CN107145827A - Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric - Google Patents
Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric Download PDFInfo
- Publication number
- CN107145827A CN107145827A CN201710213901.2A CN201710213901A CN107145827A CN 107145827 A CN107145827 A CN 107145827A CN 201710213901 A CN201710213901 A CN 201710213901A CN 107145827 A CN107145827 A CN 107145827A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- picture
- training
- metric
- 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
- 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
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (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)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of across video camera pedestrian recognition methods again learnt based on adaptive distance metric, sample pair is constituted first with pedestrian's picture from different cameras in training set, constraint is provided for learning distance metric；Then the ga s safety degree according to training sample in original feature space is different samples to adaptively distributing training weight；Then use the object function for accelerating near-end gradient algorithm to adjust the distance metric learning to be solved, obtain mahalanobis distance metric matrix；The distance matrix metric for learning to obtain finally is substituted into mahalanobis distance metric function, and calculates the mahalanobis distance between test phase pedestrian picture feature vector, similitude ranking results are obtained.The present invention has taken into full account difference of the different training samples during learning distance metric, the distance metric function that study is obtained is had stronger identification, so as to reach higher across video camera pedestrian recognition accuracy again.
Description
Technical field
It is specially that one kind is based on adaptive distance metric the present invention relates to a kind of method of technical field of video image processing
Across the video camera pedestrian recognition methods again of study.
Background technology
Have widely to specifying target pedestrian to carry out match cognization in the camera supervised network of camera lens non-overlapping copies
Application prospect, this technology is referred to as across video camera pedestrian and recognized again, and it is across video camera target following in intelligent monitor system
And basis and the precondition of behavioural analysis.Due to huge commercial value, therefore across video camera pedestrian recognizes near again
Of great interest and research over year.Emphasis and difficult point of across the video camera pedestrian again in Study of recognition are between video camera
Illumination, visual angle difference and pedestrian attitudes vibration in itself and circumstance of occlusion change.In addition, low resolution monitor video
So that the information such as face is in most cases no longer applicable.
In order to overcome above mentioned problem, Many researchers wish to design the spy high to across camera vision change robustness
Levy.However, because the difference between video camera in reality scene has very strong uncertainty, there is also all changeable for the posture of pedestrian
Number, finds out one and all these features for changing strong robustnesses but effectively differentiating different pedestrians is difficult to implement.Therefore, away from
Across video camera pedestrian identification problem again is introduced in from metric learning.Specifically, learning distance metric will have been marked
Pedestrian sample to (positive sample to representing that two pictures belong to same a group traveling together, to representing that two pictures belong to does not go together by negative sample
People) as training set, by training set is closed sample to the distance between optimize, study obtain a distance metric
Matrix, can project to all samples one new feature space, mutual between positive sample pair in this feature space
It is in small distance, and the mutual distance between negative sample pair is larger.Such as Zheng Wei poems et al. in 2012《IEEE
Transactions on Pattern Analysis and Machine Intelligence》(International Electrical and electronic engineering
Shi Xiehui pattern analyses and machine intelligence journal) paper " the Reidentification by relative distance that deliver
Comparison " (across the video camera pedestrian matching compared based on relative distance) obtains optimal probability using training sample study
Relative distance module, and distance metric is carried out to other pictures in database with this standard.By Feature Mapping
To public space, learning distance metric can solve the variability issues between different cameras to a certain extent.However, existing
There is learning distance metric algorithm generally coequally to treat all samples in the training process, do not account for the difference between different samples
The opposite sex.Because different samples has different ga s safety degrees on original feature space, therefore for learning distance metric
Significance level is substantially different.
The content of the invention
For above-mentioned the deficiencies in the prior art, the present invention provide it is a kind of based on adaptive distance metric learn across video camera
Pedestrian's recognition methods again, can according to the distribution situation on training sample original feature space to sample carry out adaptive classification and
Weighting so that different samples can play different effects during learning distance metric, and identification is obtained so as to train
Stronger distance metric function.
To achieve the above object, the present invention constitutes sample first with pedestrian's picture from different cameras in training set
It is right, provide constraint for learning distance metric；Then it is different according to ga s safety degree of the training sample in original feature space
Sample is to adaptively distributing training weight；Then the object function for accelerating near-end gradient algorithm to adjust the distance metric learning is used
Solved, obtain mahalanobis distance metric matrix；The distance matrix metric for learning to obtain finally is substituted into mahalanobis distance measurement letter
Number, and the mahalanobis distance between test phase pedestrian picture feature vector is calculated, obtain similitude ranking results.
