CN110046608A - The leaf for identifying dictionary learning based on half coupling blocks pedestrian recognition methods and system again - Google Patents
The leaf for identifying dictionary learning based on half coupling blocks pedestrian recognition methods and system again Download PDFInfo
- Publication number
- CN110046608A CN110046608A CN201910344098.5A CN201910344098A CN110046608A CN 110046608 A CN110046608 A CN 110046608A CN 201910344098 A CN201910344098 A CN 201910344098A CN 110046608 A CN110046608 A CN 110046608A
- Authority
- CN
- China
- Prior art keywords
- video
- dictionary
- pedestrian
- leaf
- feature
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to image retrieval technologies fields, it discloses a kind of leaf that dictionary learning is identified based on half coupling and blocks pedestrian recognition methods and system again, the pedestrian public from two identifies that true branches and leaves are added on data set to be blocked again, includes to block video and ordinary video；Then feature extraction is carried out to blocking video and ordinary video respectively；Followed by the sample characteristics extracted are handled, dictionary learning method is introduced, learns projection matrix from blocking in video and ordinary video；It introduces and identifies thought study dictionary pair.The present invention learns half coupled maps matrix and identifies dictionary pair, and half coupled maps matrix can compensate the difference blocked between video and ordinary video, identifies dictionary to the same person can be made more compact in different cameral, different people separates in different cameras；The results show on 2 public data collection, the method proposed have better recognition performance.
Description
Technical field
The invention belongs to image retrieval technologies fields more particularly to a kind of leaf for identifying dictionary learning based on half coupling to block
Pedestrian recognition methods and system again.
Background technique
Currently, the immediate prior art:
Pedestrian identifies again and plays an important role in video monitoring and smart city, has also obtained in recent years extensive
Research.The purpose that pedestrian identifies again is exactly to retrieve the image of the same person from another image set to identify.Mostly
Counting method solves the problems, such as that the pedestrian in normal scene identifies again to a certain extent, but in actual scene, there are illumination, view
It angle and blocks, does not especially block camera without trimming in time in summer branches and leaves, a people passes through camera shooting whithin a period of time
Head is blocked, this people is not blocked in other times section by camera, and in this case, detection video, which is had, to be blocked
Video camera capture, and picture library video is captured by common camera.Pedestrian under this scene is identified that problem is called by the present invention again
The pedestrian blocked based on video branches and leaves identifies problem again.Video or blocking for image will will lead to visual appearance feature and space-time
The loss of the effective informations such as feature, most of existing methods are suitable for normal scene, block under scene for this
Problem not can solve.
However the pedestrian based on video shelter identifies that problem is also a kind of universal and important application again, research at present is still not
It is common.Although there are some research light to block the pedestrian blocked with local body recognition methods again, it is concentrated mainly on based on image
In recognition methods.But not only include visual appearance information in video but also include space time information, these information are for base
It identifies and is very effective again in the pedestrian of video.
In conclusion problem of the existing technology is:
(1) in the prior art, to including visual space-time information in video, identification is not can be carried out.Branches and leaves screening is not can solve
The pedestrian of gear identify again in problem.
(2) data set for not having pedestrian to identify again under scene is blocked, therefore the present invention has remake 2 common datas
Collection is simulated on PRID 2011 and iLIDS-VID with really template (true branches and leaves) is blocked, and forms 2 new data
Collect LO-PRID 2011 and LO-iLIDS-VID.
Solve the difficulty of above-mentioned technical problem:
Most methods solve the problems, such as that the pedestrian in normal scene identifies again to a certain extent, but actual scene
In, it there are illumination, visual angle and blocks, especially blocks camera without trimming in time in summer branches and leaves, a people is at one section
It being blocked in time by camera, this people is not blocked in other times section by camera, in this case, detection
Video is had the video camera blocked capture, and picture library video is captured by common camera.
Solve the meaning of above-mentioned technical problem:
Pedestrian under this scene blocked by branches and leaves is identified that problem is called again and is blocked based on video branches and leaves by the present invention
Pedestrian identifies problem again.Video or blocking for image will will lead to losing for the effective informations such as visual appearance feature and space-time characteristic
It loses, most of existing methods are suitable for normal scene, this problem blocked under scene not can solve.
The present invention will propose that a kind of leaf for identifying dictionary learning based on half coupling blocks pedestrian again and recognition methods and is in view of the above problems
System.
Summary of the invention
In view of the problems of the existing technology, the present invention provides a kind of leaves for identifying dictionary learning based on half coupling to block
Pedestrian recognition methods and system again.
The invention is realized in this way a kind of leaf for identifying dictionary learning based on half coupling blocks pedestrian's recognition methods packet again
It includes:
It is that data set is collected first, the pedestrian public from two identifies that true branches and leaves are added on data set to be blocked again, wraps
It has included and has blocked video and ordinary video；Then feature extraction is carried out to blocking video and ordinary video respectively；Followed by mentioning
The sample characteristics taken out are handled, here for throwing in the feature set study collection class of each user in two cameras
Shadow；Present invention introduces dictionary learning method, the dictionary pair of video and ordinary video is blocked in study；The present invention is from blocking video again
With learn projection matrix in ordinary video；Finally present invention introduces identifying thought to learn dictionary pair, it is based on to realize one kind
The leaf that half coupling identifies dictionary learning blocks pedestrian's weight recognizer.The advantage of the invention is that solving the row that branches and leaves block
People problem and has been put forward for the first time the leaf for identifying dictionary learning based on half coupling and blocks pedestrian and identify (SCD again identify again in2L) skill
Art.The half coupled maps matrix of technological learning and identify dictionary pair, half coupled maps matrix, which can compensate, blocks video and common
Difference between video identifies dictionary to the same person can be made more compact in different cameral, and different people is in different phases
It is separated in machine.The results show on 2 public data collection, the method proposed have better recognition performance.
