CN110874576B - Pedestrian re-identification method based on typical correlation analysis fusion characteristics - Google Patents
Pedestrian re-identification method based on typical correlation analysis fusion characteristics Download PDFInfo
- Publication number
- CN110874576B CN110874576B CN201911114451.7A CN201911114451A CN110874576B CN 110874576 B CN110874576 B CN 110874576B CN 201911114451 A CN201911114451 A CN 201911114451A CN 110874576 B CN110874576 B CN 110874576B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- matrix
- features
- identification
- mapping
- 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.)
- Active
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
- G06F18/22—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
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
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 pedestrian re-identification method based on typical correlation analysis fusion characteristics, which comprises three stages: a feature extraction stage, a mapping matrix solving stage and a pedestrian re-identification stage by fusing features; in the feature extraction stage, extracting two different features X and Y from a pedestrian image; in the stage of solving the mapping matrix, the two features X and Y are respectively subjected to typical correlation analysis to obtain a pair of mapping matrices α and β, and the new feature is expressed as X' =α T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β; in the pedestrian re-recognition stage of the fusion feature, the fusion feature can be expressed asOr Z is 2 =x '+y', fusing feature Z 1 Or Z is 2 The method is divided into a training set and a testing set, the training set is used for training the model of pedestrian re-identification, and the testing set is used for testing the trained model, so that redundant information is effectively eliminated while characteristics are fused, and the calculated amount and difficulty are reduced.
Description
Technical Field
The invention belongs to the technical field of computer vision, and relates to a pedestrian re-identification method based on typical correlation analysis fusion characteristics.
Background
Pedestrian re-identification is a very popular research topic in the field of computer vision, which aims to give a pedestrian of interest, and find it out in non-overlapping monitoring devices through computer vision technology. Most existing methods, when solving the problem of pedestrian re-identification, mainly start from the following two aspects: 1 developing a distinguishing characteristic representation; 2 seek a distinguishing distance measure. The method based on the feature representation aims at extracting the feature with robustness to represent the pedestrian, and the features used for the pedestrian re-identification can be divided into three types: visual features, filter features, attribute features. The metric-based learning approach consists in learning the similarity between two pictures. The application of pedestrian re-identification is generally that on the basis of feature representation, similarity between features is utilized to judge the similarity between pedestrian images, and the distance between the same pedestrians is as small as possible and the distance between different pedestrians is as large as possible by learning a distance measurement function with stronger distinguishing power.
The characteristic is the basis of pedestrian re-identification, and the recognition force of the characteristic directly influences the final result of pedestrian re-identification. The color feature is the most widely applied feature, can represent the color distribution condition of the pedestrian image, has robustness to the change of the gesture and the visual angle, is easily influenced by illumination and shielding, and is difficult to distinguish by only the color feature when the similar pedestrian image is installed. The texture features have robustness to illumination, and the color texture features are combined, so that the recognition accuracy can be effectively improved. Generally, the characteristic of manual design is to combine a plurality of simple characteristics, and the method combines the advantages of different characteristics for representing pedestrians, so that the recognition effect is good. However, as the number of combined features increases, the dimension of the combined features increases exponentially. The existing method for fusing the features mostly fuses different features together through a serial or parallel strategy, and has the defects that the method does not consider the internal relation among different features when in combination, only stacks the different features, all feature information is reserved when in fusion, a large amount of redundant information is reserved, the dimension of the combined features is high, the calculation complexity is increased, and the accuracy and the instantaneity of identification are affected to a certain extent.
Disclosure of Invention
The invention aims to provide a pedestrian re-identification method based on typical correlation analysis fusion characteristics, which solves the problems of high result dimension, a large amount of redundant information and complex calculation of fusion characteristics in the prior art.
The technical scheme adopted by the invention is that the pedestrian re-identification method based on typical correlation analysis fusion characteristics comprises three stages: a feature extraction stage, a mapping matrix solving stage and a pedestrian re-identification stage by fusing features; in the feature extraction stage, extracting two different features X and Y from a pedestrian image; in the stage of solving the mapping matrix, the two features X and Y are respectively subjected to typical correlation analysis to obtain a pair of mapping matrices α and β, and the new feature is expressed as X' =α T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β; in the pedestrian re-recognition stage of the fusion feature, the fusion feature is expressed asOr Z is 2 =x '+y', fusing feature Z 1 Or Z is 2 The method comprises the steps of dividing a training set and a testing set, training a pedestrian re-identification model by using the training set, and testing the trained model by using the testing set.
