CN104899567A - Small weak moving target tracking method based on sparse representation - Google Patents
Small weak moving target tracking method based on sparse representation Download PDFInfo
- Publication number
- CN104899567A CN104899567A CN201510306352.4A CN201510306352A CN104899567A CN 104899567 A CN104899567 A CN 104899567A CN 201510306352 A CN201510306352 A CN 201510306352A CN 104899567 A CN104899567 A CN 104899567A
- Authority
- CN
- China
- Prior art keywords
- target
- dictionary
- sparse
- background
- complete dictionary
- 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
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Probability & Statistics with Applications (AREA)
- Closed-Circuit Television Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A small weak moving target tracking method based on sparse representation comprises the following steps: acquiring the location of an infrared image target based on a detection algorithm, and constructing an initial training sample and an initial particle set; adopting a K-means singular value decomposition method K_SVD to learn the training sample and construct an adaptive morphological ingredient over-complete dictionary of the image, then, constructing an adaptive online classification over-complete dictionary, and carrying out real-time online updating; and finally, establishing a small weak target sparse representation observation model in a particle filter tracking framework, estimating the location of the target based on the size of sparse representation residual of a particle target image block and a particle background image block in the adaptive online classification dictionary, and keeping stable target tracking in subsequent frames through repeated iteration. The method of the invention not only overcomes the defect that it is difficult for an offline structure dictionary to sparsely represent a dynamically changing image signal and improves the difference between a signal and a background in representation sparseness, but also effectively improves the capability of infrared weak small target motion detection tracking.
Description
Technical field
The invention belongs to deep space Spacecraft TT&C field, be specifically related to detect infrared weak moving target detection,
Background technology
Moving object detection is a core technology of infrared imaging acquisition and tracking system, targeted surveillance system, satellite remote sensing system, safety check system etc., all can be widely used in all kinds of military, civilian system.In various imaging detection tracker, requirement can be intercepted and captured and locking tracking target as soon as possible.During distant between detector and target, target shows as the Small object only accounting for several pixels in imaging, and be easy to be submerged in various clutter background and very noisy, and lack color, structure, Texture eigenvalue, adopt common method for tracking target often cannot obtain good tracking effect, therefore, the new method studying the infrared little weak signal target tracking under Low SNR has extremely important meaning.
In recent years, the terseness of image sparse representation theory on signal represents and high efficiency, make it enjoy extensive concern.The signal method for expressing that it adopts the super complete dictionary of redundancy to replace traditional Fourier transform, wavelet transformation etc. to construct based on mathematics basis function, utilizes less atom in super complete dictionary can form optimum expression to signal.It is few and can represent on " the intrinsic atom " of respective essential characteristic and immanent structure that echo signal and background clutter are only concentrated on quantity by rarefaction representation respectively.How selecting base signal as the atom of dictionary, determine and adopt the rarefaction representation final effect of dealing with problems, is the most root problem of sparse signal representation model.Structure characterizes the adaptive morphology constituent structure dictionary of echo signal and background clutter, effectively enhances the difference degree between both rarefaction representation coefficients., characterize the atom of echo signal in this dictionary and represent that the atom of background clutter is mixed in together, representing that coefficient interpretation is poor, be difficult to tell echo signal from background clutter.But, existing method is all that the dictionary built by off-line learning mode is defied capture the echo signal of dynamic change and all states of yo-yo background clutter, cause structure dictionary and signal condition mismatch, power dissipation is adjacent and on the dictionary atom that correlativity is strong, non-null representation coefficient presents the form of aggregation block, and match tracing class and norm class algorithm performance will be deteriorated.
Summary of the invention
For the deficiency of off-line dictionary, the present invention utilizes the further on-line automatic classification of the atom of the super complete dictionary of Gauss to adaptive morphology Ingredient Dictionary, build the sparse representation model based on the infrared image signal of the super complete dictionary of adaptive online classification, to strengthen the feature difference of object and background for starting point, in conjunction with particle filter framework, a kind of little weak motion target tracking method based on rarefaction representation is proposed.
The present invention solves the problems of the technologies described above by the following technical solutions.
The present invention proposes a kind of little weak motion target tracking method based on rarefaction representation, relates to observation and control technology field.The present invention obtains infrared image target location by detection algorithm, builds initial training sample and primary collection; Use the super complete dictionary of adaptive morphology composition of K cluster singular value decomposition method K_SVD training sample design of graphics picture, then the on-line automatic classification of the Gauss's atom of super complete dictionary to adaptive morphology Ingredient Dictionary is utilized, the super complete dictionary of structure adaptive on-line classification, the i.e. super complete dictionary of target and the super complete dictionary of background, and adopt the random method estimated to carry out real-time online renewal to dictionary subspace at interval of 5 frames; Last under particle filter tracking framework, set up the rarefaction representation observation model based on the little weak signal target of adaptive on-line classification dictionary, particle target image block and the particle background image block sparse reconstructed residual size in adaptive on-line classification dictionary is utilized to come the next frame position of estimating target, and by the robust tracking of the realize target that iterates.