Further, the leaf for identifying dictionary learning based on half coupling blocks pedestrian's recognition methods again, comprising the following steps:
Step (1): data are collected and establish database, are formed by blocking video and ordinary video；
Step (2): feature is extracted to blocking video and ordinary video respectively；
Step (3): the dictionary pair with half coupled maps is designed；
Step (4): learn dictionary pair using thought is identified；
Step (5): study blocks the aggregate projection matrix W of video and does not block the aggregate projection matrix V of video respectively；
Step (6): catalogue scalar functions are obtained；
Step (7): algorithm optimization is carried out to objective function；
Step (8): the pedestrian blocked under video identifies again.
Further, collection data described in step (1), which establish the specific method of database, is:
The invention solves problem is identified again in the pedestrian blocked in scene and normal scene, due in the case where blocking scene
There is no the data set that pedestrian identifies again, therefore the present invention has remake 2 common data sets, it is (true with template is really blocked
Real branches and leaves) it is simulated on PRID 2011 and iLIDS-VID, form 2 new data set LO-PRID 2011 and LO-
iLIDS-VID.749 people of 385 people of A video camera and B video camera is shared in 2011 data set of PRID, has 200 in 2 visual angles
Individual occurs, and each video is made of 5 to 675 picture frames.In order to guarantee the effective length of each walking period, present invention choosing
178 people of 20 frames or more are taken, therefrom 89 pairs is randomly choosed and is trained, and using remaining to testing.The present invention uses
Template is blocked to block the addition of PRID 2011 to generate 2011 data set of Lo-PRID.Likewise, 2 of iLIDS-VID image
300 couples of pedestrians are shared in head, picture frame is from 23 to 192, and average frame number is 73 frames, and the present invention covers iLIDS- with template is blocked
VID data set, Lai Shengcheng LO-iLIDS-VID data set.Simulation process and the process phase for generating 2011 data set of Lo-PRID
Seemingly.The data set that the band ultimately produced blocks can be from https: //sites.google.com/site/LODB is obtained.
Further, in step (2), respectively to video and ordinary video extraction feature is blocked, for blocking video, this hair
It is bright to pass through FEP (Wang et al.2014；Liu et al.2015) walking cycle feature is extracted, the present invention is according to normal view
The average time of walking cycle in frequency has empirically selected 20 frames as a walking period for blocking video.For commonly regarding
Frequently, the present invention is directly that each user extracts STFV3D feature.
Further, in step (3), the dictionary pair with half coupled maps is designed.For identical people, different camera shootings
There are different postures and viewpoints to change for machine, but block and normal video there are similitudes on visual appearance and space-time characteristic.
Relax equal strong in the two spaces it is assumed that the present invention considers one and half coupled maps matrixes, to reflect between character pair
Relationship.For this purpose, the present invention devises the dictionary pair with half coupled maps, it is following to indicate:
s.t.Ai=DOXi；Bi=DNYi
Wherein Erepresent() indicates that sub- dictionary indicates fidelity term, and X and Y are A and B respectively in DOAnd DNOn expression system
Number, Emapping() is mapping fidelity, and the purpose is to find the relationship between A and the code coefficient of B, Φ () is to block
With the coefficient mapping function of common dictionary pair.A=[A1, A2...AN] it is the space-time characteristic for blocking training video,It is the video features subset of i-th of people, niIt is the walking cycle number of first man, N is the number of people of being blocked
Amount, B=[B1, B2...BN] it is the space-time characteristic collection for not blocking training video, whereinIt is and njA step
The space-time characteristic collection of row period corresponding j-th of people, N are the numbers in ordinary video.
Further, in step (4), learn dictionary pair using thought is identified；In order to improve the identification of rarefaction representation coefficient
Fidelity, distance minimization of the present invention by same people in different cameras, make different people between 2 video cameras away from
From maximization.Identify reconstructed error item to define:
Wherein S indicates that same class, D indicate different classes, and θ is a balance factor.
Further, in step (5), study blocks the aggregate projection matrix W of video and does not block the set of video respectively
Projection matrix V；Due to the blocking of the posture and viewpoint of same people, the difference of posture and viewpoint in same video, deposited between sample
In certain difference.These differences can make the characteristic dispersion of same people, keep matching more difficult.Therefore, it is based on and Fisher
The similar some standards of discrimination standard, the present invention consider study one sub-spaces projection, make sample have divergence in class small, but class
Between divergence it is big.
The mapping matrix can construct the relationship blocked between video features and ordinary video feature, to a certain extent
Compensate for the information loss of the space-time characteristic caused by blocking.Wherein divergence can indicate in the class of feature set in video camera A are as follows:
The class scatter of feature set can indicate in video camera A are as follows:
Wherein μiIt is A in camera AiAverage vector, μ be other samples (not including i-th of people) in camera A it is average to
Amount,
That is:NallRefer to the number of all samples in same video camera.Therefore it is taking the photograph
As the subspace projection in head A can indicate are as follows:
Wherein W is the subspace projection matrix of feature set in video camera A, can make the feature of same people to a certain extent
Collection compresses while reducing variation in class.The projection of B is similar with A.Therefore the subspace projection in camera B can indicate are as follows:
Wherein V is the subspace projection matrix of feature set in video camera B.By learning projection matrix W, it is possible to reduce video
Internal variation, A and B can be written as in the final projection of embedded space:
Further, in step (6), catalogue scalar functions are obtained.In view of identifying reconstructed error, the half of rarefaction representation coefficient
Projection insertion subspace in coupled maps and video.It is an object of the present invention to minimize objective function:
Wherein γ, α, β, η, λ are regularization parameter balance factors,
It is regularization term, prevented from intending
It closes.Formula (8) is converted to dictionary to study and ridge regression problem, SCD2The objective function of L can indicate are as follows:
WTW=I, VTV=I (9)
SCD of the invention2Half coupling projection matrix of L combination learning and dictionary are to DOAnd DN.The projection matrix acquired can weigh
Structure blocks the relationship between video and ordinary video, can make to block losing for caused space-time characteristic information to a certain extent
Mistake is compensated.