The inventive method is realized by step in detail below：
Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric, is comprised the following steps：
Step 1：The training data of transmission range metric learning, pedestrian's picture under different cameras is expressed as looking into
Ask collectionAnd Candidate SetWhereinIt is i-th query graph
The characteristic vector of piece and jth candidate's picture, andWithFor the identity label of correspondence pedestrian, n is training stage query set
Picture sum, m is total for the picture of training stage Candidate Set；
Step 2：Choose the inquiry picture x in query setiWith candidate's picture y in Candidate SetjSample is constituted to (xi,yj),
Be sample to distribution two tag along sort zij, wherein whenWhen zij=1, (xi,yj) it is referred to as positive sample pair, and work asWhen zij=-1, (xi,yj) it is referred to as negative sample pair；Arbitrary sample is defined to (xi,yj) between mahalanobis distance measurement
Function is：
Wherein M is the distance matrix metric in mahalanobis distance；
Step 3：Using Logistic loss functions be training set in each sample to (xi,yj) set up distance metric
The loss function of habit：
Wherein ξ be training set in all samples to the Euclidean distance average between characteristic vector, as (xi,yj) it is positive sample
Pair when, loss function constraint mahalanobis distance dM(xi,yj) be less thanAnd work as (xi,yj) be negative sample pair when, loss function constraint horse
Family name is apart from dM(xi,yj) be more than
Step 4：The loss function constraint between all samples pair in training set is considered simultaneously, defines adaptive distance metric
The overall goal function of study is：
Step 5：It is the loss function ψ (x of different samples pair according to the otherness of training samplei,yj) the different instruction of distribution
Practice weight wij, it is changed into the object function that adaptive distance metric learns：
Step 6：According to object function, adaptive distance metric study is defined as optimization problem：
Step 7：Using the optimization problem accelerated in near-end gradient algorithm solution procedure 6, corresponding distance metric square is obtained
Battle array M；
Step 8：In test phase, for the characteristic vector x of given query template picturepWith it is suspicious under other video cameras
The characteristic vector group for candidate's picture that pedestrian target is constitutedN is total for the picture of test phase Candidate Set, by distance
The mahalanobis distance metric function that metric matrix M is substituted into step 2, calculates x respectivelypWith the characteristic vector of every candidate's pictureIt
Between mahalanobis distanceAnd candidate's picture is ranked up according to mahalanobis distance size, make and xpMahalanobis distance compared with
Small candidate's picture comes the front end of queue.
Further：The otherness according to training sample described in step 5, is the loss function ψ of different samples pair
(xi,yj) the different training weight w of distributionij, the process of implementing is：
Step 5.1：For the characteristic vector x of the inquiry picture in training seti, calculate Candidate SetIn all candidate's pictures
Characteristic vector and xiIn the characteristic distance of theorem in Euclid space, then the order according to distance from small to large is rightIn picture arranged
Sequence；
Step 5.2：According to the ranking results in step 5.1, by Candidate SetIt is divided into for xiDifficult setMedium setAnd simple set
Difficulty setDefinition be：
In formula, ri(yj) represent candidate's picture yjPosition in the sequencing queue that step 5.1 is obtained, andThen represent
xiIn Candidate SetIn correct matching y+Position in sequencing queue；
Medium setDefinition be：
Wherein m is Candidate SetThe sum of middle pedestrian's picture number, namely in step 5.1 sequencing queue total length；
Simple setDefinition be：
Step 5.3：According to Candidate Set in step 5.2Division result, be sample to (xi,yj) loss function it is adaptive
Training weight w should be distributedij；As (xi,yj) be positive sample pair when, by ψ (xi,yj) training weight wijIt is set to 1/N+, wherein N+For
The sum of positive sample pair in training set；And work as (xi,yj) be negative sample pair when, its loss function ψ (x are defined by following formulai,yj)
Training weight wij：
Wherein, N-The negative sample that weight is not zero in expression training set is to sum, β1And β2For regulation training weight change model
The balance parameters enclosed.
Further：Use described in step 7 accelerates the optimization problem in near-end gradient algorithm solution procedure 6, obtains
Corresponding distance matrix metric M, it specifically includes following steps：
Step 7.1：Initialize M0And M-1For unit matrix, and renewal is iterated to it；
Step 7.2：Specifically, for the t times iteration, calculate first before polymerization to matrix St：
Wherein coefficient0 is set as in first time iteration；
Step 7.3：Define matrixWhereinIt is the total of learning distance metric
Body object function Ψ (M) is in StThe gradient at place, ηtFor iteration step length；
Step 7.4：To matrix PtEigenvalues Decomposition is carried out, is denoted asThen to the t times iteration
Distance matrix metric M be updated, obtain iteration result Mt：
WhereinFor ensuring distance matrix metric MtIt is positive semi-definite；
Step 7.5：Repeat step 7.2 to 7.4, until reaching the condition of convergence：
Step 7.6：Return to M during convergencetThe training result M learnt as adaptive distance metric.