Performing step of the present invention mainly comprises:
(1) detection algorithm obtains infrared image target location, builds initial training sample and primary collection;
(2) initialization: adopt K cluster singular value decomposition method K_SVD learning training sample to build the super complete dictionary D of initial adaptive morphology composition of infrared image frame, set up initialization particle collection in initial target region
(3) the on-line automatic classification of the Gauss's atom of super complete dictionary to adaptive morphology Ingredient Dictionary is utilized, the super complete dictionary of structure adaptive on-line classification, the i.e. super complete dictionary D of target
tcomplete dictionary D super with background
b;
(4) transfering state of particle collection subsequent time is predicted according to the target movement model set up; Build the little weak signal target sparse representation model based on the super complete dictionary of adaptive on-line classification, and it can be used as the observation model in particle filter framework;
(5) adopt orthogonal matched jamming (OMP) algorithm that infrared image signal f is carried out Its Sparse Decomposition at the super complete dictionary D of adaptive on-line classification, obtain rarefaction representation factor alpha, β and the sparse reconstructed residual r of picture signal respectively in the super complete dictionary of target and the super complete dictionary of background
t(f), r
b(f), observation model value in calculation procedure (3) thus obtain particle weights;
(6) if the diversity factor of the particle target image block in tracing process and the particle background image block sparse reconstructed residual in the sparse territory of complete dictionary super based on this is greater than certain threshold value, then the movable information of this particle is preserved;
(7) accurate location of the little weak signal target of three of maximum weights intended particle predicting tracings is utilized;
(8) carry out online updating and reconstruction according to current up-to-date image information to the super complete dictionary of the online classification under current background, obtain the sparse reconstruct of the optimum of target image signal, upgrading iteration interval is 5 frames;
(9) if not meeting renewal iteration interval is 5 frames, then continue input picture and jump to step (4).
Described step (3) utilizes the super complete dictionary of Gauss to realize the on-line automatic classification of the atom of adaptive morphology Ingredient Dictionary, is specially: by super for adaptation form composition complete dictionary atom d
kat the super complete dictionary D of Gauss
gaussianin carry out Its Sparse Decomposition, be judged as target atoms or background atom by residual amount of energy.Atom d
kat Gauss's dictionary D
gaussianits Sparse Decomposition is expressed as:
for the kth of sparse matrix α arranges, after k sparse iteration, atom d
kresidual amount of energy r (d
k) be denoted as: r (d
k)=|| d
k-D
gaussianα ||
2, then by r (d
k) compare, as residual amount of energy r (d with threshold value δ
k) be greater than threshold value δ, then judge atom d
kfor background atom, on the contrary, d
kbe then target atoms, general threshold value δ is directly proportional to the size of atom; Finally, each atom in dictionary is judged, finally obtain the super complete dictionary D of target of automatic on-line classification
tcomplete dictionary D super with background
b.
The sparse model of infrared image signal f of the present invention is expressed as by the super complete dictionary of target and the super complete dictionary of background:
Wherein, D=[D
bd
t] represent comprise the super complete dictionary of target and the super complete dictionary of background combine super complete dictionary, γ=[α '
tβ '
t]
trepresent (a N
t+ N
b) vector tieed up, represent the rarefaction representation coefficient of the super complete dictionary of joint classification; If infrared image signal f is echo signal, then it can not by background dictionary rarefaction representation, and α ' should be null vector and β ' is a sparse vector; Similar, if f is background signal, then it can not by target dictionary rarefaction representation, and α ' should be sparse vector and β ' is a null vector.
Described step (4) is based on the little weak signal target sparse representation model of the super complete dictionary of adaptive on-line classification, and the target observation model namely under particle filter framework is:
Wherein, γ
tfor rarefaction representation coefficient gamma during t frame, D γ
tfor utilizing the rarefaction representation coefficient gamma in dictionary D
timage block after reconstruct, y-D γ
tfor the residual vector after reconstruct, (y-D γ
t) (i) be i-th component in residual vector, σ
sfor Gauss's variance, 0 < σ
s< 1, getting 0.5, n in reality is population.
Described step (5) adopts orthogonal matched jamming (OMP) Algorithm for Solving infrared image signal f at the super complete dictionary of adaptive on-line classification, namely the rarefaction representation factor alpha in target dictionary and background dictionary, β, then the Approximating Solutions solving its L1 norm minimum problem in certain allowable error σ is respectively:
Described step (4) adopts orthogonal matched jamming (OMP) Algorithm for Solving infrared image signal f at the super complete dictionary of adaptive on-line classification, and namely in target dictionary and background dictionary, sparse reconstructed residual is respectively
β
irepresent the sparse reconstruction coefficients of infrared image signal f in the super complete dictionary of target, α
irepresent the sparse reconstruction coefficients of infrared image signal f in the super complete dictionary of background, m is constant.