Further, in step (7), algorithm optimization is carried out to objective function.All variables in formula (9) are not convex
, the present invention solves known variables using alternative optimization strategy.In other words, when the present invention updates a variable every time, other
Variable is fixed.In order to minimize formula (9), 4 subproblems can be classified as, i.e. the sparse coding of training sample
It updates, the update of dictionary pair, the update of subspace projection matrix, the update of rarefaction representation coefficient mapping function in video.Firstly,
Mapping matrix P is initialized as unit matrix by the present invention.DOAnd DNThe method of initialization has very much, such as random matrix and PCA base.
The present invention with Frobenius norm to each column vector by dictionary to being initialized as random matrix, and pass through solution formula (10)
(11) rarefaction representation coefficient of X and Y is initialized:
(1) fixed other variable update W and V.Formula (9) can be rewritten are as follows:
By the way that the derivative of W and V is respectively set, formula (12) and (13) can be by solving as follows:
(2) fixed other variable update X and Y.X is updated first, and formula (9) can be rewritten are as follows:
WhereinExpression and XiSubset in the camera B of relevant correct matching or erroneous matching.By by XiLead
Number is set as 0 to solve, and can be obtained by following formula:
Y is updated, similar to formula (16), formula (9) can be rewritten are as follows:
WhereinExpression and YiSubset in the camera A of relevant correct matching or erroneous matching.By by Yi's
Derivative is set as 0 to solve, and can be obtained by following formula:
(3) fixed other variable update DOAnd DN.Formula (9) can be rewritten are as follows:
The present invention can use ADMM algorithm and obtain the solution of formula (20) and (21).
(4) fixed other variable update projection matrix P.Formula (9) can be rewritten are as follows:
It is solved by setting 0 for the derivative of P, the present invention can obtain:
P=(XXT+((λ/γ)I)-1(YXT) (23)
Further, in step (8), the pedestrian blocked under video identifies again.By study dictionary to DOAnd DN, mapping square
Subspace projection W and V in battle array P, video, can respectively obtain the robust rarefaction representation and effective rarefaction representation of test video.
Since feature of the video as the feature and ordinary video feature of probe collection F as picture library collection G will be blocked, matched process is executed
It is as follows:
(1) the expression coefficient f for blocking probe collection on dictionary is carried out by solution formula (10) using acquired P, W, V
Coding,
(2) it is encoded by expression coefficient g of the solution formula (11) to common dictionary on picture library collection,
(3) picture library concentrate identify with probe concentrate people image.By obtained rarefaction representation coefficient, originally
Invention can calculate the distance between picture library collection and the feature of probe collection, then adjust the distance and be ranked up, apart from the smallest figure
The image of library collection is exactly and probe collection matches correct image, i.e., the present invention has been matched on correct probe collection on picture library collection
The image of people.
Another object of the present invention is to provide the leaves for identifying dictionary learning based on half coupling described in a kind of implementation to block row
The leaf for identifying dictionary learning based on half coupling of people's recognition methods again blocks pedestrian's weight identifying system.
Another object of the present invention is to provide the leaves for identifying dictionary learning based on half coupling described in a kind of implementation to block row
The traffic route pedestrian image of people's recognition methods again knows equipment again.
In conclusion advantages of the present invention and good effect are as follows:
Whether there is a good superiority to verify method of the invention, the present invention is by the SCD of proposition2L method and it is several most
Recognition methods compares advanced pedestrian again, the recognition methods again of the pedestrian including 2 feature learnings: STFV3D, RFA-
Net；1 dictionary learning method: the pedestrian of PHDL and distance metric recognition methods again: RDC, TDL, KISSME, SI2DL.Respectively
It is tested on LO-PRID 2011 and LO-iLIDS-VID two datasets.
Experimental result:
Table 1 has counted the Top R matching rate on LO-PRID 2011 and LO-iLIDS-VID two datasets.
Fig. 2 has counted the Top R matching rate on LO-PRID 2011 and LO-iLIDS-VID two datasets.
Table 1
By table 1 and Fig. 2, the present invention can be seen that one kind more proposed by the invention compared with comparing algorithm and be based on half
The leaf that coupling identifies dictionary learning blocks pedestrian's weight recognizer and gets well than the performance of other control methods.Table 1 is shown in detail
Top-R matching rate, more precisely, compared with the optimum match method on Lo-iLIDS-VID data set, Rank-1
5% (27.3%-22.3%) is improved with rate.As can be seen that these are right for identifying problem again based on the pedestrian for blocking video
The performance of ratio method is poor, reason may be that many people in original iLIDS-vid data set exist in some picture frames
Certain blocks, for example, certain people may be blocked by certain objects or other pedestrians, this will have more to a certain extent is chosen
War property.It is projected using identifying in the video that thought learns, the feature set of each user can be made more compact, the spy of different people
Collection is separated.Therefore, method of the invention can be more preferable than the performance of other control methods.
Experimental result: Fig. 3 (a): counted that whether there is or not aggregate projection matrixes in video in LO-PRID 2011 and LO-iLIDS-
1 matching rate of Top in VID two datasets.Fig. 3 (b): counted that whether there is or not half coupling projection matrixes in 2011 He of LO-PRID
1 matching rate of Top in LO-iLIDS-VID two datasets.Fig. 4 (a): different dictionary size has been counted in LO-PRID
1 matching rate of Top on 2011 data sets.Fig. 4 (b): the convergence curve figure on 2011 data set of LO-PRID has been counted.