The invention is also characterized in that:
the method comprises the following specific steps:
step 1, extracting two characteristics from a pedestrian re-identification data set:
extracting features from the dataset of the pedestrian image by using different feature extraction algorithms, respectively marking as:
X∈R p*N ，Y∈R q*N
p and q represent dimensions of two features, respectively, and N represents the number of pictures contained in the dataset;
step 2, respectively carrying out typical correlation analysis on the two features X and Y extracted in the step 1, and solving by a singular value decomposition method to obtain a pair of featuresMapping matrices α and β, the new feature is expressed as X' =α T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β;
step 3, pedestrian re-identification is carried out by using the fusion characteristics:
step 3.1, obtaining fusion expression of typical relevant characteristics through the following fusion strategy through the mapping matrixes alpha and beta obtained in step 2Or Z is 2 ＝X'+Y'＝α T X+β T Y, feature Z of fusion 1 Or Z is 2 Dividing the different data sets into a training set I and a testing set I of a visual angle I, a training set II and a testing set II of the visual angle II according to the dividing rule of the different data sets in the pedestrian re-identification, training a model of the pedestrian re-identification by using the training set I and the training set II, and testing the trained model by using the testing set I and the testing set II;
and 3.2, evaluating the test result in the step 3.1 by using the cumulative matching curve CMC, and taking the identification rate of rank1 as the most important evaluation index, wherein the larger the value of rank1 is, the better the identification effect is.
In the step 2, a singular value decomposition method is used for solving a projection matrix, and the solving process is as follows:
1) Normalizing the two features to obtain standard data with a mean value of 0 and a variance of 1;
2) Calculating the variance S of X XX Variance S of Y YY Covariance S of X and Y XY ；
3) Calculating a matrix
4) Singular value decomposition is carried out on the matrix M to obtain a maximum singular value sigma, and left and right singular vectors u and v corresponding to the maximum singular value;
5) Calculating mapping matrices alpha and beta for X and Y,
6) The representation of the two features in the correlation subspace is X' =α T X，Y'＝β T Y。
The specific process for solving the projection matrix by using the singular value decomposition method in the step 2 is as follows:
(1) Let the mapping matrices of X and Y be α and β, respectively, their representation in subspace be: x' =α T X and Y' =β T Y, their correlation coefficients can be expressed as:
the objective function is:
namely solving mapping matrixes alpha and beta corresponding to the maximum correlation coefficient;
(2) Before projection, the original data is normalized first to obtain data with mean value of 0 and variance of 1,
Cov(α T X,β T Y)＝E(＜α T X,β T Y＞)＝E((α T X)(β T Y) T )＝α T E(XY T )β
similarly, var (. Beta.) T Y)＝β T E(YY T )β，μ x Is the mean value of X;
(3) Since the average value of X and Y is 0, then
Var(X)＝Cov(X,X)＝E(XX T )
Var(Y)＝Cov(Y,Y)＝E(YY T )
Cov(X,Y)＝E(XY T )
Cov(Y,X)＝E(YX T )；
(4) Let S XX ＝Var(X,X)，S YY ＝Var(Y,Y)，S XY =cov (X, Y), then the objective function is converted into
(5) Since the numerator denominator is increased by the same multiple, the optimization target result is unchanged, the denominator is fixed, and the numerator is optimized, namely:
s.t.α T S XX α＝1,β T S YY β＝1；
(6) When solving the objective function in (5), adopting a singular value decomposition method, wherein u and v are two unit vectors,
order the
Alpha is then T S XX α＝1，β T S YY β＝1，
At the same time, from alpha T S XX α=1, obtainable:
from beta T S YY Beta=1, obtainable:
at this time, the objective function is:
s.t.u T u＝1,v T v＝1；
(7) For the objective function in (6), let the matrixAt this time, U and V represent left and right singular vectors corresponding to a singular value of the matrix M, and m=u Σv is obtained by singular value decomposition T Wherein U, V are respectively the matrix composed of the left singular vector and the right singular vector of M, and Sigma is the diagonal matrix composed of the singular values of M; since all columns of U, V are orthonormal, then U T U and V T v gets a vector with only one scalar 1 and the remaining scalar 0; at this time, the liquid crystal display device,
maximization ofThe corresponding maximum value is the maximum value of the singular values corresponding to a certain group of left and right singular vectors, namely after M is subjected to singular value decomposition, the maximum singular value is the maximum value of an optimization target, namely the maximum correlation coefficient between X and Y;
(8) The mapping matrix of original X and Y is obtained by using the corresponding left and right singular vectors u, v
In the step 3.1, an XQDA algorithm is used in the process of training the pedestrian re-identification model, and a training set and a training sample label are taken as inputs and output as subspace mapping matrixes W and WWherein Sigma 'is' I Is an intra-class covariance matrix, sigma' E Is an inter-class covariance matrix;
during testing, the mahalanobis distance is used for measuring the similarity between two pedestrian images, and the mapping of M and the training features on the subspace W is input to obtain the mahalanobis distance of the original features on the subspace.
