CN106651912A - Compressed sensing-based robust target tracking method - Google Patents

Compressed sensing-based robust target tracking method Download PDF

Info

Publication number
CN106651912A
CN106651912A CN201611020862.6A CN201611020862A CN106651912A CN 106651912 A CN106651912 A CN 106651912A CN 201611020862 A CN201611020862 A CN 201611020862A CN 106651912 A CN106651912 A CN 106651912A
Authority
CN
China
Prior art keywords
feature
sample
high identification
positive sample
compressed sensing
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
Application number
CN201611020862.6A
Other languages
Chinese (zh)
Inventor
黄鸿胜
何昭水
谢胜利
王伟华
杨森泉
王沛涛
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.)
Guangdong University of Technology
Original Assignee
Guangdong University of 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 Guangdong University of Technology filed Critical Guangdong University of Technology
Priority to CN201611020862.6A priority Critical patent/CN106651912A/en
Publication of CN106651912A publication Critical patent/CN106651912A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G06F18/24155—Bayesian classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a compressed sensing-based robust target tracking method. The method includes an initial stage and a tracking stage. According to the method provided by the technical schemes of the invention, further feature selection is performed on features extracted through compressed sensing, so that features with small classification effect contribution can be eliminated, so that compressed features with high discrimination performance are obtained; and the compressed features with high discrimination performance are classified. The compressed sensing-based robust target tracking method of the invention can track a target in a video in real time, has high accuracy, low computational complexity and enables an excellent tracking effect for a scene where illumination change is large, and a target is blocked.