Compared with prior art, beneficial effects of the present invention are as follows：
1) present invention according to ga s safety degree of the training sample on original feature space to sample to carrying out adaptivity
Classification and weighting, all training samples, this hair are coequally treated compared to existing learning distance metric technology in the training process
It is bright to better profit from the otherness between training sample；
2) present invention has cast out training concentrated part primitive character by the way that the training weight of part sample pair is set into zero
The negative sample easily distinguished in space, not only effectively alleviate pedestrian again in identification problem negative sample to it is excessive the problem of, also subtract
The amount of calculation in training process is lacked；
3) present invention can utilize the markup information of training sample, and study obtains specific mahalanobis distance metric function, made
Picture feature with a group traveling together is closer to the distance, without same pedestrian's picture characteristic distance farther out, effectively overcome different cameras
Between the influence that is recognized again to across video camera pedestrian of difference.
Brief description of the drawings
Across the video camera pedestrian recognition methods overall flow signal again that Fig. 1 is learnt for the present invention based on adaptive distance metric
Figure.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below in conjunction with the accompanying drawings and specific implementation
Example, is described in further detail to technical scheme.
Following examples are implemented lower premised on technical solution of the present invention, give detailed embodiment and tool
The operating process of body, but protection scope of the present invention is not limited to following embodiments.
Embodiment
The present embodiment is regarded by the pedestrian that will occur in video camera A as target, and with video camera A in the absence of overlapping
Find and the most like suspicious pedestrian's object of pedestrian target, recognized again with completing across video camera pedestrian in the video camera B in domain, at this
In the embodiment of invention, this method comprises the following steps：
Step 1：The training data of transmission range metric learning, the pedestrian's picture set occurred in video camera A is defined
For query setAnd by the pedestrian's picture set Candidate Set occurred in video camera BWhereinIt is the characteristic vector of i-th inquiry picture and jth candidate's picture, andWithFor the identity label of correspondence pedestrian.N is the picture sum of training stage query set, and m is the figure of training stage Candidate Set
Piece sum；
Step 2：Choose the inquiry picture x in query setiWith candidate's picture y in Candidate SetjSample is constituted to (xi,yj),
Be sample to distribution two tag along sort zij, wherein whenWhen zij=1, (xi,yj) it is referred to as positive sample pair, and work asWhen zij=-1, (xi,yj) it is referred to as negative sample pair.Arbitrary sample is defined to (xi,yj) between mahalanobis distance measurement
Function is：
Wherein M is the distance matrix metric in mahalanobis distance.
Step 3：Using Logistic loss functions be training set in each sample to (xi,yj) set up distance metric
The loss function of habit：
Wherein ξ is all samples in training set to the Euclidean distance average between characteristic vector.As (xi,yj) it is positive sample
Pair when, loss function constraint mahalanobis distance dM(xi,yj) be less thanAnd work as (xi,yj) be negative sample pair when, loss function constraint horse
Family name is apart from dM(xi,yj) be more than
Step 4：The loss function constraint between all samples pair in training set is considered simultaneously, defines adaptive distance metric
The overall goal function of study is：
Step 5：It is the loss function ψ (x of different samples pair according to the otherness of training samplei,yj) the different instruction of distribution
Practice weight wij, it is changed into the object function that adaptive distance metric learns：
In the present embodiment, the otherness according to training sample described in step 5, is the loss function ψ of different samples pair
(xi,yj) the different training weight w of distributionij, it is as follows that it implements process：
Step 5.1：For the characteristic vector x of the inquiry picture in training seti, calculate Candidate SetIn all candidate's pictures
Characteristic vector and xiIn the characteristic distance of theorem in Euclid space, then the order according to distance from small to large is rightIn picture arranged
Sequence；
Step 5.2：According to the ranking results in step 5.1, by Candidate SetIt is divided into for xiDifficult setMedium setAnd simple set
Difficulty setDefinition be:
In formula, ri(yj) represent candidate's picture yjPosition in the sequencing queue that step 5.1 is obtained, andThen represent
xiIn Candidate SetIn correct matching y+Position in sequencing queue.