In order to assess the effect of aggregate projection item W and V in video, and in LO-PRID 2011 and LO-iLIDS-VID two
It carries out Top 1 on data set to match, the purpose is to separate the video set of same people and different people.In order to assess the effect of W and V
Fruit, present invention employs the methods of removal W and V, are referred to as SCD2L-W, SCD2L-V and SCD2L-WV.The experiment knot of Fig. 3 (a)
Fruit shows that matching rate can reduce when deleting each, and especially not there are two item, performance is substantially reduced.Cause
This, projection item plays an important role in identifying again based on feature set in video.
In order to assess the effect of half coupling projection matrix, and in LO-PRID 2011 and LO-iLIDS-VID two datasets
Mapping matrix is set as a unit matrix, referred to as SCD by upper progress Top 1 matching, the present invention2L-P.The experimental result of Fig. 3 (b)
Show that half coupled maps item is able to reflect the relationship between the shaded coefficient of feature and unobstructed sparse coefficient, can compensate for
The difference with feature under common camera is blocked, identifying again for people is conducive to.
Formula (9) is a combined optimization problem, and the present invention uses alternating iteration optimization algorithm, Fig. 4 (b) to this process
Convergence curve of the inventive algorithm on 2011 data set of LO-PRID is illustrated, the present invention can see curve and decline rapidly
It tends towards stability after 19 iteration.The algorithm of the invention iteration on 2011 data set of LO-PRID can be restrained less than 16 times, in LO-
Iteration can be restrained less than 19 times on iLIDS-VID data set.
Detailed description of the invention
Fig. 1 is that the leaf provided in an embodiment of the present invention for identifying dictionary learning based on half coupling blocks pedestrian's recognition methods stream again
Cheng Tu.
Fig. 2 is statistics provided in an embodiment of the present invention on LO-PRID 2011 and LO-iLIDS-VID two datasets
Top R matching rate figure.
Fig. 3 is 1 matching rate figure of Top provided in an embodiment of the present invention.
In figure: (a): having counted that whether there is or not aggregate projection matrixes in video in LO-PRID 2011 and LO-iLIDS-VID two
1 matching rate of Top on data set.(b): having counted that whether there is or not half coupling projection matrixes in LO-PRID 2011 and LO-iLIDS-
1 matching rate of Top in VID two datasets.
Fig. 4 is convergence curve figure provided in an embodiment of the present invention.
In figure: (a): having counted the 1 matching rate curve of Top of different dictionary size on 2011 data set of LO-PRID
Figure；(b): having counted the convergence curve figure on 2011 data set of LO-PRID.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
In the prior art, to including visual space-time information in video, identification is not can be carried out.It not can solve what branches and leaves blocked
Pedestrian identify again in problem.
To solve the above problems, below with reference to concrete scheme, the present invention is described in detail.
The leaf provided in an embodiment of the present invention for identifying dictionary learning based on half coupling blocks pedestrian's recognition methods again, is first
Data set is collected, and the pedestrian public from two identify that true branches and leaves are added on data set to be blocked again, include block video and
Ordinary video；Then feature extraction is carried out to blocking video and ordinary video respectively；It is followed by special to the sample extracted
Sign is handled, here for projecting in the feature set study collection class of each user in two cameras；Present invention introduces
The dictionary pair of video and ordinary video is blocked in dictionary learning method, study；The present invention is from blocking in video and ordinary video again
Learn projection matrix；Finally present invention introduces identifying thought to learn dictionary pair, word is identified based on half coupling to realize one kind
The leaf of allusion quotation study blocks pedestrian's weight recognizer.The advantage of the invention is that solve the pedestrian that branches and leaves block identify again in ask
It inscribes and has been put forward for the first time the leaf that dictionary learning is identified based on half coupling and block pedestrian and identify (SCD again2L) technology.The technology
It practises half coupled maps matrix and identifies dictionary pair, half coupled maps matrix can compensate the difference blocked between video and ordinary video
It is different, identify dictionary to the same person can be made more compact in different cameral, different people separates in different cameras.2
The results show on a public data collection, the method proposed have better recognition performance
As shown in Figure 1, in embodiments of the present invention, the leaf for identifying dictionary learning based on half coupling blocks the pedestrian side of identification again
Method specifically includes the following steps:
Step (1): data are collected and establish database, are formed by blocking video and ordinary video.
Step (2): feature is extracted to blocking video and ordinary video respectively.
Step (3): the dictionary pair with half coupled maps is designed.
Step (4): learn dictionary pair using thought is identified.
Step (5): study blocks the aggregate projection matrix W of video and does not block the aggregate projection matrix V of video respectively.
Step (6): catalogue scalar functions are obtained.
Step (7): algorithm optimization is carried out to objective function.
Step (8): the pedestrian blocked under video identifies again.
As the preferred embodiment of the present invention, in step (1), the specific method that the collection data establish database is:
What the technology solved is to identify problem again in the pedestrian blocked in scene and normal scene, due in the case where blocking scene
There is no the data set that pedestrian identifies again, therefore the present invention has remake 2 common data sets, it is (true with template is really blocked
Real branches and leaves) it is simulated on PRID 2011 and iLIDS-VID, form 2 new data set LO-PRID 2011 and LO-
iLIDS-VID.749 people of 385 people of A video camera and B video camera is shared in 2011 data set of PRID, has 200 in 2 visual angles
Individual occurs, and each video is made of 5 to 675 picture frames.In order to guarantee the effective length of each walking period, present invention choosing
178 people of 20 frames or more are taken, therefrom 89 pairs is randomly choosed and is trained, and using remaining to testing.The present invention uses
Template is blocked to block the addition of PRID 2011 to generate 2011 data set of Lo-PRID.Likewise, 2 of iLIDS-VID image
300 couples of pedestrians are shared in head, picture frame is from 23 to 192, and average frame number is 73 frames, and the present invention covers iLIDS- with template is blocked
VID data set, Lai Shengcheng LO-iLIDS-VID data set.Simulation process and the process phase for generating 2011 data set of Lo-PRID
Seemingly.The data set that the band ultimately produced blocks can be from https: //sites.google.com/site/LODB is obtained.