The beneficial effects of the invention are as follows: the invention discloses a pedestrian re-identification method based on characteristic fusion of typical correlation analysis. Aiming at the problems of high dimension, large amount of redundant information and complex calculation of the fusion result of the current fusion feature method, the typical correlation analysis algorithm is used for analyzing the internal relations among different features of the same target and searching for a linear combination of the features, so that most of information of the original features is reserved by the new features, and the new features have the maximum correlation with the other new features. The two new features are fused according to a certain strategy, so that the purpose of feature fusion is achieved, and redundant information among the features is eliminated.
Drawings
FIG. 1 is a feature fusion process diagram of a pedestrian re-identification method based on typical correlation analysis fusion features of the present invention;
FIG. 2 is a graph of the results of two features of a pedestrian re-recognition method based on a typical correlation analysis fusion feature and the fusion feature on a VIPER dataset in accordance with the present invention;
fig. 3 is a specific result diagram when rank1, rank5, rank10, rank20 are used as evaluation indexes in fig. 2.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
The invention relates to a pedestrian re-identification method based on typical correlation analysis fusion characteristics, which comprises three stages as shown in fig. 1: a feature extraction stage, a mapping matrix solving stage and a pedestrian re-identification stage by fusing features; in the feature extraction stage, extracting two different features X and Y from a pedestrian image; in the stage of solving the mapping matrix, the two features X and Y are respectively subjected to typical correlation analysis to obtain a pair of mapping matrices α and β, and the new feature is expressed as X' =α T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β; in fusion withThe pedestrian re-identification stage is carried out by combining features, and the combined features are expressed asOr Z is 2 =x '+y', fusing feature Z 1 Or Z is 2 The method comprises the steps of dividing a training set and a testing set, training a pedestrian re-identification model by using the training set, and testing the trained model by using the testing set.
The invention discloses a pedestrian re-identification method based on typical correlation analysis fusion characteristics, which comprises the following specific steps:
step 1, extracting two characteristics from a pedestrian re-identification data set:
extracting features from the dataset of the pedestrian image by using different feature extraction algorithms, respectively marking as:
X∈R p*N ，Y∈R q*N
p and q represent dimensions of two features, respectively, and N represents the number of pictures contained in the dataset;
step 2, respectively performing typical correlation analysis on the two features X and Y extracted in the step 1, and solving by using a singular value decomposition method to obtain a pair of mapping matrixes alpha and beta, wherein the new features are expressed as X' =alpha T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β;
step 3, pedestrian re-identification is carried out by using the fusion characteristics:
step 3.1, obtaining fusion expression of typical relevant characteristics through the following fusion strategy through the mapping matrixes alpha and beta obtained in step 2Or Z is 2 ＝X'+Y'＝α T X+β T Y, feature Z of fusion 1 Or Z is 2 Dividing the different data sets into a training set I and a testing set I of a visual angle I, a training set II and a testing set II of the visual angle II according to the dividing rule of the different data sets in the pedestrian re-identification, training a model of the pedestrian re-identification by using the training set I and the training set II, and testing the trained model by using the testing set I and the testing set II;
and 3.2, evaluating the test result in the step 3.1 by using the cumulative matching curve CMC, and taking the identification rate of rank1 as the most important evaluation index, wherein the larger the value of rank1 is, the better the identification effect is.