Medium setDefinition be：
Wherein m is Candidate SetThe sum of middle pedestrian's picture number, namely in step 5.1 sequencing queue total length.
Simple setDefinition be：
Step 5.3：According to Candidate Set in step 5.2Division result, be sample to (xi,yj) loss function it is adaptive
Training weight w should be distributedij.As (xi,yj) be positive sample pair when, by ψ (xi,yj) training weight wijIt is set to 1/N+, wherein N+For
The sum of positive sample pair in training set.And work as (xi,yj) be negative sample pair when, its loss function ψ (x are defined by following formulai,yj)
Training weight wij：
Wherein, N-The negative sample that weight is not zero in expression training set is to sum, β1And β2For regulation training weight change model
The balance parameters enclosed, in the present embodiment β1=0.75 and β2=0.25.
Step 6：According to object function, adaptive distance metric study is defined as optimization problem：
Step 7：Using the optimization problem accelerated in near-end gradient algorithm solution procedure 6, corresponding distance metric square is obtained
Battle array M.
In the present embodiment, the use described in step 7 accelerates the optimization problem in near-end gradient algorithm solution procedure 6,
Corresponding distance matrix metric M is obtained, it is as follows that it implements step：
Step 7.1：Initialize M0And M-1For unit matrix, and renewal is iterated to it；
Step 7.2：Specifically, for the t times iteration, calculate first before polymerization to matrix St：
Wherein coefficient0 is set as in first time iteration；
Step 7.3：Define matrixWhereinIt is the mesh of learning distance metric
Scalar functions Ψ (M) is in StThe gradient at place, ηtFor iteration step length；In the present embodiment, ηt2 are initialized in each iteration8,
And check whether following condition meets：
Wherein<·,·>The inner product of representing matrix, | | | |FThe Frobenius norms of representing matrix, when the bar in above formula
When part is met, η is usedtIt is used as iteration step length；Otherwise η is usedt/ 2 replace ηt, until condition is met；
Step 7.4：To matrix PtEigenvalues Decomposition is carried out, is denoted asThen to the t times repeatedly
The distance matrix metric M in generation is updated, and obtains iteration result Mt：
WhereinFor ensuring distance matrix metric MtIt is positive semi-definite；
Step 7.5：Repeat step 7.2 to 7.4, until reaching the condition of convergence：
Step 7.6：Return to M during convergencetThe training result M learnt as adaptive distance metric.
Step 8：After distance matrix metric M is obtained, for any pedestrian target Prototype drawing occurred in video camera A
Piece, extracts its characteristic vector xp, while the unknown pedestrian's picture set of the identity occurred in video camera B is defined as into candidate
Collection, extracts the characteristic vector of every candidate's picture, composition characteristic Vector GroupsN is total for the picture of test phase Candidate Set
Number.The mahalanobis distance metric function that the distance matrix metric M tried to achieve in step 7 is substituted into step 2, calculates x respectivelypWith every
The characteristic vector of candidate's pictureBetween mahalanobis distanceAnd candidate's picture is entered according to mahalanobis distance size
Row sequence, makes and xpThe less candidate's picture of mahalanobis distance comes the front end of queue.Finally, more forward candidate's picture is sorted more
It is likely to be correct matching of the pedestrian target in video camera B.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
Any modifications, equivalent substitutions and improvements made within refreshing and principle etc., should be included in the scope of the protection.