As the preferred embodiment of the present invention, in step (2), feature is extracted to blocking video and ordinary video respectively, it is right
In blocking video, the present invention passes through FEP (Wang et al.2014；Liu et al.2015) walking cycle feature is extracted, this
Invention has empirically selected 20 frames as a walking week for blocking video according to the average time of walking cycle in normal video
Phase.For ordinary video, the present invention is directly that each user extracts STFV3D feature.
As the preferred embodiment of the present invention, in step (3), the dictionary pair with half coupled maps is designed.For identical
People, there are different postures and viewpoints to change for different video cameras, but blocks and normal video is special in visual appearance and space-time
There are similitudes in sign.Relax equal strong in the two spaces it is assumed that the present invention considers one and half coupled maps matrixes, with anti-
Reflect the relationship between character pair.For this purpose, the present invention devises the dictionary pair with half coupled maps, it is following to indicate:
s.t.Ai=DOXi；Bi=DNYi
Wherein Erepresent() indicates that sub- dictionary indicates fidelity term, and X and Y are A and B respectively in DOAnd DNOn expression system
Number, Emapping() is mapping fidelity, and the purpose is to find the relationship between A and the code coefficient of B, Φ () is to block
With the coefficient mapping function of common dictionary pair.A=[A1, A2...AN] it is the space-time characteristic for blocking training video,It is the video features subset of i-th of people, niIt is the walking cycle number of first man, N is the number of people of being blocked
Amount, B=[B1, B2...BN] it is the space-time characteristic collection for not blocking training video, whereinIt is and njA step
The space-time characteristic collection of row period corresponding j-th of people, N are the numbers in ordinary video.
According to claim 1, a kind of leaf identifying dictionary learning based on half coupling blocks pedestrian's recognition methods again,
It is characterized in that, in step (4), learns dictionary pair using thought is identified；In order to improve the identification fidelity of rarefaction representation coefficient,
Distance minimization of the present invention by same people in different cameras keeps distance of the different people between 2 video cameras maximum
Change.Identify reconstructed error item to define:
Wherein S indicates that same class, D indicate different classes, and θ is a balance factor.
According to claim 1, a kind of leaf identifying dictionary learning based on half coupling blocks pedestrian's recognition methods again,
It is characterized in that, in step (5), study blocks the aggregate projection matrix W of video and do not block the aggregate projection square of video respectively
Battle array V；Due to the blocking of the posture and viewpoint of same people, the difference of posture and viewpoint in same video, there is one between sample
Fixed difference.These differences can make the characteristic dispersion of same people, keep matching more difficult.Therefore, it is based on differentiating with Fisher and mark
Quasi- similar some standards, the present invention consider study one sub-spaces projection, make sample have divergence in class small, but class scatter
Greatly.
The mapping matrix can construct the relationship blocked between video features and ordinary video feature, to a certain extent
Compensate for the information loss of the space-time characteristic caused by blocking.Wherein divergence can indicate in the class of feature set in video camera A are as follows:
The class scatter of feature set can indicate in video camera A are as follows:
Wherein μiIt is A in camera AiAverage vector, μ be other samples (not including i-th of people) in camera A it is average to
Amount,
That is:NallRefer to the number of all samples in same video camera.Therefore it is taking the photograph
As the subspace projection in head A can indicate are as follows:
Wherein W is the subspace projection matrix of feature set in video camera A, can make the feature of same people to a certain extent
Collection compresses while reducing variation in class.The projection of B is similar with A.Therefore the subspace projection in camera B can indicate are as follows:
Wherein V is the subspace projection matrix of feature set in video camera B.By learning projection matrix W, it is possible to reduce video
Internal variation, A and B can be written as in the final projection of embedded space:
It is a kind of according to claim 1 that the leaf for identifying dictionary learning is coupled based on half as the preferred embodiment of the present invention
Pedestrian's recognition methods again is blocked, in step (6), obtains catalogue scalar functions.In view of identifying reconstructed error, rarefaction representation coefficient
Half coupled maps and video in projection insertion subspace.It is an object of the present invention to minimize objective function:
Wherein γ, α, β, η, λ are regularization parameter balance factors,
It is regularization term, prevented from intending
It closes.Formula (8) is converted to dictionary to study and ridge regression problem, SCD2The objective function of L can indicate are as follows:
WTW=I, VTV=I (9).
SCD of the invention2Half coupling projection matrix of L combination learning and dictionary are to DOAnd DN.The projection matrix acquired can weigh
Structure blocks the relationship between video and ordinary video, can make to block losing for caused space-time characteristic information to a certain extent
Mistake is compensated.
As the preferred embodiment of the present invention, in step (7), algorithm optimization is carried out to objective function.Institute in formula (9)
Variable be not it is convex, the present invention using alternative optimization strategy solve known variables.In other words, the present invention updates one every time
When a variable, dependent variable is fixed.In order to minimize formula (9), 4 subproblems can be classified as, that is, train sample
The update of this sparse coding, the update of dictionary pair, the update of subspace projection matrix, rarefaction representation coefficient map letter in video
Several updates.Firstly, mapping matrix P is initialized as unit matrix by the present invention.DOAnd DNThe method of initialization has very much, such as with
Machine matrix and PCA base.The present invention with Frobenius norm to each column vector by dictionary to being initialized as random matrix, and lead to
Solution formula (10) and (11) are crossed to initialize the rarefaction representation coefficient of X and Y:
Fixed other variable update W and V.Formula (9) can be rewritten are as follows:
By the way that the derivative of W and V is respectively set, formula (12) and (13) can be by solving as follows:
Fixed other variable update X and Y.X is updated first, and formula (9) can be rewritten are as follows:
WhereinExpression and XiSubset in the camera B of relevant correct matching or erroneous matching.By by XiLead
Number is set as 0 to solve, and can be obtained by following formula:
Y is updated, similar to formula (16), formula (9) can be rewritten are as follows:
WhereinExpression and YiSubset in the camera A of relevant correct matching or erroneous matching.By by Yi's
Derivative is set as 0 to solve, and can be obtained by following formula:
Fixed other variable update DOAnd DN.Formula (9) can be rewritten are as follows:
The present invention can use ADMM algorithm and obtain the solution of formula (20) and (21).