In the step 2, a singular value decomposition method is used for solving a projection matrix, and the solving process is as follows:
1) Normalizing the two features to obtain standard data with a mean value of 0 and a variance of 1;
2) Calculating the variance S of X XX Variance S of Y YY Covariance S of X and Y XY ；
3) Calculating a matrix
4) Singular value decomposition is carried out on the matrix M to obtain a maximum singular value sigma, and left and right singular vectors u and v corresponding to the maximum singular value;
5) Calculating mapping matrices alpha and beta for X and Y,
6) The representation of the two features in the correlation subspace is X' =α T X，Y'＝β T Y。
The specific process for solving the projection matrix by using the singular value decomposition method in the step 2 is as follows:
(1) Let the mapping matrices of X and Y be α and β, respectively, their representation in subspace be: x' =α T X and Y' =β T Y, their correlation coefficients can be expressed as:
the objective function is:
namely solving mapping matrixes alpha and beta corresponding to the maximum correlation coefficient;
(2) Before projection, the original data is normalized first to obtain data with mean value of 0 and variance of 1,
Cov(α T X,β T Y)＝E(＜α T X,β T Y＞)＝E((α T X)(β T Y) T )＝α T E(XY T )β
similarly, var (. Beta.) T Y)＝β T E(YY T )β，μ x Is the mean value of X;
(3) Since the average value of X and Y is 0, then
Var(X)＝Cov(X,X)＝E(XX T )
Var(Y)＝Cov(Y,Y)＝E(YY T )
Cov(X,Y)＝E(XY T )
Cov(Y,X)＝E(YX T )；
(4) Let S XX ＝Var(X,X)，S YY ＝Var(Y,Y)，S XY =cov (X, Y), then the objective function is converted into
(5) Since the numerator denominator is increased by the same multiple, the optimization target result is unchanged, the denominator is fixed, and the numerator is optimized, namely:
s.t.α T S XX α＝1,β T S YY β＝1；
(6) When solving the objective function in (5), adopting a singular value decomposition method, wherein u and v are two unit vectors,
order the
Alpha is then T S XX α＝1，β T S YY β＝1，
At the same time, from alpha T S XX α=1, obtainable:
from beta T S YY Beta=1, obtainable:
at this time, the objective function is:
s.t.u T u＝1,v T v＝1；
(7) For the objective function in (6), let the matrixAt this time, U and V represent left and right singular vectors corresponding to a singular value of the matrix M, and m=u Σv is obtained by singular value decomposition T Wherein U, V are respectively the matrix composed of the left singular vector and the right singular vector of M, and Sigma is the diagonal matrix composed of the singular values of M; since all columns of U, V are orthonormal, then U T U and V T v gets a vector with only one scalar 1 and the remaining scalar 0; at this time, the liquid crystal display device,
maximization ofThe corresponding maximum value is the maximum value of the singular values corresponding to a certain group of left and right singular vectors, namely after M is subjected to singular value decomposition, the maximum singular value is the maximum value of an optimization target, namely the maximum correlation coefficient between X and Y;
(8) The mapping matrix of original X and Y is obtained by using the corresponding left and right singular vectors u, v
In the step 3.1, an XQDA algorithm is used in the process of training the pedestrian re-identification model, and a training set and a training sample label are taken as inputs and output as subspace mapping matrixes W and WWherein Sigma 'is' I Is an intra-class covariance matrix, sigma' E Is an inter-class covariance matrix;
during testing, the mahalanobis distance is used for measuring the similarity between two pedestrian images, and the mapping of M and the training features on the subspace W is input to obtain the mahalanobis distance of the original features on the subspace.
The pedestrian re-identification method based on typical correlation analysis fusion characteristics has the advantages that: in the method, a typical correlation analysis and fusion strategy is adopted in a feature fusion stage, so that the maximum correlation of different spatial features in a public subspace is analyzed, the maximum correlation feature between two features is used as discrimination information, redundant information is effectively eliminated while features are fused, and the calculated amount and difficulty are reduced.
Example 1
The invention discloses a pedestrian re-identification method based on typical correlation analysis fusion characteristics, which is implemented according to the following steps:
step 1: extracting two features from pedestrian re-identification dataset
A pedestrian re-identification dataset VIPeR is used, which contains 632 pairs of images of pedestrians, 1264 in total, each pair containing two pictures of a person from different perspectives, and each image being scaled to a size of 128 x 48 pixels, the dataset being extracted WHOS (Weighted Histogram of Overlapping Stripes) features and LOMO (Local Maximal Occurrence) features in combination with existing feature extraction means.