Claims (3)
1. a kind of across video camera pedestrian recognition methods again learnt based on adaptive distance metric, it is characterised in that including following step
Suddenly：
Step 1：The training data of transmission range metric learning, query set is expressed as by pedestrian's picture under different camerasAnd Candidate SetWhereinBe i-th inquiry picture and
The characteristic vector of jth candidate's picture, andWithFor the identity label of correspondence pedestrian, n is the picture of training stage query set
Sum, m is total for the picture of training stage Candidate Set；
Step 2：Choose the inquiry picture x in query setiWith candidate's picture y in Candidate SetjSample is constituted to (xi,yj), it is sample
This is to two tag along sort z of distributionij, wherein whenWhen zij=1, (xi,yj) it is referred to as positive sample pair, and work asWhen
zij=-1, (xi,yj) it is referred to as negative sample pair；Arbitrary sample is defined to (xi,yj) between mahalanobis distance metric function be：
<mrow>
<msub>
<mi>d</mi>
<mi>M</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mi>M</mi>
<mn>2</mn>
</msubsup>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
<mi>M</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Wherein M is the distance matrix metric in mahalanobis distance；
Step 3：Using Logistic loss functions be training set in each sample to (xi,yj) set up learning distance metric
Loss function：
<mrow>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>l</mi>
<mi>o</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>+</mo>
<msup>
<mi>e</mi>
<mrow>
<msub>
<mi>z</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>d</mi>
<mi>M</mi>
</msub>
<mo>(</mo>
<mrow>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mi>&xi;</mi>
<mo>)</mo>
</mrow>
</mrow>
</msup>
<mo>)</mo>
</mrow>
</mrow>
Wherein ξ be training set in all samples to the Euclidean distance average between characteristic vector, as (xi,yj) be positive sample pair when,
Loss function constraint mahalanobis distance dM(xi,yj) it is less than ξ；And work as (xi,yj) be negative sample pair when, loss function constraint mahalanobis distance
dM(xi,yj) it is more than ξ；
Step 4：The loss function constraint between all samples pair in training set is considered simultaneously, defines adaptive distance metric study
Overall goal function be：
<mrow>
<mi>&Psi;</mi>
<mrow>
<mo>(</mo>
<mi>M</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</msubsup>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Step 5：It is the loss function ψ (x of different samples pair according to the otherness of training samplei,yj) the different training power of distribution
Weight wij, it is changed into the object function that adaptive distance metric learns：
<mrow>
<mi>&Psi;</mi>
<mrow>
<mo>(</mo>
<mi>M</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</msubsup>
<msub>
<mi>w</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>&CenterDot;</mo>
<mi>&psi;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>y</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
Step 6：According to object function, adaptive distance metric study is defined as optimization problem：
<mrow>
<munder>
<mi>min</mi>
<mi>M</mi>
</munder>
<mi>&Psi;</mi>
<mrow>
<mo>(</mo>
<mi>M</mi>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mi>s</mi>
<mo>.</mo>
<mi>t</mi>
<mo>.</mo>
<mi>M</mi>
<mo>&GreaterEqual;</mo>
<mn>0</mn>
</mrow>
Step 7：Using the optimization problem accelerated in near-end gradient algorithm solution procedure 6, corresponding distance matrix metric M is obtained；
Step 8：In test phase, for the characteristic vector x of given query template picturepWith suspicious pedestrian under other video cameras
The characteristic vector group of candidate's picture of target configurationN is total for the picture of test phase Candidate Set, by distance metric
The mahalanobis distance metric function that matrix M is substituted into step 2, calculates x respectivelypWith the characteristic vector of every candidate's pictureBetween
Mahalanobis distanceAnd candidate's picture is ranked up according to mahalanobis distance size, make and xpMahalanobis distance is less
Candidate's picture comes the front end of queue.
2. a kind of across video camera pedestrian recognition methods again learnt based on adaptive distance metric according to claim 1,
It is characterized in that：The otherness according to training sample described in step 5, is the loss function ψ (x of different samples pairi,yj) point
With different training weight wij, the process of implementing is：
Step 5.1：For the characteristic vector x of the inquiry picture in training seti, calculate Candidate SetIn all candidate's picture features
Vector and xiIn the characteristic distance of theorem in Euclid space, then the order according to distance from small to large is rightIn picture be ranked up；
Step 5.2：According to the ranking results in step 5.1, by Candidate SetIt is divided into for xiDifficult setIn
Deng setAnd simple set
Difficulty setDefinition be：
In formula, ri(yj) represent candidate's picture yjPosition in the sequencing queue that step 5.1 is obtained, andThen represent xi
Candidate SetIn correct matching y+Position in sequencing queue；
Medium setDefinition be：
Wherein m is Candidate SetThe sum of middle pedestrian's picture number, namely in step 5.1 sequencing queue total length；
Simple setDefinition be：
Step 5.3：According to Candidate Set in step 5.2Division result, be sample to (xi,yj) loss function adaptively divide
With training weight wij；As (xi,yj) be positive sample pair when, by ψ (xi,yj) training weight wijIt is set to 1/N+, wherein N+For training
Concentrate the sum of positive sample pair；And work as (xi,yj) be negative sample pair when, its loss function ψ (x are defined by following formulai,yj) instruction
Practice weight wij：
Wherein, N-The negative sample that weight is not zero in expression training set is to sum, β1And β2For regulation training weight excursion
Balance parameters.