Described step (5) sparse reconstructed error size target function is defined as: D (f)=r
b(f)-r
t(f)
In formula, D (f) represents that signal reconstructs the diversity factor of rear error in target dictionary and background dictionary, and η is error threshold, as D (f) > η, then can judge that little weak objective image signal is echo signal, otherwise, be background signal.
In described step (8), employing is random estimates that (Stochastic Approximation) method upgrades dictionary subspace, upgrades and is spaced apart every 5 frames once: during note echo signal tracking former frame, adaptive morphology compositional classification dictionary is D
k-1, then the dictionary after upgrading is designated as D
k, by formula
upgrade, wherein, f
krepresent k time chart image signal, γ
krepresent that k time chart image signal is at D
k-1coefficient after Its Sparse Decomposition.
Described step (6), in target following identifying, takes limited Space domain sampling training sample, supposes that kth-1 frame has identified that the center of target is designated as L
k-1=(x
k-1, y
k-1), from meeting | L
target-L
k-1| the target image signal of the extracted region kth frame of < m, as training sample, upgrades target dictionary; At m < | L
target-L
k-1| select background training sample to upgrade background dictionary within the scope of < n, wherein m and n is sample radius.
Described step (7) t preserves observation sample corresponding to three particles of maximum weight
get
the mean center point position of three particles is as final output particle X
t_finaland demonstrating the region, target location of its correspondence, realize target is followed the tracks of.
The present invention adopts adaptive sparse classification to represent the tracking combined with particle filter, the shortcoming that off-line structure dictionary is difficult to the picture signal of sparse representation dynamic change can not only be overcome, improve the expression degree of rarefication difference of signal and background, effectively can also improve the motion detecting and tracking ability of infrared little weak signal target.
Accompanying drawing explanation
Fig. 1 is the FB(flow block) of the inventive method;
Fig. 2 is infrared image;
Fig. 3 is adaptive morphology Ingredient Dictionary;
Fig. 4 is Gauss's dictionary;
Fig. 5 (a) and Fig. 5 (b) is adaptive on-line classification dictionary;
Fig. 6 (a) and Fig. 6 (b) is echo signal rarefaction representation coefficient in the super complete dictionary of adaptive on-line classification;
Fig. 7 (a) and Fig. 7 (b) is background signal rarefaction representation coefficient in the super complete dictionary of adaptive on-line classification;
Fig. 8 (1)-Fig. 8 (6) is tracking results.
Embodiment
In order to understand technical scheme of the present invention better, below in conjunction with accompanying drawing, embodiments of the present invention are further described.Fig. 1 is the FB(flow block) of the little weak motion target tracking method that the present invention is based on rarefaction representation.Detection algorithm obtains infrared image target location, builds initial training sample and primary collection; Use the super complete dictionary of adaptive morphology composition of K cluster singular value decomposition method K_SVD training sample design of graphics picture, then the on-line automatic classification of the Gauss's atom of super complete dictionary to adaptive morphology Ingredient Dictionary is utilized, the super complete dictionary of structure adaptive on-line classification, the i.e. super complete dictionary of target and the super complete dictionary of background, and adopt the random method estimated to carry out real-time online renewal to dictionary subspace at interval of 5 frames; Then under particle filter tracking framework, set up the little weak signal target rarefaction representation observation model based on adaptive on-line classification dictionary, utilize particle target image block and the particle background image block sparse reconstructed residual size in adaptive on-line classification dictionary to come the next frame position of estimating target, and keep the tenacious tracking to the target in subsequent frame by iterating.
The concrete implementation detail of each several part is discussed in detail as follows for Infrared cloud image:
Obtain infrared image target location by detection algorithm as seen from Figure 2, build initial training sample and primary collection; White box is by being drawn a circle to approve target initial position.
1. the super complete dictionary of adaptive morphology composition
Adopt K average singular value decomposition method (K-SVD) from the form of a large amount of training sample learning target and background and motion feature, build when can adapt to dynamic object and background empty and cross complete destructuring dictionary.Space-time joint crosses the training pattern following [7] of complete dictionary D:
in formula, || ||
0with || ||
2represent respectively
with
by residual energy
under certain condition, image sequence signal F can be reconstructed by a small amount of atom and sparse coefficient matrix γ thereof in dictionary D during excessively complete sky in restriction.Crossing complete dictionary when setting up empty is an iterative process, and each iteration comprises two stages, and namely sparse coding and dictionary upgrade.
(1) sparse coding: first given initial dictionary D and infrared sequence image F, adopts the sparse coefficient γ of orthogonal matching pursuit algorithm sequence of calculation image in dictionary, namely solves
in formula, ε is the approximate error limited.