Step 2: the characteristic is subjected to typical correlation analysis, a mapping matrix is solved, and the process of solving by a singular value decomposition method is as follows:
1) Normalizing the two features to obtain standard data with a mean value of 0 and a variance of 1;
2) Calculating the variance S of X XX Variance S of Y YY Covariance S of X and Y XY ；
3) Calculating a matrix
4) Singular value decomposition is carried out on the matrix M to obtain a maximum singular value sigma, and left and right singular vectors u and v corresponding to the maximum singular value;
5) Calculating mapping matrices alpha and beta for X and Y,
6) The representation of the two features in the correlation subspace is X' =α T X，Y'＝β T Y。
Step 3: the fusion characteristics are used for pedestrian re-identification, and the specific process is as follows:
1) Fusion feature Z ε R d*N Where d is the dimension of the fused feature, N is the number of pictures contained in the dataset, and the fused feature is represented asOr Z is 2 ＝X'+Y'＝α T X+β T For the VIPeR dataset, n=1264, with 1-632 columns of features as the query set and 633-1264 columns of features as the candidate set;
2) For the query set and the candidate set, 316 columns of features are randomly selected as two training sets, and the rest 316 columns are two test sets;
3) The recognition process uses an XQDA (Cross-view Quadratic Discriminant Analysis) algorithm, takes a training set and a training sample label as inputs and outputs as subspace mapping matrices W andwherein Sigma 'is' I Is an intra-class covariance matrix, sigma' E Is an inter-class covariance matrix;
4) During testing, the similarity between two pedestrian images is measured by using the mahalanobis distance, and the mahalanobis distance of the original feature in the subspace can be obtained by inputting the M and the mapping of the training feature on the subspace W;
5) As shown in fig. 2 and 3, CMC curves are used as evaluation results, and rank1, rank5, rank10, rank20 are used as evaluation indexes, wherein the rank1 value is particularly important in evaluating the effect of pedestrian re-recognition.
Claims (5)
1. A pedestrian re-identification method based on typical correlation analysis fusion characteristics is characterized by comprising three stages: a feature extraction stage, a mapping matrix solving stage and a pedestrian re-identification stage by fusing features; in the feature extraction stage, extracting two different features X and Y from a pedestrian image; in the stage of solving the mapping matrix, the two features X and Y are respectively subjected to typical correlation analysis to obtain a pair of mapping matrices α and β, and the new feature is expressed as X' =α T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β; in the pedestrian re-recognition stage of the fusion feature, the fusion feature is expressed asOr Z is 2 =x '+y', fusing feature Z 1 Or Z is 2 The method comprises the steps of dividing a training set and a testing set, training a pedestrian re-identification model by using the training set, and testing the trained model by using the testing set.
2. The pedestrian re-identification method based on typical correlation analysis fusion characteristics according to claim 1, wherein the specific steps are as follows:
step 1, extracting two characteristics from a pedestrian re-identification data set:
extracting features from the dataset of the pedestrian image by using different feature extraction algorithms, respectively marking as:
X∈R p*N ，Y∈R q*N
p and q represent dimensions of two features, respectively, and N represents the number of pictures contained in the dataset;
step 2, respectively performing typical correlation analysis on the two features X and Y extracted in the step 1, and solving by using a singular value decomposition method to obtain a pair of mapping matrixes alpha and beta, wherein the new features are expressed as X' =alpha T X，Y'＝β T Y，α T To map the transpose of matrix alpha, beta T Transpose of the mapping matrix β;
step 3, pedestrian re-identification is carried out by using the fusion characteristics:
step 3.1, obtaining fusion expression of typical relevant characteristics through the following fusion strategy through the mapping matrixes alpha and beta obtained in step 2Or Z is 2 ＝X'+Y'＝α T X+β T Y, feature Z of fusion 1 Or Z is 2 Dividing the different data sets into a training set I and a testing set I of a visual angle I, a training set II and a testing set II of the visual angle II according to the dividing rule of the different data sets in the pedestrian re-identification, training a model of the pedestrian re-identification by using the training set I and the training set II, and testing the trained model by using the testing set I and the testing set II;
and 3.2, evaluating the test result in the step 3.1 by using the cumulative matching curve CMC, and taking the identification rate of rank1 as the most important evaluation index, wherein the larger the value of rank1 is, the better the identification effect is.