Fixed other variable update projection matrix P.Formula (9) can be rewritten are as follows:
It is solved by setting 0 for the derivative of P, the present invention can obtain:
P=(XXT+((λ/γ)I)-1(YXT) (23)。
It is a kind of according to claim 1 that the leaf for identifying dictionary learning is coupled based on half as the preferred embodiment of the present invention
Pedestrian's recognition methods again is blocked, in step (8), the pedestrian blocked under video identifies again.By study dictionary to DOAnd DN, reflect
It penetrates matrix P, subspace projection W and V in video, the robust rarefaction representation and effective sparse table of test video can be respectively obtained
Show.Since feature of the video as the feature and ordinary video feature of probe collection F as picture library collection G will be blocked, execute matched
Process is as follows:
(1) the expression coefficient f for blocking probe collection on dictionary is carried out by solution formula (10) using acquired P, W, V
Coding,
(2) it is encoded by expression coefficient g of the solution formula (11) to common dictionary on picture library collection,
(3) picture library concentrate identify with probe concentrate people image.By obtained rarefaction representation coefficient, originally
Invention can calculate the distance between picture library collection and the feature of probe collection, then adjust the distance and be ranked up, apart from the smallest figure
The image of library collection is exactly and probe collection matches correct image, i.e., the present invention has been matched on correct probe collection on picture library collection
The image of people.
In embodiments of the present invention, Fig. 2 is statistics provided in an embodiment of the present invention in LO-PRID 2011 and LO-
Top R matching rate figure in iLIDS-VID two datasets.
Fig. 3 is 1 matching rate figure of Top provided in an embodiment of the present invention.
In figure: (a): having counted that whether there is or not aggregate projection matrixes in video in LO-PRID 2011 and LO-iLIDS-VID two
1 matching rate of Top on data set.(b): having counted that whether there is or not half coupling projection matrixes in LO-PRID 2011 and LO-iLIDS-
1 matching rate of Top in VID two datasets.
Fig. 4 is convergence curve figure provided in an embodiment of the present invention.
In figure: (a): having counted the 1 matching rate curve of Top of different dictionary size on 2011 data set of LO-PRID
Figure；(b): having counted the convergence curve figure on 2011 data set of LO-PRID.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of leaf for identifying dictionary learning based on half coupling blocks pedestrian's recognition methods again, which is characterized in that coupled based on half
The leaf of identification dictionary learning blocks pedestrian, and recognition methods includes: again
Data set collect, identify that true branches and leaves are added on data set to be blocked again from public pedestrian, include block video and
Ordinary video；Feature extraction is carried out to blocking video and ordinary video respectively；
The sample characteristics of extraction are handled, dictionary learning method, study are introduced for projection in feature set study collection class
The dictionary pair for blocking video and ordinary video learns projection matrix from blocking in video and ordinary video；
And introduce and identify thought study dictionary pair, the leaf for carrying out coupling identification dictionary learning based on half blocks pedestrian and identifies again.
2. the leaf for identifying dictionary learning based on half coupling as described in claim 1 blocks pedestrian's recognition methods again, feature exists
In blocking pedestrian based on the leaf that half coupling identifies dictionary learning, recognition methods specifically includes again:
Step (1): data are collected and establish database, are formed by blocking video and ordinary video；
Step (2): feature is extracted to blocking video and ordinary video respectively；
Step (3): the dictionary pair with half coupled maps is designed；
Step (4): learn dictionary pair using thought is identified；
Step (5): study blocks the aggregate projection matrix W of video and does not block the aggregate projection matrix V of video respectively；
Step (6): catalogue scalar functions are obtained；
Step (7): algorithm optimization is carried out to objective function；
Step (8): the pedestrian blocked under video identifies again.
3. the leaf for identifying dictionary learning based on half coupling as claimed in claim 2 blocks pedestrian's recognition methods again, feature exists
In the specific method that collection data described in step (1) establish database includes:
Two common data sets are remake, are simulated with really template is blocked on PRID 2011 and iLIDS-VID,
Form 2 new data set LO-PRID 2011 and LO-iLIDS-VID；178 people for choosing 20 frames or more, therefrom randomly choose
89 pairs are trained, and using remaining to testing；Generation Lo- is blocked to the addition of PRID 2011 using template is blocked
2011 data set of PRID；Using template covering iLIDS-VID data set is blocked, LO-iLIDS-VID data set is generated；Finally obtain
The data set that the band that must be generated blocks.
4. the leaf for identifying dictionary learning based on half coupling as claimed in claim 2 blocks pedestrian's recognition methods again, feature exists
In extracting feature to blocking video and ordinary video respectively, for blocking video, pass through FEP and extract trip in step (2)
Periodic characteristic is walked, according to the average time of walking cycle in normal video, selects 20 frames as a walking week for blocking video
Phase；For ordinary video, STFV3D feature directly is extracted for each user；
In step (3), the dictionary pair with half coupled maps is designed, following to indicate:
s.t.Ai=DOXi；Bi=DNYi
Wherein Erepresent() indicates that sub- dictionary indicates fidelity term, and X and Y are A and B respectively in DOAnd DNOn expression coefficient,
Emapping() is mapping fidelity, it is therefore an objective to find the relationship between A and the code coefficient of B, Φ () be block with it is general
The coefficient mapping function of logical dictionary pair；A=[A1,A2...AN] it is the space-time characteristic for blocking training video,
It is the video features subset of i-th of people, ni is the walking cycle number of first man, and N is the quantity of people of being blocked, B=[B1,
B2...BN] it is the space-time characteristic collection for not blocking training video,It is and njA walking period is corresponding
The space-time characteristic collection of j-th of people, N are the numbers in ordinary video.