3. a kind of across video camera pedestrian recognition methods again learnt based on adaptive distance metric according to claim 1,
It is characterized in that：Use described in step 7 accelerates the optimization problem in near-end gradient algorithm solution procedure 6, obtains corresponding
Distance matrix metric M, it specifically includes following steps：
Step 7.1：Initialize M0And M-1For unit matrix, and renewal is iterated to it；
Step 7.2：Specifically, for the t times iteration, calculate first before polymerization to matrix st：
<mrow>
<msub>
<mi>S</mi>
<mi>t</mi>
</msub>
<mo>=</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>+</mo>
<mfrac>
<mrow>
<msub>
<mi>&alpha;</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>-</mo>
<mn>1</mn>
</mrow>
<msub>
<mi>&alpha;</mi>
<mi>t</mi>
</msub>
</mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>2</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
2
Wherein coefficient0 is set as in first time iteration；
Step 7.3：Define matrixWhereinIt is the overall mesh of learning distance metric
Scalar functions Ψ (M) is in stThe gradient at place, ηtFor iteration step length；
Step 7.4：To matrix PtEigenvalues Decomposition is carried out, is denoted asThen to the t times iteration away from
It is updated from metric matrix M, obtains iteration result Mt：
<mrow>
<msub>
<mi>M</mi>
<mi>t</mi>
</msub>
<mo>=</mo>
<msub>
<mi>U</mi>
<mi>t</mi>
</msub>
<msubsup>
<mi>&Lambda;</mi>
<mi>t</mi>
<mo>+</mo>
</msubsup>
<msubsup>
<mi>U</mi>
<mi>t</mi>
<mi>T</mi>
</msubsup>
</mrow>
WhereinFor ensuring distance matrix metric MtIt is positive semi-definite；
Step 7.5：Repeat step 7.2 to 7.4, until reaching the condition of convergence：
<mrow>
<mfrac>
<mrow>
<mo>|</mo>
<mi>&Psi;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>M</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>&Psi;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>|</mo>
</mrow>
<mrow>
<mo>|</mo>
<mi>&Psi;</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>M</mi>
<mrow>
<mi>t</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>|</mo>
</mrow>
</mfrac>
<mo>&le;</mo>
<msup>
<mn>10</mn>
<mrow>
<mo>-</mo>
<mn>3</mn>
</mrow>
</msup>
</mrow>
Step 7.6：Return to M during convergencetThe training result M learnt as adaptive distance metric.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710213901.2A CN107145827A (en) | 2017-04-01 | 2017-04-01 | Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710213901.2A CN107145827A (en) | 2017-04-01 | 2017-04-01 | Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107145827A true CN107145827A (en) | 2017-09-08 |
Family
ID=59773501
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710213901.2A Pending CN107145827A (en) | 2017-04-01 | 2017-04-01 | Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107145827A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563452A (en) * | 2017-09-18 | 2018-01-09 | 天津师范大学 | A kind of cross-domain ground cloud atlas sorting technique that study is estimated based on differentiation |
| CN107704890A (en) * | 2017-10-27 | 2018-02-16 | 北京旷视科技有限公司 | A kind of generation method and device of four-tuple image |
| CN107909049A (en) * | 2017-11-29 | 2018-04-13 | 广州大学 | Pedestrian's recognition methods again based on least square discriminant analysis distance study |
| CN107944340A (en) * | 2017-10-25 | 2018-04-20 | 宁波大学 | A kind of combination is directly measured and the pedestrian of indirect measurement recognition methods again |
| CN108021898A (en) * | 2017-12-14 | 2018-05-11 | 西北工业大学 | A kind of personage's recognition methods again that study is measured based on adapted distance |
| CN108133230A (en) * | 2017-12-14 | 2018-06-08 | 西北工业大学 | A kind of personage's recognition methods again of object-oriented personage's distance measure study |
| CN108229555A (en) * | 2017-12-29 | 2018-06-29 | 深圳云天励飞技术有限公司 | Sample weights distribution method, model training method, electronic equipment and storage medium |
| CN108985168A (en) * | 2018-06-15 | 2018-12-11 | 江南大学 | A kind of video face identification method based on the study of minimum normalized cumulant |
| CN109492702A (en) * | 2018-11-21 | 2019-03-19 | 中国科学院自动化研究所 | Pedestrian based on sorting measure function recognition methods, system, device again |
| CN109934849A (en) * | 2019-03-08 | 2019-06-25 | 西北工业大学 | Online multi-object tracking method based on track metric learning |
| CN109977882A (en) * | 2019-03-29 | 2019-07-05 | 广东石油化工学院 | A kind of half coupling dictionary is to the pedestrian of study again recognition methods and system |