3. The pedestrian re-recognition method based on the typical correlation analysis fusion characteristic according to claim 2, wherein the step 2 is characterized in that a singular value decomposition method is used for solving a projection matrix, and the solving process is as follows:
1) Normalizing the two features to obtain standard data with a mean value of 0 and a variance of 1;
2) Calculating the variance S of X XX Variance S of Y YY Covariance S of X and Y XY ；
3) Calculating a matrix
4) Singular value decomposition is carried out on the matrix M to obtain a maximum singular value sigma, and left and right singular vectors u and v corresponding to the maximum singular value;
5) Calculating mapping matrices alpha and beta for X and Y,
6) The representation of the two features in the correlation subspace is X' =α T X，Y'＝β T Y。
4. The pedestrian re-recognition method based on typical correlation analysis fusion features according to claim 3, wherein the specific process of solving the projection matrix by using the singular value decomposition method in the step 2 is as follows:
(1) Let the mapping matrices of X and Y be α and β, respectively, their representation in subspace be: x' =α T X and Y' =β T Y, their correlation coefficients can be expressed as:
the objective function is:
namely solving mapping matrixes alpha and beta corresponding to the maximum correlation coefficient;
(2) Before projection, the original data is normalized first to obtain data with mean value of 0 and variance of 1,
Cov(α T X,β T Y)＝E(＜α T X,β T Y＞)
＝E((α T X)(β T Y) T )＝α T E(XY T )β
similarly, var (. Beta.) T Y)＝β T E(YY T )β，μ x Is the mean value of X;
(3) Since the average value of X and Y is 0, then
Var(X)＝Cov(X,X)＝E(XX T )
Var(Y)＝Cov(Y,Y)＝E(YY T )
Cov(X,Y)＝E(XY T )
Cov(Y,X)＝E(YX T )；
(4) Let S XX ＝Var(X,X)，S YY ＝Var(Y,Y)，S XY =cov (X, Y), then the objective function is converted into
(5) Since the numerator denominator is increased by the same multiple, the optimization target result is unchanged, the denominator is fixed, and the numerator is optimized, namely:
s.t.α T S XX α＝1,β T S YY β＝1；
(6) When solving the objective function in (5), adopting a singular value decomposition method, wherein u and v are two unit vectors,
order the
Alpha is then T S XX α＝1，β T S YY β＝1，
At the same time, from alpha T S XX α=1, obtainable:
from beta T S YY Beta=1, obtainable:
at this time, the objective function is:
s.t.u T u＝1,v T v＝1；
(7) For the objective function in (6), let the matrixAt this time, U and V represent left and right singular vectors corresponding to a singular value of the matrix M, and m=u Σv is obtained by singular value decomposition T Wherein U, V are respectively the matrix composed of the left singular vector and the right singular vector of M, and Sigma is the diagonal matrix composed of the singular values of M; since all columns of U, V are orthonormal, then U T U and V T v gets a vector with only one scalar 1 and the remaining scalar 0; at this time, the liquid crystal display device,
maximization ofThe corresponding maximum value is the maximum value of the singular values corresponding to a certain group of left and right singular vectors, namely after M is subjected to singular value decomposition, the maximum singular value is the maximum value of an optimization target, namely the maximum correlation coefficient between X and Y;
(8) The mapping matrix of original X and Y is obtained by using the corresponding left and right singular vectors u, v
5. The pedestrian re-recognition method based on the fusion feature of the canonical correlation analysis of claim 2, wherein the training of the pedestrian re-recognition model in step 3.1 uses the XQDA algorithm, and the training set and the training sample label are input and output as the subspace mapping matrix W and the subspace mapping matrix WWherein Sigma 'is' I Is an intra-class covariance matrix, sigma' E Is an inter-class covariance matrix;
during testing, the mahalanobis distance is used for measuring the similarity between two pedestrian images, and the mapping of M and the training features on the subspace W is input to obtain the mahalanobis distance of the original features on the subspace.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911114451.7A CN110874576B (en) | 2019-11-14 | 2019-11-14 | Pedestrian re-identification method based on typical correlation analysis fusion characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911114451.7A CN110874576B (en) | 2019-11-14 | 2019-11-14 | Pedestrian re-identification method based on typical correlation analysis fusion characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110874576A CN110874576A (en) | 2020-03-10 |
| CN110874576B true CN110874576B (en) | 2023-10-27 |
Family
ID=69718334