Description

A kind of robust target tracking method based on compressed sensing
Technical field
The present invention relates to area of pattern recognition, more particularly to a kind of robust target tracking method based on compressed sensing.
Background technology
Target tracking has many actual applications, has widely in fields such as interpersonal interaction, video monitoring, behavior analysiss Using being one of key technology of computer video monitoring field research.However, in actual complex application scenarios, illumination becomes Change, shade, block, move the various factors such as mutation, background clutter and greatly challenge is brought to video object tracer technique.
On traditional method for tracing, as video camera captures a large amount of structural integrities and the very high data of redundancy, cause Data transmission period is long, and information processing is computationally intensive, is extremely difficult to the purpose of quick real-time tracing.
In order to adapt to the various demands of practical application area, accurate and quick video object tracer technique becomes One of hot issue of art circle and industrial circle extensive concern.
The content of the invention
To overcome the deficiencies in the prior art, the present invention to propose a kind of robust target tracking method based on compressed sensing.
The technical scheme is that what is be achieved in that：
A kind of robust target tracking method based on compressed sensing, including step
S1：Target location is gone out to the first frame flag in any one section of video image；
S2：Sparse projection matrix is generated to the place target area of target location；
S3：Around target location to the target area, sample positive sample masterplate and negative sample masterplate；
S4：Multiple dimensioned normalization characteristic is constructed to the positive sample masterplate and negative sample masterplate by convolution masterplate, by institute State multiple dimensioned normalization characteristic to be connected, obtain a long column vector；
S5：The long column vector is compressed using the sparse projection matrix, obtains low-dimensional feature；
S6：Retraining is carried out to the low-dimensional feature, a Projection Character matrix is obtained, is selected by this feature projection matrix Select out the high identification feature of positive sample and the high identification feature of negative sample；
S7：With the high identification feature of the positive sample and the high identification features training Bayes classifier of negative sample；
S8：Next frame picture is obtained, is sampled around the target location of previous frame, obtain a collection of candidate samples, Gu Determine the Projection Character matrix in step S6, repeat step S4-S6 obtains the high identification feature of candidate samples；
S9：The high identification feature of the candidate samples is sent into into Bayes classifier, the time of positive maximum probability is categorized as It is target location that sampling sheet is；
S10：Projection Character matrix in fixing step S6, repeat step S3-S7 update Bayes classifier；
S11：Repeat step S8-S11, until video terminates.
Further, in step S6, the construction process of Projection Character matrix includes step
S61：Solution formula
WhereinIt is by npIndividual positive sample feature A+And nnIndividual negative sample feature A-Constitute, K is the dimension of feature Number, λ is weight parameter, vectorIn element represent the property of the respective sample in training set A, i.e. ,+1 just represents Sample, -1 represents negative sample, draws Projection Character matrix；
S62：By A'=SA and x'=Sx, obtain sample form and candidate samples the high identification shadow feature of positive sample and The high identification feature of negative sample.
Further, low-dimensional described in step S5 is characterized in that intrinsic dimensionality is 300 low-dimensional features.
Further, the high identification feature of positive sample described in step S6 and the high identification of negative sample are characterized in that dimension is The high identification feature of positive sample and the high identification feature of negative sample of 60-80 dimensions.
Further, the high identification feature of the positive sample and the high identification of negative sample be characterized in that dimension be 70 dimensions just The high identification feature of sample and the high identification feature of negative sample.
The beneficial effects of the present invention is, compared with prior art, the present invention extracts feature based on compressed sensing can be effective Intrinsic dimensionality is reduced, the computation complexity of algorithm is substantially reduced, is improve the real-time of algorithm, while using proposed by the present invention Projection Character matrix carries out adaptive features select and can filter out high identification feature, strengthens the robust of tracing algorithm system Property, therefore, the present invention has good robustness and real-time on target tracking, with very strong using value.
Description of the drawings
Fig. 1 is robust target tracking method flow chart of the present invention based on compressed sensing.
Specific embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than the embodiment of whole.It is based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made Embodiment, belongs to the scope of protection of the invention.
Refer to Fig. 1, a kind of robust target tracking method based on compressed sensing of the present invention, including initial phase and chase after The track stage.
In initial phase, including step
S1：Target location I is gone out to the t frame flags in any one section of video imaget(x, y, w, h), wherein x, y represent mesh The coordinate in the target upper left corner, w, h represent the width and height of target area rectangle.
S2：The target area that target location is located is generated, the sparse projection matrix R of n × m dimensions, wherein n is Feature Compression Dimension afterwards, takes n=300 here, and m is the dimension of feature before compression, wherein
Arrange parameter：Studying factors λ=0.9, positive sample scope α=5, negative sample scope ξ=10, β=25, target search Scope γ=40；
S3：In target ItAround (x, y, w, h), sampling positive sample template D α=z | | | I (z)-It| | ＜ α } and negative sample This template Dξ,β=z | ξ ＜ | | I (z)-It| | ＜ β }, α, β, ξ are the firm search parameter for arranging, and z represents sample；
S4：It is provided with template set T={ hij}I=1:W, j=1:h, wherein element definition is：
For a width of w, the piecemeal of a height of h, convolution is carried out with template set T with sample, m=(wh) can be generated altogether2It is individual many Dimension normalization feature, these features are carried out connecting obtains column vector X of a m dimension.
S5：Low-dimensional feature F for obtaining that intrinsic dimensionality is n=300 is compressed with the sparse projection matrix to m dimensional features =R × X.
S6：To the low-dimensional feature retraining after these projections, obtain a Projection Character matrix, it is adaptively selected go out probably The high identification feature of 70 dimensions, this feature projection matrix are obtained by following methods, solve following formula first：
WhereinIt is by npIndividual positive sample feature A+And nnIndividual negative sample feature A- is constituted, and K is the dimension of feature Number, λ is weight parameter.VectorIn element represent the property of the respective sample in training set A, that is,+1 generation Table positive sample, -1 represents negative sample.
During nonzero element in sparse solution vector s of the formula represents K dimensional features, corresponding feature has high identification.By In l1Constraint it is openness, can be with the adaptively selected feature with identification by solving the formula our algorithm.Utilize Solution vector s we can construct projection matrix S.S is obtained by the full zero row removed in diagonal matrix S', and the element of S' is by following formula Be given：
Therefore the high identification projection properties of sample form and candidate samples can be obtained by A'=SA and x'=Sx.
S7：Bayes classifier is trained with the positive and negative template characteristic of high identification that step S6 is obtained
Naive Bayes Classifier is defined as：
Wherein p (y=1)=p (y=0), y ∈ { 0,1 } are a positive negative sample marks, and y=0 represents negative sample, and y=1 generations Table positive sample.It is assumed herein that conditional probability p (vi| y=1) and p (vi| y=0) Gaussian distributed
I.e.：
WhereinWithThe respectively average and variance of positive sample probability, andWithRespectively negative sample probability is equal Value and variance.
When wherein for i=1m, the initial value of Gaussian Distribution Parameters is all
In track phase, including step：
S8：In t+1 frames, sample Dγ=z | | | I (z)-It| | ＜ γ } obtain candidate samples, repeat step S3-S6, to waiting The multiple dimensioned normalization characteristic of sample extraction is selected, and Feature Compression is carried out with sparse projection matrix and is carried out using Projection Character matrix Feature selection.
S9：The feature of candidate samples is sent into Bayes classifier, the candidate samples for being categorized as positive maximum probability are considered as It is target probability of occurrence highest position It+1(x, y, w, h), i.e. target location.
S10：In target It+1Around (x, y, w, h), repeat step S3-S7, positive sample template D of samplingα=z | | | I (z)-It+1| | ＜ α } and negative sample template Dξ,β=z | ξ ＜ | | I (z)-It+1| | ＜ β }, equally align negative sample template extraction many Dimension normalization feature, and Feature Compression is carried out with sparse projection matrix and feature selection is carried out using Projection Character matrix.
Update Bayes classifier parameter
The update method of Bayes classifier is defined as：
WhereinIt is the average of grader, μ1It is the average of the positive sample in the collection of t+1 frames,It is the variance of grader, δ1 It is the variance of the positive sample in the collection of t+1 frames.
Can update after the same methodWith
S11：Repeat step S8-S10 proceeds target tracking, until video terminates.
The above is the preferred embodiment of the present invention, it is noted that for those skilled in the art For, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications are also considered as Protection scope of the present invention.