(2) dictionary D upgrades: each row upgrading dictionary D, namely upgrade an atom d
k.Atom d
kthe error of Approximate Sequence figure F is
sVD is utilized to decompose
obtain one group of best approximation solution (d
k, γ
k).Formula is repeated to each atom
until
upgrade all atoms and namely complete a dictionary updating.Along with the increase of iterations, sparse bayesian learning error exponentially is decayed, and namely upgrades through iteration several times, can train the adaptive morphology composition adapted with infrared sequence image F cross complete empty time sparse dictionary D.
Fig. 3 is the super complete dictionary of adaptive morphology composition using K cluster singular value decomposition method K_SVD training sample design of graphics picture.
2. the super complete dictionary of adaptive on-line classification
Utilize the on-line automatic classification of Fig. 4 Gauss atom of super complete dictionary to adaptive morphology Ingredient Dictionary, be specially: by super for adaptation form composition complete dictionary atom d
kat the super complete dictionary D of Gauss
gaussianin carry out Its Sparse Decomposition, be judged as target atoms or background atom by residual amount of energy.Atom d
kat Gauss's dictionary D
gaussianits Sparse Decomposition is expressed as:
s.t.|| α ||
0=k, after k sparse iteration, atom d
kresidual amount of energy r (d
k) be denoted as: r (d
k)=|| d
k-D
gaussianα ||
2, then by r (d
k) compare, as residual amount of energy r (d with threshold value δ
k) be greater than threshold value δ, then judge atom d
kfor background atom, on the contrary, d
kbe then target atoms, general threshold value δ is directly proportional to the size of atom; Finally, each atom in dictionary is judged, finally obtain the super complete dictionary D of target of automatic on-line classification
tcomplete dictionary D super with background
b.
Fig. 5 (a) and Fig. 5 (b) is for utilizing the super complete dictionary of the Gauss in Fig. 4 to the on-line automatic classification of atom of adaptive morphology Ingredient Dictionary in Fig. 3, the super complete dictionary of structure adaptive on-line classification, i.e. the super complete dictionary of target and the super complete dictionary of background.
The sparse model of infrared image signal f is expressed as by the super complete dictionary of target and the super complete dictionary of background:
wherein, D=[D
bd
t] represent comprise the super complete dictionary of target and the super complete dictionary of background combine super complete dictionary, γ=[α '
tβ '
t]
trepresent (a N
t+ N
b) vector tieed up, represent the rarefaction representation coefficient of the super complete dictionary of joint classification; If infrared image signal f is echo signal, then it can not by background dictionary rarefaction representation, and α ' should be null vector and β ' is a sparse vector; Similar, if f is background signal, then it can not by target dictionary rarefaction representation, and α ' should be sparse vector and β ' is a null vector.
Fig. 6 (a) and Fig. 6 (b) for echo signal is respectively at Fig. 5 (a), rarefaction representation coefficient in the super complete dictionary of adaptive on-line classification in (b);
Fig. 7 (a) and Fig. 7 (b) for background signal is respectively at Fig. 5 (a), rarefaction representation coefficient in the super complete dictionary of adaptive on-line classification in (b).
3. particle filter
Particle filter algorithm flow process is as follows:
(1) initialization: as t=0, according to prior imformation p (x
0) stochastic sampling, generate initialization particle state variables set
select the importance density function q (x
t| x
0:t-1, y
1:t).
(2) importance sampling: the transfering state predicting particle collection according to the motion model of target and formula (4.1), namely by the state in a upper moment
obtain new particle collection
(3) right value update and normalization: obtain observed reading corresponding to each particle of particle set according to the observation model of target
calculate the weights of each particle
and normalization weights
(4) resampling: if N
effduring≤N, resampling is carried out to particle, otherwise terminate.
4. set up the little weak signal target rarefaction representation observation model based on adaptive on-line classification dictionary
Utilize background image block and the degree of rarefication otherness of target image block in the super complete classifying dictionary of self-adaptation, adopt maximum 5-10 the sparse decomposition coefficients reconstructed image block of testing image signal in adaptive on-line classification dictionary, the residual energy after reconstruct is as the criterion of similarity.Suppose when the i-th frame, all reconstructed residual, noise obeys independent distribution, and noise is white Gaussian noise, then the target observation model under particle filter framework is
obtain thus: residual energy value is less, then the probable value in target observation model is larger, illustrates that candidate image signal is more similar to the atom signals in super complete dictionary, and so candidate image signal is that the probability of echo signal to be tracked is larger.Based in the particle filter tracking algorithm of this target observation model, i-th component (y-D γ in residual vector
t) (i) represent i-th candidate's particle, the mean place choosing maximum probability 3 particles is the tracking position of object of present frame.