| CN110287822A (en) * | 2019-06-10 | 2019-09-27 | 浙江大学城市学院 | The snail pest control method of view-based access control model target detection in a kind of officinal dendrobium stem plantation |
| CN111680697A (en) * | 2020-03-17 | 2020-09-18 | 北京大学 | Method, apparatus, electronic device, and medium for implementing domain adaptation |
| CN112101217A (en) * | 2020-09-15 | 2020-12-18 | 镇江启迪数字天下科技有限公司 | Pedestrian re-identification method based on semi-supervised learning |
| CN112381056A (en) * | 2020-12-02 | 2021-02-19 | 山西大学 | Cross-domain pedestrian re-identification method and system fusing multiple source domains |
| CN113361568A (en) * | 2021-05-18 | 2021-09-07 | 北京迈格威科技有限公司 | Target identification method, device and electronic system |
| CN113536003A (en) * | 2021-06-08 | 2021-10-22 | 支付宝(杭州)信息技术有限公司 | Feature extraction model training method, image retrieval method, device and equipment |
| CN115965817A (en) * | 2023-01-05 | 2023-04-14 | 北京百度网讯科技有限公司 | Training method and device of image classification model and electronic equipment |
| CN116244612A (en) * | 2023-05-12 | 2023-06-09 | 国网江苏省电力有限公司信息通信分公司 | HTTP traffic clustering method and device based on self-learning parameter measurement |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850857A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Trans-camera pedestrian target matching method based on visual space significant constraints |
-
2017
- 2017-04-01 CN CN201710213901.2A patent/CN107145827A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850857A (en) * | 2015-05-15 | 2015-08-19 | 浙江大学 | Trans-camera pedestrian target matching method based on visual space significant constraints |
Non-Patent Citations (1)
| Title |
|---|
| YI XIE等: "Adaptive Metric Learning and Probe-Specific Reranking for Person Reidentification", 《IEEE SIGNAL PROCESSING LETTERS》 * |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563452B (en) * | 2017-09-18 | 2020-03-27 | 天津师范大学 | Cross-domain foundation cloud picture classification method based on discriminant measure learning |
| CN107563452A (en) * | 2017-09-18 | 2018-01-09 | 天津师范大学 | A kind of cross-domain ground cloud atlas sorting technique that study is estimated based on differentiation |
| CN107944340A (en) * | 2017-10-25 | 2018-04-20 | 宁波大学 | A kind of combination is directly measured and the pedestrian of indirect measurement recognition methods again |
| CN107944340B (en) * | 2017-10-25 | 2020-06-16 | 宁波大学 | Pedestrian re-identification method combining direct measurement and indirect measurement |
| CN107704890A (en) * | 2017-10-27 | 2018-02-16 | 北京旷视科技有限公司 | A kind of generation method and device of four-tuple image |
| CN107909049A (en) * | 2017-11-29 | 2018-04-13 | 广州大学 | Pedestrian's recognition methods again based on least square discriminant analysis distance study |
| CN107909049B (en) * | 2017-11-29 | 2020-07-31 | 广州大学 | Pedestrian re-identification method based on least square discriminant analysis distance learning |
| CN108021898A (en) * | 2017-12-14 | 2018-05-11 | 西北工业大学 | A kind of personage's recognition methods again that study is measured based on adapted distance |
| CN108133230A (en) * | 2017-12-14 | 2018-06-08 | 西北工业大学 | A kind of personage's recognition methods again of object-oriented personage's distance measure study |
| CN108229555A (en) * | 2017-12-29 | 2018-06-29 | 深圳云天励飞技术有限公司 | Sample weights distribution method, model training method, electronic equipment and storage medium |
| CN108985168A (en) * | 2018-06-15 | 2018-12-11 | 江南大学 | A kind of video face identification method based on the study of minimum normalized cumulant |
| CN108985168B (en) * | 2018-06-15 | 2022-05-24 | 江南大学 | Video face recognition method based on minimum normalized distance learning |
| CN109492702A (en) * | 2018-11-21 | 2019-03-19 | 中国科学院自动化研究所 | Pedestrian based on sorting measure function recognition methods, system, device again |
| CN109492702B (en) * | 2018-11-21 | 2020-09-22 | 中国科学院自动化研究所 | Pedestrian re-identification method, system and device based on ranking measurement function |
| CN109934849A (en) * | 2019-03-08 | 2019-06-25 | 西北工业大学 | Online multi-object tracking method based on track metric learning |
| CN109977882B (en) * | 2019-03-29 | 2019-12-03 | 广东石油化工学院 | A kind of half coupling dictionary is to the pedestrian of study again recognition methods and system |
| CN109977882A (en) * | 2019-03-29 | 2019-07-05 | 广东石油化工学院 | A kind of half coupling dictionary is to the pedestrian of study again recognition methods and system |