5. the leaf for identifying dictionary learning based on half coupling as claimed in claim 2 blocks pedestrian's recognition methods again, feature exists
In by distance minimization of the same people in different cameras, making different people between 2 video cameras in step (4)
Distance maximizes；Definition identifies reconstructed error item:
Wherein S indicates that same class, D indicate different classes, and θ is a balance factor；
In step (5), study blocks the aggregate projection matrix W of video and does not block the aggregate projection matrix V of video respectively；It learns
Sub-spaces projection is practised, makes sample have divergence in class small, but class scatter is big；The mapping matrix, which constructs, blocks video spy
The relationship between ordinary video feature is levied, divergence indicates in the class of feature set in video camera A are as follows:
The class scatter of feature set indicates in video camera A are as follows:
Wherein μiIt is A in camera AiAverage vector, μ is the average vector of other samples in camera A,NallRefer to the number of all samples in same video camera；Subspace in camera A
Projective representation are as follows:
Wherein W is the subspace projection matrix of feature set in video camera A；Subspace projection in camera B indicates are as follows:
V is the subspace projection matrix of feature set in video camera B；A and B is written as in the final projection of embedded space:
6. the leaf for identifying dictionary learning based on half coupling as claimed in claim 2 blocks pedestrian's recognition methods again, feature exists
In in step (6), target is to minimize objective function:
γ, α, β, η, λ are regularization parameter balance factors,
It is regularization term, prevents over-fitting；
SCD2The objective function of L indicates are as follows:
WTW=I, VTV=I.
7. the leaf for identifying dictionary learning based on half coupling as claimed in claim 2 blocks pedestrian's recognition methods again, feature exists
In using alternative optimization strategy solution known variables in step (7)；When updating a variable every time, dependent variable is fixed；
In order to minimize SCD2The objective function of L is divided into the update of the sparse coding of training sample, the update of dictionary pair, son in video
The update of space projection matrix, rarefaction representation coefficient mapping function 4 subproblems of update；Mapping matrix P is initialized as list
Bit matrix；With Frobenius norm to each column vector by dictionary to being initialized as random matrix, and by solving following formula
Initialize the rarefaction representation coefficient of X and Y:
Fixed other variable update W and V；SCD2The objective function of L is rewritten are as follows:
By the way that the derivative of W and V is respectively set, by solving as follows:
Fixed other variable update X and Y；X, SCD are updated first2The objective function of L is rewritten are as follows:
WhereinExpression and XiSubset in the camera B of relevant correct matching or erroneous matching；By by XiDerivative
0 is set as to solve, is obtained by following formula:
Update Y, SCD2The objective function of L is rewritten are as follows:
WhereinExpression and YiSubset in the camera A of relevant correct matching or erroneous matching；By by YiDerivative
It is set as 0 solution, is obtained by following formula:
Fixed other variable update DOAnd DN；SCD2The objective function of L is rewritten are as follows:
Fixed other variable update projection matrix P；SCD2The objective function Equation of L is rewritten are as follows:
It is solved, is obtained by setting 0 for the derivative of P:
P=(XXT+((λ/γ)I)-1(YXT)。
8. the leaf for identifying dictionary learning based on half coupling as claimed in claim 2 blocks pedestrian's recognition methods again, feature exists
In in step (8), by study dictionary to DOAnd DN, mapping matrix P, subspace projection W and V in video, respectively obtain survey
Try the robust rarefaction representation and effective rarefaction representation of video；It is special as the feature of probe collection F and ordinary video that video will be blocked
The feature as picture library collection G is levied, it is as follows to execute matched process:
1) using acquired P, W, V, pass through solution formulaTo blocking
The expression coefficient f of probe collection is encoded on dictionary,
2) pass through solution formulaTo the expression system of common dictionary on picture library collection
Number g is encoded,
3) picture library concentrate identify with probe concentrate people image；By obtained rarefaction representation coefficient, calculate
The distance between picture library collection and the feature of probe collection, then adjust the distance and are ranked up, and the image apart from the smallest picture library collection is exactly
Correct image is matched with probe collection.
9. a kind of leaf for implementing to identify dictionary learning based on half coupling described in claim 1 block pedestrian again recognition methods based on
The leaf that half coupling identifies dictionary learning blocks pedestrian's weight identifying system.
10. the friendship that a kind of leaf for implementing to identify dictionary learning based on half coupling described in claim 1 blocks pedestrian's recognition methods again
Passway pedestrian image knows equipment again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910344098.5A CN110046608B (en) | 2019-04-26 | 2019-04-26 | Leaf-shielded pedestrian re-recognition method and system based on semi-coupling identification dictionary learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910344098.5A CN110046608B (en) | 2019-04-26 | 2019-04-26 | Leaf-shielded pedestrian re-recognition method and system based on semi-coupling identification dictionary learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110046608A true CN110046608A (en) | 2019-07-23 |
| CN110046608B CN110046608B (en) | 2020-01-07 |
Family
ID=67279623
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910344098.5A Active CN110046608B (en) | 2019-04-26 | 2019-04-26 | Leaf-shielded pedestrian re-recognition method and system based on semi-coupling identification dictionary learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110046608B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113313074A (en) * | 2021-06-28 | 2021-08-27 | 严大莲 | Feature sparse representation multi-dictionary pair learning pedestrian re-identification method |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0736989A (en) * | 1993-07-20 | 1995-02-07 | Brother Ind Ltd | Automatic certificate issue device |
| CN104317838A (en) * | 2014-10-10 | 2015-01-28 | 浙江大学 | Cross-media Hash index method based on coupling differential dictionary |
| CN105550649A (en) * | 2015-12-09 | 2016-05-04 | 武汉工程大学 | Extremely low resolution human face recognition method and system based on unity coupling local constraint expression |
| CN106066992A (en) * | 2016-05-13 | 2016-11-02 | 哈尔滨工业大学深圳研究生院 | Differentiation dictionary learning algorithm based on adaptive local constraint and face identification system |
| CN106980848A (en) * | 2017-05-11 | 2017-07-25 | 杭州电子科技大学 | Facial expression recognizing method based on warp wavelet and sparse study |
| WO2018081135A1 (en) * | 2016-10-25 | 2018-05-03 | Vmaxx Inc. | Point to set similarity comparison and deep feature learning for visual recognition |
| CN108664951A (en) * | 2018-05-22 | 2018-10-16 | 南京邮电大学 | Pedestrian's recognition methods again based on color name feature |
| CN108985177A (en) * | 2018-06-21 | 2018-12-11 | 南京师范大学 | A kind of facial image classification method of the quick low-rank dictionary learning of combination sparse constraint |
| US20180373962A1 (en) * | 2017-06-27 | 2018-12-27 | Canon Kabushiki Kaisha | method and apparatus for determining similarity of objects in images |
| US20180374233A1 (en) * | 2017-06-27 | 2018-12-27 | Qualcomm Incorporated | Using object re-identification in video surveillance |
-
2019
- 2019-04-26 CN CN201910344098.5A patent/CN110046608B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0736989A (en) * | 1993-07-20 | 1995-02-07 | Brother Ind Ltd | Automatic certificate issue device |
| CN104317838A (en) * | 2014-10-10 | 2015-01-28 | 浙江大学 | Cross-media Hash index method based on coupling differential dictionary |
| CN105550649A (en) * | 2015-12-09 | 2016-05-04 | 武汉工程大学 | Extremely low resolution human face recognition method and system based on unity coupling local constraint expression |
| CN106066992A (en) * | 2016-05-13 | 2016-11-02 | 哈尔滨工业大学深圳研究生院 | Differentiation dictionary learning algorithm based on adaptive local constraint and face identification system |
| WO2018081135A1 (en) * | 2016-10-25 | 2018-05-03 | Vmaxx Inc. | Point to set similarity comparison and deep feature learning for visual recognition |
| CN106980848A (en) * | 2017-05-11 | 2017-07-25 | 杭州电子科技大学 | Facial expression recognizing method based on warp wavelet and sparse study |
| US20180373962A1 (en) * | 2017-06-27 | 2018-12-27 | Canon Kabushiki Kaisha | method and apparatus for determining similarity of objects in images |
| US20180374233A1 (en) * | 2017-06-27 | 2018-12-27 | Qualcomm Incorporated | Using object re-identification in video surveillance |
| CN108664951A (en) * | 2018-05-22 | 2018-10-16 | 南京邮电大学 | Pedestrian's recognition methods again based on color name feature |
| CN108985177A (en) * | 2018-06-21 | 2018-12-11 | 南京师范大学 | A kind of facial image classification method of the quick low-rank dictionary learning of combination sparse constraint |
Non-Patent Citations (2)
| Title |
|---|
| WANG CAILING: "《Person re-identification based on saliency》", 《PROCEEDINGS OF THE 35TH CHINESE CONTROL CONFERENCE 》 * |
| 朱小柯: "《行人重识别关键匹配技术研究》", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113313074A (en) * | 2021-06-28 | 2021-08-27 | 严大莲 | Feature sparse representation multi-dictionary pair learning pedestrian re-identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110046608B (en) | 2020-01-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sarfraz et al. | A pose-sensitive embedding for person re-identification with expanded cross neighborhood re-ranking | |
| McLaughlin et al. | Data-augmentation for reducing dataset bias in person re-identification | |
| CN105574505B (en) | The method and system that human body target identifies again between a kind of multiple-camera | |
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| CN109961051A (en) | A kind of pedestrian's recognition methods again extracted based on cluster and blocking characteristic | |
| CN107463920A (en) | A kind of face identification method for eliminating partial occlusion thing and influenceing | |
| CN108509854B (en) | Pedestrian re-identification method based on projection matrix constraint and discriminative dictionary learning | |
| CN103679142B (en) | A kind of recognition method for target human body based on space constraint | |
| CN104598883A (en) | Method for re-recognizing target in multi-camera monitoring network | |
| CN105574475B (en) | A kind of rarefaction representation classification method based on common vector dictionary | |
| CN109711366A (en) | A kind of recognition methods again of the pedestrian based on group information loss function | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN107463954B (en) | A kind of template matching recognition methods obscuring different spectrogram picture | |
| CN109902662A (en) | A kind of pedestrian recognition methods, system, device and storage medium again | |
| CN109447123B (en) | Pedestrian re-identification method based on label consistency constraint and stretching regularization dictionary learning | |
| CN109409201A (en) | A kind of pedestrian's recognition methods again based on shared and peculiar dictionary to combination learning | |
| CN111563404B (en) | Global local time representation method for video-based person re-identification | |
| CN109635634A (en) | A kind of pedestrian based on stochastic linear interpolation identifies data enhancement methods again | |
| Reddy et al. | Ocularnet: deep patch-based ocular biometric recognition | |
| CN114299542A (en) | Video pedestrian re-identification method based on multi-scale feature fusion | |
| CN109165698A (en) | A kind of image classification recognition methods and its storage medium towards wisdom traffic | |
| CN113947814A (en) | Cross-visual angle gait recognition method based on space-time information enhancement and multi-scale saliency feature extraction | |
| Choi et al. | Data insufficiency in sketch versus photo face recognition | |
| CN114882537A (en) | Finger new visual angle image generation method based on nerve radiation field | |
| CN109886251A (en) | A kind of recognition methods again of pedestrian end to end guiding confrontation study based on posture |
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 |