Fig. 8 (1)-Fig. 8 (6) is under particle filter tracking framework, set up the rarefaction representation observation model based on the little weak signal target of adaptive on-line classification dictionary, particle target image block and the particle background image block sparse reconstructed residual size in adaptive on-line classification dictionary is utilized to come the next frame position of estimating target, and by the robust tracking of the realize target that iterates.
Claims (10)
1., based on a little weak motion target tracking method for rarefaction representation, it is characterized in that, comprise step:
(1) detection algorithm obtains infrared image target location, builds initial training sample and primary collection;
(2) initialization: adopt K cluster singular value decomposition method K_SVD learning training sample to build the super complete dictionary D of initial adaptive morphology composition of infrared image frame, set up initialization particle collection in initial target region
(3) the on-line automatic classification of the Gauss's atom of super complete dictionary to adaptive morphology Ingredient Dictionary is utilized, the super complete dictionary of structure adaptive on-line classification, the i.e. super complete dictionary D of target
tcomplete dictionary D super with background
b;
(4) transfering state of particle collection subsequent time is predicted according to the target movement model set up; Build the little weak signal target sparse representation model based on the super complete dictionary of adaptive on-line classification, and it can be used as the observation model in particle filter framework;
(5) adopt orthogonal matched jamming (OMP) algorithm that infrared image signal f is carried out Its Sparse Decomposition at the super complete dictionary D of adaptive on-line classification, obtain rarefaction representation factor alpha, β and the sparse reconstructed residual r of picture signal respectively in the super complete dictionary of target and the super complete dictionary of background
t(f), r
b(f), observation model value in calculation procedure (4) thus obtain particle weights;
(6) if the diversity factor of the particle target image block in tracing process and the particle background image block sparse reconstructed residual in the sparse territory of complete dictionary super based on this is greater than certain threshold value, then the movable information of this particle is preserved;
(7) accurate location of the little weak signal target of three of maximum weights intended particle predicting tracings is utilized;
(8) carry out online updating and reconstruction according to current up-to-date image information to the super complete dictionary of the online classification under current background, obtain the sparse reconstruct of the optimum of target image signal, upgrading iteration interval is 5 frames;
(9) if not meeting renewal iteration interval is 5 frames, then continue input picture and jump to step (4).
2. the little weak motion target tracking method based on rarefaction representation according to claim 1, it is characterized in that described step (3) utilizes the super complete dictionary of Gauss to realize the on-line automatic classification of the atom of adaptive morphology Ingredient Dictionary, be specially: by super for adaptation form composition complete dictionary atom d
kat the super complete dictionary D of Gauss
gaussianin carry out Its Sparse Decomposition, be judged as target atoms or background atom by residual amount of energy.Atom d
kat Gauss's dictionary D
gaussianits Sparse Decomposition is expressed as:
for the kth of sparse matrix α arranges, after k sparse iteration, atom d
kresidual amount of energy r (d
k) be denoted as: r (d
k)=|| d
k-D
gaussianα ||
2, then by r (d
k) compare, as residual amount of energy r (d with threshold value δ
k) be greater than threshold value δ, then judge atom d
kfor background atom, on the contrary, d
kbe then target atoms, general threshold value δ is directly proportional to the size of atom; Finally, each atom in dictionary is judged, finally obtain the super complete dictionary D of target of automatic on-line classification
tcomplete dictionary D super with background
b.
3. the little weak motion target tracking method based on rarefaction representation according to claim 1, is characterized in that, the sparse model of infrared image signal f is expressed as by the super complete dictionary of target and the super complete dictionary of background:
Wherein, D=[D
bd
t] represent comprise the super complete dictionary of target and the super complete dictionary of background combine super complete dictionary, γ=[α '
tβ '
t]
trepresent (a N
t+ N
b) vector tieed up, represent the rarefaction representation coefficient of the super complete dictionary of joint classification; If infrared image signal f is echo signal, then it can not by background dictionary rarefaction representation, and α ' should be null vector and β ' is a sparse vector; Similar, if f is background signal, then it can not by target dictionary rarefaction representation, and α ' should be sparse vector and β ' is a null vector.
4. the little weak motion target tracking method based on rarefaction representation according to claim 1, it is characterized in that the little weak signal target sparse representation model of described step (4) based on the super complete dictionary of adaptive on-line classification, the target observation model namely under particle filter framework is:
Wherein, γ
tfor rarefaction representation coefficient gamma during t frame, D γ
tfor utilizing the rarefaction representation coefficient gamma in dictionary D
timage block after reconstruct, y-D γ
tfor the residual vector after reconstruct, (y-D γ
t) (i) be i-th component in residual vector, σ
sfor Gauss's variance, 0 < σ
s< 1, getting 0.5, n in reality is population.
5. the little weak motion target tracking method based on rarefaction representation according to claim 1, it is characterized in that described step (5) adopts orthogonal matched jamming (OMP) Algorithm for Solving infrared image signal f at the super complete dictionary of adaptive on-line classification, namely the rarefaction representation factor alpha in target dictionary and background dictionary, β, then the Approximating Solutions solving its L1 norm minimum problem in certain allowable error σ is respectively:
6. the little weak motion target tracking method based on rarefaction representation according to claim 1, it is characterized in that described step (5) adopts orthogonal matched jamming (OMP) Algorithm for Solving infrared image signal f at the super complete dictionary of adaptive on-line classification, namely in target dictionary and background dictionary, sparse reconstructed residual is respectively
β
irepresent the sparse reconstruction coefficients of infrared image signal f in the super complete dictionary of target, α
irepresent the sparse reconstruction coefficients of infrared image signal f in the super complete dictionary of background, m is constant.
7. the little weak motion target tracking method based on rarefaction representation according to claim 1, is characterized in that described step (6) sparse reconstructed error size target function is defined as: D (f)=r
b(f)-r
t(f)
In formula, D (f) represents that signal reconstructs the diversity factor of rear error in target dictionary and background dictionary, and η is error threshold, as D (f) > η, then can judge that little weak objective image signal is echo signal, otherwise, be background signal.
8. the little weak motion target tracking method based on rarefaction representation according to claim 1, it is characterized in that in described step (8), the random estimation of employing (Stochastic Approximation) method upgrades dictionary subspace, upgrades and is spaced apart every 5 frames once: during note echo signal tracking former frame, adaptive morphology compositional classification dictionary is D
k-1, then the dictionary after upgrading is designated as D
k, by formula
upgrade, wherein, f
krepresent k time chart image signal, γ
krepresent that k time chart image signal is at D
k-1coefficient after Its Sparse Decomposition.
9. the little weak motion target tracking method based on rarefaction representation according to claim 1, it is characterized in that described step (6) is in target following identifying, take limited Space domain sampling training sample, suppose that kth-1 frame has identified that the center of target is designated as L
k-1=(x
k-1, y
k-1), from meeting | L
target-L
k-1| the target image signal of the extracted region kth frame of < m, as training sample, upgrades target dictionary; At m < | L
target-L
k-1| select background training sample to upgrade background dictionary within the scope of < n, wherein m and n is sample radius.
10. the little weak motion target tracking method based on rarefaction representation according to claim 1, is characterized in that the observation sample that three particles of described step (7) t preservation maximum weight are corresponding
get
the mean center point position of three particles is as final output particle X
t_finaland demonstrating the region, target location of its correspondence, realize target is followed the tracks of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510306352.4A CN104899567A (en) | 2015-06-05 | 2015-06-05 | Small weak moving target tracking method based on sparse representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510306352.4A CN104899567A (en) | 2015-06-05 | 2015-06-05 | Small weak moving target tracking method based on sparse representation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104899567A true CN104899567A (en) | 2015-09-09 |
Family
ID=54032223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510306352.4A Pending CN104899567A (en) | 2015-06-05 | 2015-06-05 | Small weak moving target tracking method based on sparse representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104899567A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678787A (en) * | 2016-02-03 | 2016-06-15 | 西南交通大学 | Heavy-duty lorry driving barrier detection and tracking method based on binocular fisheye camera |
| CN106203291A (en) * | 2016-07-01 | 2016-12-07 | 昆明理工大学 | A kind of method based on anatomic element analysis with the scene image words detection of self-adapting dictionary study |
| CN106408019A (en) * | 2016-09-14 | 2017-02-15 | 南京信息工程大学 | Adaptive optical celestial target detection method on strong skylight background |
| CN106971176A (en) * | 2017-05-10 | 2017-07-21 | 河海大学 | Tracking infrared human body target method based on rarefaction representation |
| CN107330912A (en) * | 2017-05-10 | 2017-11-07 | 南京邮电大学 | A kind of target tracking method of rarefaction representation based on multi-feature fusion |
| CN107564029A (en) * | 2017-07-24 | 2018-01-09 | 南京信息工程大学 | Moving target detecting method based on the filtering of Gauss extreme value and the sparse RPCA of group |
| CN108229505A (en) * | 2018-02-05 | 2018-06-29 | 南京邮电大学 | Image classification method based on FISHER multistage dictionary learnings |
| CN105809643B (en) * | 2016-03-14 | 2018-07-06 | 浙江外国语学院 | A kind of image enchancing method based on adaptive block channel extrusion |
| CN108685570A (en) * | 2017-04-12 | 2018-10-23 | 中国科学院微电子研究所 | A kind of processing method of excessively complete dictionary, apparatus and system |
| CN109375205A (en) * | 2018-09-28 | 2019-02-22 | 清华大学 | Multiple types unmanned plane scene recognition method dictionary-based learning and device |
| CN110298865A (en) * | 2019-05-22 | 2019-10-01 | 西华大学 | The space-based Celestial Background small point target tracking of cluster device is separated based on threshold value |
| CN110632566A (en) * | 2019-08-31 | 2019-12-31 | 南京理工大学 | Radio fuse foil strip interference resisting method based on sparse representation |
| CN110677363A (en) * | 2019-10-28 | 2020-01-10 | 重庆邮电大学 | Multi-user detection method and device based on compressed sensing under MUSA (multi user application architecture) system |
| CN112114300A (en) * | 2020-09-14 | 2020-12-22 | 哈尔滨工程大学 | Underwater weak target detection method based on image sparse representation |
| CN113256687A (en) * | 2021-06-29 | 2021-08-13 | 西南石油大学 | Online video multi-target tracking method |
| CN114463619A (en) * | 2022-04-12 | 2022-05-10 | 西北工业大学 | Infrared dim target detection method based on integrated fusion features |
Citations (3)
| 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 |
| US8620093B2 (en) * | 2010-03-15 | 2013-12-31 | The United States Of America As Represented By The Secretary Of The Army | Method and system for image registration and change detection |
| CN104361609A (en) * | 2014-11-18 | 2015-02-18 | 电子科技大学 | Target tracking method based on sparse representation |
-
2015
- 2015-06-05 CN CN201510306352.4A patent/CN104899567A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8620093B2 (en) * | 2010-03-15 | 2013-12-31 | The United States Of America As Represented By The Secretary Of The Army | Method and system for image registration and change detection |
| CN103440645A (en) * | 2013-08-16 | 2013-12-11 | 东南大学 | Target tracking algorithm based on self-adaptive particle filter and sparse representation |
| CN104361609A (en) * | 2014-11-18 | 2015-02-18 | 电子科技大学 | Target tracking method based on sparse representation |
Non-Patent Citations (3)
| Title |
|---|
| YUAN XIE等: "Discriminative Object Tracking via Sparse", 《IEEE TRANSACTIONS ON CYBERNETICS》 * |
| ZHENG-ZHOU LI等: "Sparse Representation for Infrared Dim Target Detection via a Discriminative Over-Complete Dictionary Learned Online", 《SENSORS》 * |
| 唐峥远等: "基于稀疏表示模型的红外目标跟踪算法", 《红外与激光工程》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678787A (en) * | 2016-02-03 | 2016-06-15 | 西南交通大学 | Heavy-duty lorry driving barrier detection and tracking method based on binocular fisheye camera |
| CN105809643B (en) * | 2016-03-14 | 2018-07-06 | 浙江外国语学院 | A kind of image enchancing method based on adaptive block channel extrusion |
| CN106203291A (en) * | 2016-07-01 | 2016-12-07 | 昆明理工大学 | A kind of method based on anatomic element analysis with the scene image words detection of self-adapting dictionary study |
| CN106203291B (en) * | 2016-07-01 | 2019-07-05 | 昆明理工大学 | A method of it is detected based on anatomic element analysis and the scene image words of self-adapting dictionary study |
| CN106408019A (en) * | 2016-09-14 | 2017-02-15 | 南京信息工程大学 | Adaptive optical celestial target detection method on strong skylight background |
| CN108685570B (en) * | 2017-04-12 | 2021-01-22 | 中国科学院微电子研究所 | Method, device and system for processing over-complete dictionary |
| CN108685570A (en) * | 2017-04-12 | 2018-10-23 | 中国科学院微电子研究所 | A kind of processing method of excessively complete dictionary, apparatus and system |
| CN106971176A (en) * | 2017-05-10 | 2017-07-21 | 河海大学 | Tracking infrared human body target method based on rarefaction representation |
| CN107330912A (en) * | 2017-05-10 | 2017-11-07 | 南京邮电大学 | A kind of target tracking method of rarefaction representation based on multi-feature fusion |
| CN107330912B (en) * | 2017-05-10 | 2021-06-11 | 南京邮电大学 | Target tracking method based on sparse representation of multi-feature fusion |
| CN107564029A (en) * | 2017-07-24 | 2018-01-09 | 南京信息工程大学 | Moving target detecting method based on the filtering of Gauss extreme value and the sparse RPCA of group |
| CN107564029B (en) * | 2017-07-24 | 2021-09-03 | 南京信息工程大学 | Moving target detection method based on Gaussian extreme value filtering and group sparse RPCA |
| CN108229505A (en) * | 2018-02-05 | 2018-06-29 | 南京邮电大学 | Image classification method based on FISHER multistage dictionary learnings |
| CN108229505B (en) * | 2018-02-05 | 2022-02-18 | 南京邮电大学 | Image classification method based on FISER multi-level dictionary learning |
| CN109375205A (en) * | 2018-09-28 | 2019-02-22 | 清华大学 | Multiple types unmanned plane scene recognition method dictionary-based learning and device |
| CN110298865A (en) * | 2019-05-22 | 2019-10-01 | 西华大学 | The space-based Celestial Background small point target tracking of cluster device is separated based on threshold value |
| CN110298865B (en) * | 2019-05-22 | 2023-07-07 | 深空探测科技(北京)有限责任公司 | Space-based starry sky background weak small point target tracking method based on threshold separation clustering device |
| CN110632566A (en) * | 2019-08-31 | 2019-12-31 | 南京理工大学 | Radio fuse foil strip interference resisting method based on sparse representation |
| CN110677363A (en) * | 2019-10-28 | 2020-01-10 | 重庆邮电大学 | Multi-user detection method and device based on compressed sensing under MUSA (multi user application architecture) system |
| CN112114300A (en) * | 2020-09-14 | 2020-12-22 | 哈尔滨工程大学 | Underwater weak target detection method based on image sparse representation |
| CN112114300B (en) * | 2020-09-14 | 2022-06-21 | 哈尔滨工程大学 | Underwater weak target detection method based on image sparse representation |
| CN113256687A (en) * | 2021-06-29 | 2021-08-13 | 西南石油大学 | Online video multi-target tracking method |
| CN113256687B (en) * | 2021-06-29 | 2022-02-22 | 西南石油大学 | Online video multi-target tracking method |
| CN114463619A (en) * | 2022-04-12 | 2022-05-10 | 西北工业大学 | Infrared dim target detection method based on integrated fusion features |
| CN114463619B (en) * | 2022-04-12 | 2022-07-08 | 西北工业大学 | Infrared dim target detection method based on integrated fusion features |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104899567A (en) | Small weak moving target tracking method based on sparse representation | |
| CN106355151B (en) | A kind of three-dimensional S AR images steganalysis method based on depth confidence network | |
| CN111046967A (en) | Underwater image classification method based on convolutional neural network and attention mechanism | |
| CN112395987B (en) | SAR image target detection method based on unsupervised domain adaptive CNN | |
| CN109584303B (en) | Infrared weak and small target detection method based on Lp norm and nuclear norm | |
| CN103955701B (en) | Multi-level-combined multi-look synthetic aperture radar image target recognition method | |
| CN103077506B (en) | In conjunction with local and non-local adaptive denoising method | |
| CN111160268A (en) | Multi-angle SAR target recognition method based on multi-task learning | |
| CN103473755B (en) | Based on the sparse denoising method of SAR image that change detects | |
| CN107133648B (en) | One-dimensional range profile identification method based on adaptive multi-scale fusion sparse preserving projection | |
| CN106971189B (en) | A kind of noisy method for recognising star map of low resolution | |
| Du et al. | Robust statistical recognition and reconstruction scheme based on hierarchical Bayesian learning of HRR radar target signal | |
| CN107742133A (en) | A kind of sorting technique for Polarimetric SAR Image | |
| CN110135344A (en) | Based on the method for detecting infrared puniness target for weighting fixed order expression | |
| CN108765313A (en) | High spectrum image denoising method based on low-rank representation in class | |
| CN110333489A (en) | The processing method to SAR echo data Sidelobe Suppression is combined with RSVA using CNN | |
| CN104680182B (en) | Polarization SAR sorting technique based on NSCT and differentiation dictionary learning | |
| Yu et al. | Application of a convolutional autoencoder to half space radar hrrp recognition | |
| CN114428234A (en) | Radar high-resolution range profile noise reduction identification method based on GAN and self-attention | |
| CN101833752B (en) | Pretreatment method for decomposed and reconstituted infrared small targets based on singular values | |
| CN106952287A (en) | A kind of video multi-target dividing method expressed based on low-rank sparse | |
| Zhang et al. | Distributed compressive sensing via LSTM-Aided sparse Bayesian learning | |
| CN111144316B (en) | Sea surface target detection method based on stack self-encoder | |
| Li et al. | Automatic modulation recognition based on a new deep K-SVD denoising algorithm | |
| Tu et al. | A novel HRRP target recognition method based on LSTM and HMM decision-making |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| CB03 | Change of inventor or designer information |
Inventor after: Li Zhengzhou Inventor after: Jin Gang Inventor after: Cao Dong Inventor after: Liu Depeng Inventor after: Fu Hongxia Inventor after: Shao Wanxing Inventor after: Chen Jing Inventor before: Li Zhengzhou Inventor before: Fu Hongxia Inventor before: Liu Depeng Inventor before: Li Jianing Inventor before: Shao Wanxing Inventor before: Chen Jing |
|
| COR | Change of bibliographic data | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20170104 Address after: 400030 Shapingba District, Sha Sha Street, No. 174, Chongqing Applicant after: Chongqing Univ Applicant after: 63821 people's Liberation Army Address before: 400030 Shapingba District, Sha Sha Street, No. 174, Chongqing Applicant before: Chongqing Univ |
|
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20150909 |
|
| RJ01 | Rejection of invention patent application after publication |