| CN110287822A (en) * | 2019-06-10 | 2019-09-27 | 浙江大学城市学院 | The snail pest control method of view-based access control model target detection in a kind of officinal dendrobium stem plantation |
| CN111680697A (en) * | 2020-03-17 | 2020-09-18 | 北京大学 | Method, apparatus, electronic device, and medium for implementing domain adaptation |
| CN111680697B (en) * | 2020-03-17 | 2024-03-22 | 北京大学 | Method, device, electronic equipment and medium for realizing field adaptation |
| CN112101217A (en) * | 2020-09-15 | 2020-12-18 | 镇江启迪数字天下科技有限公司 | Pedestrian re-identification method based on semi-supervised learning |
| CN112101217B (en) * | 2020-09-15 | 2024-04-26 | 镇江启迪数字天下科技有限公司 | Pedestrian re-identification method based on semi-supervised learning |
| CN112381056A (en) * | 2020-12-02 | 2021-02-19 | 山西大学 | Cross-domain pedestrian re-identification method and system fusing multiple source domains |
| CN113361568A (en) * | 2021-05-18 | 2021-09-07 | 北京迈格威科技有限公司 | Target identification method, device and electronic system |
| CN113536003B (en) * | 2021-06-08 | 2024-03-12 | 支付宝(杭州)信息技术有限公司 | Feature extraction model training method, image retrieval method, device and equipment |
| CN113536003A (en) * | 2021-06-08 | 2021-10-22 | 支付宝(杭州)信息技术有限公司 | Feature extraction model training method, image retrieval method, device and equipment |
| CN115965817A (en) * | 2023-01-05 | 2023-04-14 | 北京百度网讯科技有限公司 | Training method and device of image classification model and electronic equipment |
| CN116244612A (en) * | 2023-05-12 | 2023-06-09 | 国网江苏省电力有限公司信息通信分公司 | HTTP traffic clustering method and device based on self-learning parameter measurement |
| CN116244612B (en) * | 2023-05-12 | 2023-08-29 | 国网江苏省电力有限公司信息通信分公司 | HTTP traffic clustering method and device based on self-learning parameter measurement |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107145827A (en) | Across the video camera pedestrian recognition methods again learnt based on adaptive distance metric | |
| Huang et al. | Building deep networks on grassmann manifolds | |
| CN107194341B (en) | Face recognition method and system based on fusion of Maxout multi-convolution neural network | |
| CN112308158A (en) | Multi-source field self-adaptive model and method based on partial feature alignment | |
| CN105069481B (en) | Natural scene multiple labeling sorting technique based on spatial pyramid sparse coding | |
| CN110334806A (en) | A kind of confrontation sample generating method based on production confrontation network | |
| CN105205449B (en) | Sign Language Recognition Method based on deep learning | |
| CN111985581B (en) | Sample-level attention network-based few-sample learning method | |
| CN112101176A (en) | User identity recognition method and system combining user gait information | |
| CN105138998B (en) | Pedestrian based on the adaptive sub-space learning algorithm in visual angle recognition methods and system again | |
| CN107808113B (en) | Facial expression recognition method and system based on differential depth features | |
| CN107122375A (en) | The recognition methods of image subject based on characteristics of image | |
| CN104408405B (en) | Face representation and similarity calculating method | |
| CN103065158B (en) | The behavior recognition methods of the ISA model based on relative gradient | |
| CN108062574A (en) | A kind of Weakly supervised object detection method based on particular category space constraint | |
| CN106709528A (en) | Method and device of vehicle reidentification based on multiple objective function deep learning | |
| CN109284406A (en) | Intension recognizing method based on difference Recognition with Recurrent Neural Network | |
| CN107316059A (en) | Learner's gesture recognition method | |
| CN110363156A (en) | A kind of Facial action unit recognition methods that posture is unrelated | |
| CN113553908A (en) | Heterogeneous iris identification method based on equipment unique perception | |
| CN109145947A (en) | A kind of Fashionable women dress image fine grit classification method based on component detection and visual signature | |
| CN107330363A (en) | A kind of quick Internet advertising board detection method | |
| CN114492634A (en) | Fine-grained equipment image classification and identification method and system | |
| Huynh et al. | An efficient model for copy-move image forgery detection | |
| Tan et al. | Wide Residual Network for Vision-based Static Hand Gesture Recognition. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170908 |