Claims (5)

1. a kind of robust target tracking method based on compressed sensing, it is characterised in that including step
S1：Target location is gone out to the first frame flag in any one section of video image；
S2：Sparse projection matrix is generated to the place target area of target location；
S3：Around target location to the target area, sample positive sample masterplate and negative sample masterplate；
S4：Multiple dimensioned normalization characteristic is constructed to the positive sample masterplate and negative sample masterplate by convolution masterplate, will be described many Dimension normalization feature is connected, and obtains a long column vector；
S5：The long column vector is compressed using the sparse projection matrix, obtains low-dimensional feature；
S6：Retraining is carried out to the low-dimensional feature, a Projection Character matrix is obtained, is selected by this feature projection matrix The high identification feature of positive sample and the high identification feature of negative sample；
S7：With the high identification feature of the positive sample and the high identification features training Bayes classifier of negative sample；
S8：Next frame picture is obtained, is sampled around the target location of previous frame, obtain a collection of candidate samples, fixed step Projection Character matrix in rapid S6, repeat step S4-S6 obtain the high identification feature of candidate samples；
S9：The high identification feature of the candidate samples is sent into into Bayes classifier, candidate's sample of positive maximum probability is categorized as It is target location that this is；
S10：Projection Character matrix in fixing step S6, repeat step S3-S7 update Bayes classifier；
S11：Repeat step S8-S11, until video terminates.
2. the robust target tracking method based on compressed sensing as claimed in claim 1, it is characterised in that step S6 includes step Suddenly
S61：Solution formula
m i n s | | A T s - p | | 2 2 + λ | | s | | 1 ,
WhereinIt is by npIndividual positive sample feature A+And nnIndividual negative sample feature A-Constitute, K is the dimension of feature, λ is Weight parameter, vectorIn element represent the property of the respective sample in training set A, i.e. ,+1 represents positive sample ,- 1 represents negative sample, draws Projection Character matrix；
S62：By A'=SA and x'=Sx, the high identification shadow feature of positive sample and negative sample of sample form and candidate samples are obtained This high identification feature.
3. the robust target tracking method based on compressed sensing as claimed in claim 1, it is characterised in that in step S5 The low-dimensional is characterized in that intrinsic dimensionality is 300 low-dimensional features.
4. the robust target tracking method based on compressed sensing as claimed in claim 1, it is characterised in that described in step S6 The high identification feature of positive sample and the high identification of negative sample are characterized in that dimension for the high identification feature of positive sample of 60-80 dimensions and bear The high identification feature of sample.
5. the robust target tracking method based on compressed sensing as claimed in claim 3, it is characterised in that the positive sample is high Identification feature and the high identification of negative sample are characterized in that the high identification feature of positive sample that dimension is 70 dimensions and the high differentiation of negative sample Property feature.
CN201611020862.6A 2016-11-21 2016-11-21 Compressed sensing-based robust target tracking method Pending CN106651912A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611020862.6A CN106651912A (en) 2016-11-21 2016-11-21 Compressed sensing-based robust target tracking method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611020862.6A CN106651912A (en) 2016-11-21 2016-11-21 Compressed sensing-based robust target tracking method

Publications (1)

Publication Number Publication Date
CN106651912A true CN106651912A (en) 2017-05-10

Family

ID=58808153

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611020862.6A Pending CN106651912A (en) 2016-11-21 2016-11-21 Compressed sensing-based robust target tracking method

Country Status (1)

Country Link
CN (1) CN106651912A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107844739A (en) * 2017-07-27 2018-03-27 电子科技大学 Robustness target tracking method based on adaptive rarefaction representation simultaneously
CN111860189A (en) * 2020-06-24 2020-10-30 北京环境特性研究所 Target tracking method and device

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103839273A (en) * 2014-03-25 2014-06-04 武汉大学 Real-time detection tracking frame and tracking method based on compressed sensing feature selection

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103839273A (en) * 2014-03-25 2014-06-04 武汉大学 Real-time detection tracking frame and tracking method based on compressed sensing feature selection

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
XU JIA等: "Visual Tracking via Adaptive Structural Local Sparse Appearance Model", 《2012 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 *
王振等: "基于视频序列的目标跟踪算法优化", 《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
CN107844739A (en) * 2017-07-27 2018-03-27 电子科技大学 Robustness target tracking method based on adaptive rarefaction representation simultaneously
CN107844739B (en) * 2017-07-27 2020-09-04 电子科技大学 Robust target tracking method based on self-adaptive simultaneous sparse representation
CN111860189A (en) * 2020-06-24 2020-10-30 北京环境特性研究所 Target tracking method and device
CN111860189B (en) * 2020-06-24 2024-01-19 北京环境特性研究所 Target tracking method and device

Similar Documents

Publication Publication Date Title
CN111062973B (en) Vehicle tracking method based on target feature sensitivity and deep learning
Tao et al. Smoke detection based on deep convolutional neural networks
CN111709909B (en) General printing defect detection method based on deep learning and model thereof
CN106778604B (en) Pedestrian re-identification method based on matching convolutional neural network
CN108416266B (en) Method for rapidly identifying video behaviors by extracting moving object through optical flow
CN103455797B (en) Detection and tracking method of moving small target in aerial shot video
Li et al. Adaptive deep convolutional neural networks for scene-specific object detection
CN107481264A (en) A kind of video target tracking method of adaptive scale
CN107688829A (en) A kind of identifying system and recognition methods based on SVMs
CN111639564B (en) Video pedestrian re-identification method based on multi-attention heterogeneous network
CN107808376B (en) Hand raising detection method based on deep learning
CN105160310A (en) 3D (three-dimensional) convolutional neural network based human body behavior recognition method
CN116448019B (en) Intelligent detection device and method for quality flatness of building energy-saving engineering
CN104537381B (en) A kind of fuzzy image recognition method based on fuzzy invariant features
CN112329771B (en) Deep learning-based building material sample identification method
CN112084952B (en) Video point location tracking method based on self-supervision training
CN110555377A (en) pedestrian detection and tracking method based on fisheye camera overlook shooting
CN111523586A (en) Noise-aware-based full-network supervision target detection method
CN116402769A (en) High-precision intelligent detection method for textile flaws considering size targets
CN114926826A (en) Scene text detection system
CN106651912A (en) Compressed sensing-based robust target tracking method
CN107291813B (en) Example searching method based on semantic segmentation scene
CN110570450B (en) Target tracking method based on cascade context-aware framework
CN108520539A (en) A kind of image object detection method based on sparse study variable model
Zhu et al. ID card number detection algorithm based on convolutional neural network

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: 20170510