomposition Feature Dimension Reduction
CN109087337B (en) Long-time target tracking method and system based on hierarchical convolution characteristics
CN113111963B (en) Method for re-identifying pedestrian by black box attack
CN106887012A (en) A kind of quick self-adapted multiscale target tracking based on circular matrix
CN110245587B (en) Optical remote sensing image target detection method based on Bayesian transfer learning
Liu et al. Correlation filter with motion detection for robust tracking of shape-deformed targets
Xu et al. Robust tracking via weighted spatio-temporal context learning
CN110349176A (en) Method for tracking target and system based on triple convolutional networks and perception interference in learning
Mbelwa et al. Visual tracking tracker via object proposals and co-trained kernelized correlation filters
CN108734722A (en) A kind of vision tracking error correcting method based on PSR
CN109658436B (en) Fast moving target tracking method for mixed GOA and TLBO search
CN104182990B (en) A kind of Realtime sequence images motion target area acquisition methods
CN113327273A (en) Infrared target tracking method based on variable window function correlation filtering

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