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911114451.7A Active CN110874576B (en) | 2019-11-14 | 2019-11-14 | Pedestrian re-identification method based on typical correlation analysis fusion characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110874576B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112270228A (en) * | 2020-10-16 | 2021-01-26 | 西安工程大学 | Pedestrian re-identification method based on DCCA fusion characteristics |
| CN114139614B (en) * | 2021-11-18 | 2022-09-23 | 南京工业大学 | Fisher photovoltaic module hot spot diagnosis method and system based on typical correlation analysis feature extraction |
| CN115984193A (en) * | 2022-12-15 | 2023-04-18 | 东北林业大学 | PDL1 expression level detection method fusing histopathology image and CT image |
| CN117978240A (en) * | 2024-01-29 | 2024-05-03 | 中国人民解放军军事科学院系统工程研究院 | Material state monitoring method and device based on high-flux satellite Internet of things |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273825B (en) * | 2017-05-25 | 2020-09-08 | 西安电子科技大学 | Physiological signal fusion identity recognition method based on improved canonical correlation analysis |
| CN107506700B (en) * | 2017-08-07 | 2021-03-16 | 苏州经贸职业技术学院 | Pedestrian re-identification method based on generalized similarity measurement learning |
| CN110414294B (en) * | 2018-04-26 | 2023-09-26 | 北京京东尚科信息技术有限公司 | Pedestrian re-identification method and device |
-
2019
- 2019-11-14 CN CN201911114451.7A patent/CN110874576B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110874576A (en) | 2020-03-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110874576B (en) | Pedestrian re-identification method based on typical correlation analysis fusion characteristics | |
| CN106250870B (en) | A kind of pedestrian's recognition methods again of joint part and global similarity measurement study | |
| CN106295124B (en) | The method of a variety of image detecting technique comprehensive analysis gene subgraph likelihood probability amounts | |
| CN105005772B (en) | A kind of video scene detection method | |
| CN103136516B (en) | The face identification method that visible ray and Near Infrared Information merge and system | |
| US9070041B2 (en) | Image processing apparatus and image processing method with calculation of variance for composited partial features | |
| CN110659665B (en) | Model construction method of different-dimension characteristics and image recognition method and device | |
| CN104504366A (en) | System and method for smiling face recognition based on optical flow features | |
| CN102422324B (en) | Age estimation device and method | |
| CN106778517A (en) | A kind of monitor video sequence image vehicle knows method for distinguishing again | |
| CN108960142B (en) | Pedestrian re-identification method based on global feature loss function | |
| CN106980848A (en) | Facial expression recognizing method based on warp wavelet and sparse study | |
| CN103034838A (en) | Special vehicle instrument type identification and calibration method based on image characteristics | |
| CN112836671B (en) | Data dimension reduction method based on maximized ratio and linear discriminant analysis | |
| CN104281835B (en) | Face recognition method based on local sensitive kernel sparse representation | |
| CN109284668B (en) | Pedestrian re-identification method based on distance regularization projection and dictionary learning | |
| CN106557740A (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| Li et al. | Human sperm health diagnosis with principal component analysis and K-nearest neighbor algorithm | |
| CN104598888A (en) | Human face gender recognition method | |
| CN109766748B (en) | Pedestrian re-recognition method based on projection transformation and dictionary learning | |
| CN110781828A (en) | Fatigue state detection method based on micro-expression | |
| CN108875448B (en) | Pedestrian re-identification method and device | |
| CN116612335A (en) | Few-sample fine-granularity image classification method based on contrast learning | |
| CN103714340A (en) | Self-adaptation feature extracting method based on image partitioning | |
| CN116468935A (en) | Multi-core convolutional network-based stepwise classification and identification method for traffic signs |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20230918 Address after: 150000 No. 113-3, Zhongshan Road, Nangang District, Harbin City, Heilongjiang Province Applicant after: PENGBOPUHUA TECHNOLOGY Co.,Ltd. Address before: 518000 1002, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Applicant before: Shenzhen Wanzhida Technology Co.,Ltd. Effective date of registration: 20230918 Address after: 518000 1002, Building A, Zhiyun Industrial Park, No. 13, Huaxing Road, Henglang Community, Longhua District, Shenzhen, Guangdong Province Applicant after: Shenzhen Wanzhida Technology Co.,Ltd. Address before: 710048 Shaanxi province Xi'an Beilin District Jinhua Road No. 19 Applicant before: XI'AN POLYTECHNIC University |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |