CN107423697B - Behavior identification method based on nonlinear fusion depth 3D convolution descriptor - Google Patents
Behavior identification method based on nonlinear fusion depth 3D convolution descriptor Download PDFInfo
- Publication number
- CN107423697B CN107423697B CN201710568540.3A CN201710568540A CN107423697B CN 107423697 B CN107423697 B CN 107423697B CN 201710568540 A CN201710568540 A CN 201710568540A CN 107423697 B CN107423697 B CN 107423697B
- Authority
- CN
- China
- Prior art keywords
- sample
- kernel
- feature set
- matrix
- global
- 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
Images
Classifications
-
- 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
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a behavior identification method based on a nonlinear fusion depth 3D convolution descriptor, and mainly solves the problem of low identification accuracy rate in the prior art. The scheme is as follows: 1. inputting each sample into a C3D network to obtain each layer of activation values; 2. processing each layer of the C3D network to obtain a feature vector of each layer; 3. fusing the feature vectors of different layers to obtain a global feature set and a local feature set; 4. carrying out discriminant nonlinear fusion on the global feature set and the local feature set to obtain a depth 3D convolution descriptor; 5. acquiring depth features of training samples for training a linear SVM classifier; 6. and acquiring the depth features of the test samples, and inputting the depth features into a linear SVM classifier for recognition. The invention improves the accuracy of behavior recognition, obtains 94.67% recognition rate on the UCF-Sports library, and can be applied to human-computer interaction, video monitoring and video retrieval.
Description
Technical Field
The invention belongs to the technical field of video processing, and particularly relates to a behavior identification method which can be applied to man-machine interaction, video monitoring and video retrieval.
Background
At present, behavior recognition methods in the field of video processing mainly include two methods, namely artificial features and deep learning. In which artificial features are usually designed based on domain knowledge of the controlled environment, however, video data in real scenes cannot always be correctly modeled, and thus the generalization capability of artificial features is not sufficient. Because the video contains abundant semantic information, the traditional artificial features are directly used for behavior recognition, certain semantic information and enough discrimination capability are lacked, and behavior recognition confusion is easily caused.
In recent years, behavior recognition methods based on deep learning have enjoyed great success and progress. Deep learning generally utilizes a deep convolutional neural network for behavior recognition, and the deep convolutional neural network for behavior recognition mainly includes: 2D convolutional networks, 3D convolutional networks, and C3D networks. Among them, the 3D convolutional network model is superior to the conventional 2D convolutional network model. However, the 3D convolutional network model requires a human body detector and a head tracking algorithm to segment the video, and the segmented video segment is used as an input of the 3D convolutional neural network, which has great limitations. Compared with a 3D convolutional network, the C3D network can learn the space-time information in the video, and can directly take the complete video as input, does not depend on any preprocessing, and is easy to expand to a large-scale data set. However, when performing behavior recognition, the C3D network only uses the global features at the top level, and the bottom-level features, which are important local features in the network, are not fully regarded.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a behavior identification method based on a nonlinear fusion depth 3D convolution descriptor, which obtains more discriminant feature representation by fusing different layer features of a C3D network and improves the behavior identification rate.
The technical key point for realizing the invention is to construct a discriminant nonlinear fusion method, fuse the global features and the local features extracted from the C3D network by using the method to obtain a deep 3D convolution descriptor, and classify data by using an SVM, wherein the implementation steps comprise the following steps:
(1) acquiring L eigenvectors u of each sample by using a C3D network, wherein L is the layer number of the C3D network;
(2) obtaining a global feature vector X and a local feature vector Y of each sample according to the feature vector u to obtain a global feature set X and a local feature set Y;
(3) obtaining a depth 3D convolution descriptor D according to the global feature set X and the local feature set YC3D;
(4) Descriptor D by depth 3D convolutionC3DObtaining a depth feature vector z of each training sampletrainAnd a depth feature vector z for each test sampletest;
(5)Depth feature vector z from training samplestrainTraining a linear SVM classifier;
(6) depth feature vector z for each test sample according to a linear SVM classifiertestAnd classifying to obtain a classification result of each test sample.
Compared with the prior art, the invention has the following advantages:
according to the method, the global features and the local features of the data are extracted by using the C3D network, a more discriminant deep 3D convolution descriptor is obtained through nonlinear fusion, the deep 3D convolution descriptor is used for training the SVM classifier, and the accuracy of behavior recognition is improved.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention.
Detailed Description
Referring to fig. 1, the behavior recognition method based on the nonlinear fusion depth 3D convolution descriptor of the present invention includes the following steps:
step 1, a training data set and a test data set are obtained.
(1a) Acquiring a human behavior video set V, wherein the category number of the human behavior video set V is C, and the total number of samples is N;
(1b) selecting a samples from each category of the human behavior video set V as test samples to obtain a test data set VtrainTaking the residual samples in the human behavior video set V as training samples to obtain a training data set VtrainWherein, a ∈ {1,2k-1},NkThe number of samples in class k, k is 1, 2.
And 2, acquiring a feature vector u of each sample.
(2a) Dividing each sample into a plurality of continuous video segments, wherein the length of each video segment is the same;
(2b) inputting the video clips obtained in the step (2a) into a C3D network, and obtaining activation values of each layer of each video clip in the C3D network, wherein the number of the layers of the C3D network is L;
(2c) according to the activation values of all layers of each video clip obtained in the step (2b), summing the activation values of the same layer of all the video clips, and averaging to obtain an average activation value of each layer;
(2d) and (3) carrying out dimensionality reduction on the average activation value of each layer obtained in the step (2c) by using principal component analysis to obtain L eigenvectors u of each sample.
And 3, acquiring a global feature set X.
(3a) According to the L feature vectors u of the training samples obtained in the step 2, the number from the C3D network is the firstB feature vectors u are selected from the L-th layer, different feature vectors u are connected in series to obtain a global feature vector x, the dimension of the global feature vector x is q, represents rounding down;
(3b) repeating the same process for each sample according to the step (3a) to obtain a global feature set: x ═ X1,x2,...,xn,...,xN]Wherein, X ∈ Rq×N,Rq×NVector space of dimension q × N, xnIs the global feature vector of the nth sample, N is 1, 2.
And 4, acquiring a local feature set Y.
(4a) According to the L eigenvectors u of the training sample obtained in the step 2, from the 1 st layer to the 1 st layer of the C3D networkE feature vectors u are selected from the layer, different feature vectors u are connected in series to obtain a local feature vector y, the dimension of the local feature vector y is p,
(4b) according to the step (4a) pairRepeating the same process for each sample to obtain a local feature set: y ═ Y1,y2,...,ym,...,yN]Wherein, Y ∈ Rp×N,Rp×NVector space of dimension p × N, ymIs the local feature vector of the mth sample, m is 1, 2.
Step 5, calculating a depth 3D convolution descriptor DC3D。
(5a) Computing a kernel matrix K of a global feature set X using a kernel functionXAnd a kernel matrix K of the local feature set YYThe kernel function may be a polynomial kernel function, a gaussian kernel function, a laplacian kernel function, a power exponent kernel function, or other different types of kernel functions, and the polynomial kernel function is selected in this example, but is not limited to this method;
(5a1) calculating a kernel matrix K of the global feature set X by utilizing a polynomial kernel function according to the global feature set XXEach of the elements of (a):
(KX)ij=GX(xi,xj),
wherein, i is 1,2, N, j is 1,2, N, (K)X)ijKernel matrix K being a global feature set XXThe ith row and the jth column of elements,<·>representing the calculated inner product, xiGlobal feature vector for ith sample in global feature set X, XjIs the global feature vector of the jth sample in the global feature set X, theta1Kernel parameters are polynomial kernel functions;
(5a2) according to the local feature set Y, a kernel matrix K of the local feature set Y is calculated by utilizing a polynomial kernel functionYEach of the elements of (a):
(KY)ηξ=GY(yη,yξ),
wherein η is 1,2, N, ξ is 1,2, N, (K)Y)ηξKernel matrix K being a local feature set YYLine η column ξ,yηis the local feature vector of the η th sample in the local feature set Y, YξIs the local feature vector of the ξ th sample in the local feature set Y, theta2Kernel parameters are polynomial kernel functions;
(5b) kernel matrix K from global feature set XXAnd a kernel matrix K of the local feature set YYPerforming discriminant nonlinear fusion to obtain a depth 3D convolution descriptor DC3D:
(5b1) Calculating an intra-kernel-class divergence matrix for a global feature set XAnd the inter-kernel-class divergence matrix of the global feature set X
Wherein the content of the first and second substances, for the non-linear mapping of the global feature vector for the u-th sample in the kth class of samples,global feature vector of u-th sample in k-th sample, u ═ 1,2kT is matrix transposition;
(5b2) computing a set of local featuresNuclear-class internal divergence matrix of YAnd the inter-kernel-class divergence matrix of the local feature set Y
Wherein the content of the first and second substances, for the non-linear mapping of the local feature vector of the g-th sample in the kth class of samples,local feature vector of the g-th sample in the k-th sample, g ═ 1,2k;
(5b3) According to the kernel class divergence matrix of the global feature set X obtained in the step (5b1)And the kernel-class divergence matrix of the local feature set Y obtained in the step (5b2)Obtaining a cross covariance matrix KxyAnd cross covariance matrix Kyx:
Wherein cov (·) represents the computational covariance;
(5b4) constructing an objective function:and calculates a global projection vector α for each eigenvector x and a local projection vector β for each eigenvector y using the objective function, wherein,
(5b5) and (3) solving the obtained objective function according to a Lagrange multiplier method (5b1), namely converting the problem of solving the objective function into the problem of solving the generalized characteristic value, wherein the formula for solving the generalized characteristic value is as follows:
wherein λ is a generalized eigenvalue, the global projection vector α is composed of the first N elements of the eigenvector corresponding to the generalized eigenvalue λ, and the local projection vector β is composed of the last N elements of the eigenvector corresponding to the generalized eigenvalue λ;
(5b6) solving the generalized eigenvalues according to the step (5b5) to obtain the first s maximum eigenvalues to obtain a projection matrix W of the global eigenvalue set XX=[α1,α2,...,αs]And a projection matrix W of the local feature set YY=[β1,β2,...,βs]Where s ═ min (q, p), min (·) denotes the minimum value, α1,α2,...,αsThe global projection vectors corresponding to the first s maximum eigenvalues obtained for solving the generalized eigenvalues, β1,β2,...,βsLocal projection vectors corresponding to the first s maximum eigenvalues obtained by solving the generalized eigenvalues;
(5b7) from a global feature setKernel matrix K of XXKernel matrix K of local feature set YYProjection matrix W of global feature set XXAnd a projection matrix W of the local feature set YYObtaining a depth 3D convolution descriptor:
and 6, training a linear SVM classifier.
(6a) Obtaining a depth 3D convolution descriptor D according to the step 5C3DFrom which the depth feature vector z of each training sample is obtainedtrainWherein the depth 3D convolution descriptor DC3DEach column of (a) corresponds to a depth feature vector of a sample;
(6b) depth feature vector z using training samplestrainAnd training the linear SVM classifier.
And 7, obtaining the classification result of the test sample.
(7a) Obtaining a depth 3D convolution descriptor D according to the step 5C3DFrom which depth feature vector z for each test sample is obtainedtest;
(7b) The depth feature vector z of each test sampletestAnd inputting the data into a linear SVM classifier to obtain the recognition result of each test sample.
The foregoing description is only an example of the present invention and should not be construed as limiting the invention, as it will be apparent to those skilled in the art that various modifications and variations in form and detail can be made without departing from the principle and structure of the invention after understanding the present disclosure and the principles, but such modifications and variations are considered to be within the scope of the appended claims.
Claims (1)
1. The behavior identification method based on the nonlinear fusion depth 3D convolution descriptor comprises the following steps:
(1) acquiring L eigenvectors v of each sample by using a C3D network, wherein L is the layer number of the C3D network;
(1a) dividing each sample into a plurality of continuous video segments, wherein the length of each video segment is the same;
(1b) inputting the video clips obtained in the step (1a) into a C3D network, and obtaining activation values of each layer of each video clip in the C3D network, wherein the number of the layers of the C3D network is L;
(1c) according to the activation values of all layers of each video clip obtained in the step (1b), summing the activation values of the same layer of all the video clips, and averaging to obtain an average activation value of each layer;
(1d) performing dimensionality reduction on the average value of the activation value of each layer obtained in the step (1c) by using principal component analysis to obtain L eigenvectors u of each sample;
(2) obtaining a global feature vector X and a local feature vector Y of each sample according to the feature vector v to obtain a global feature set X and a local feature set Y; the method comprises the following implementation steps:
(2a) according to the L feature vectors v of the training samples obtained in the step (1), obtaining the L feature vectors v from the C3D networkB feature vectors v are selected from the L-th layer, different feature vectors v are connected in series to obtain a global feature vector x, the dimension of the global feature vector x is q, represents rounding down;
(2b) repeating the same process for each sample according to the step (2a) to obtain a global feature set: x ═ X1,x2,...,xn,...,xN]Wherein, X ∈ Rq×N,Rq×NVector space of dimension q × N, xnA global feature vector of an nth sample, N being 1, 2.
(2c) According to the L eigenvectors v of the training sample obtained in the step (1), from the layer 1 to the layer 1 of the C3D networkE feature vectors v are selected from the layer, different feature vectors v are connected in series to obtain a local feature vector y, the dimension of the local feature vector y is p, represents rounding down;
(2d) repeating the same process for each sample according to the step (2c) to obtain a local feature set: y ═ Y1,y2,...,ym,...,yN]Wherein, Y ∈ Rp×N,Rp×NVector space of dimension p × N, ymIs the local feature vector of the mth sample, m ═ 1, 2.., N;
(3) obtaining a depth 3D convolution descriptor D according to the global feature set X and the local feature set YC3D(ii) a The method comprises the following concrete steps:
(3a) computing a kernel matrix K of a global feature set X using a kernel functionXAnd a kernel matrix K of the local feature set YY:
(3a1) Calculating a kernel matrix K of the global feature set X by utilizing a polynomial kernel function according to the global feature set XXEach of the elements of (a):
(KX)ij=GX(xi,xj),
wherein, i is 1,2, N, j is 1,2, N, (K)X)ijIs a kernel matrix KXThe ith row and the jth column of elements,<·>representing the calculated inner product, xiGlobal feature vector for the ith sample in feature set X, XjGlobal feature vector, θ, for the jth sample in feature set X1Kernel parameters are polynomial kernel functions;
(3a2) according to local characteristicsA characteristic set Y, a kernel matrix K of the local characteristic set Y is calculated by utilizing a polynomial kernel functionYEach of the elements of (a):
(KY)ηξ=GY(yη,yξ),
wherein η is 1,2, N, ξ is 1,2, N, (K)Y)ηξIs a kernel matrix KYLine η column ξ,<·>representing the calculated inner product, yηIs the local feature vector of the η th sample in the local feature set Y, YξIs the local feature vector of the ξ th sample in the local feature set Y, theta2Kernel parameters are polynomial kernel functions;
(3b) obtaining a kernel matrix K of the global feature set X according to the step (3a)XAnd a kernel matrix K of the local feature set YYCalculating the depth 3D convolution descriptor DC3D:
(3b1) Calculating an intra-kernel-class divergence matrix for a global feature set XAnd the inter-kernel divergence matrix
Wherein the content of the first and second substances, for the non-linear mapping of the global feature vector for the u-th sample in the kth class of samples,global feature vector of the u-th sample in the k-th sample, NkN is the number of samples of class k, u is 1,2kK is 1,2, C, T is a matrix transpose;
(3b2) calculating the intra-kernel-class divergence matrix of the local feature set YAnd the inter-kernel divergence matrix
Wherein the content of the first and second substances, for the non-linear mapping of the local feature vector of the g-th sample in the kth class of samples,local feature vector of the g-th sample in the k-th sample, g ═ 1,2k;
(3b3) Kernel-to-kernel divergence matrix from global feature set XAnd the inter-kernel-class divergence matrix of the local feature set YObtaining a cross covariance matrix KxyAnd cross covariance matrix Kyx:
Wherein cov (·) represents the computational covariance;
(3b4) constructing an objective function:and calculates a projection vector α for each eigenvector x and a projection vector β for each eigenvector y using the objective function, wherein,
(3b5) solving the objective function obtained by (3b4) according to a Lagrange multiplier method, namely converting the problem of solving the objective function into the problem of solving the generalized characteristic value, wherein the formula for solving the generalized characteristic value is as follows:
wherein λ is a generalized eigenvalue, the global projection vector α is composed of the first N elements of the eigenvector corresponding to the generalized eigenvalue λ, and the local projection vector β is composed of the last N elements of the eigenvector corresponding to the generalized eigenvalue λ;
(3b6) solving the generalized eigenvalues according to the step (3b5) to obtain the first s maximum eigenvalues to obtain a projection matrix W of the global eigenvalue set XX=[α1,α2,...,αs]And a projection matrix W of the local feature set YY=[β1,β2,...,βs]Where s ═ min (q, p), min (·) denotes the minimum value, α1,α2,...,αsThe global projection vectors corresponding to the first s maximum eigenvalues obtained for solving the generalized eigenvalues, β1,β2,...,βsLocal projection vectors corresponding to the first s maximum eigenvalues obtained by solving the generalized eigenvalues;
(3b7) kernel matrix K from global feature set XXKernel matrix K of local feature set YYProjection matrix W of global feature set XXAnd a projection matrix W of the local feature set YYObtaining a depth 3D convolution descriptor:
(4) descriptor D by depth 3D convolutionC3DObtaining a depth feature vector z of each training sampletrainAnd a depth feature vector z for each test sampletest;
(5) Depth feature vector z from training samplestrainTraining a linear SVM classifier;
(6) depth feature vector z for each test sample according to a linear SVM classifiertestAnd classifying to obtain the identification result of each test sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710568540.3A CN107423697B (en) | 2017-07-13 | 2017-07-13 | Behavior identification method based on nonlinear fusion depth 3D convolution descriptor |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710568540.3A CN107423697B (en) | 2017-07-13 | 2017-07-13 | Behavior identification method based on nonlinear fusion depth 3D convolution descriptor |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107423697A CN107423697A (en) | 2017-12-01 |
CN107423697B true CN107423697B (en) | 2020-09-08 |
Family
ID=60427184
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710568540.3A Active CN107423697B (en) | 2017-07-13 | 2017-07-13 | Behavior identification method based on nonlinear fusion depth 3D convolution descriptor |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107423697B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108298393A (en) * | 2017-12-20 | 2018-07-20 | 浙江新再灵科技股份有限公司 | Method based on the wrong report of depth network filtering elevator malfunction |
CN108828533B (en) * | 2018-04-26 | 2021-12-31 | 电子科技大学 | Method for extracting similar structure-preserving nonlinear projection features of similar samples |
CN109104728B (en) * | 2018-07-11 | 2021-09-14 | 浙江理工大学 | ELM classification intrusion detection method based on improved LDA dimension reduction |
CN110610145B (en) * | 2019-08-28 | 2022-11-08 | 电子科技大学 | Behavior identification method combined with global motion parameters |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102542067A (en) * | 2012-01-06 | 2012-07-04 | 上海交通大学 | Automatic image semantic annotation method based on scale learning and correlated label dissemination |
CN102930302A (en) * | 2012-10-18 | 2013-02-13 | 山东大学 | On-line sequential extreme learning machine-based incremental human behavior recognition method |
CN103699578A (en) * | 2013-12-01 | 2014-04-02 | 北京航空航天大学 | Image retrieval method based on spectrum analysis |
WO2015008279A1 (en) * | 2013-07-15 | 2015-01-22 | Tel Hashomer Medical Research Infrastructure And Services Ltd. | Mri image fusion methods and uses thereof |
CN104715254A (en) * | 2015-03-17 | 2015-06-17 | 东南大学 | Ordinary object recognizing method based on 2D and 3D SIFT feature fusion |
CN105912991A (en) * | 2016-04-05 | 2016-08-31 | 湖南大学 | Behavior identification method based on 3D point cloud and key bone nodes |
-
2017
- 2017-07-13 CN CN201710568540.3A patent/CN107423697B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102542067A (en) * | 2012-01-06 | 2012-07-04 | 上海交通大学 | Automatic image semantic annotation method based on scale learning and correlated label dissemination |
CN102930302A (en) * | 2012-10-18 | 2013-02-13 | 山东大学 | On-line sequential extreme learning machine-based incremental human behavior recognition method |
WO2015008279A1 (en) * | 2013-07-15 | 2015-01-22 | Tel Hashomer Medical Research Infrastructure And Services Ltd. | Mri image fusion methods and uses thereof |
CN103699578A (en) * | 2013-12-01 | 2014-04-02 | 北京航空航天大学 | Image retrieval method based on spectrum analysis |
CN104715254A (en) * | 2015-03-17 | 2015-06-17 | 东南大学 | Ordinary object recognizing method based on 2D and 3D SIFT feature fusion |
CN105912991A (en) * | 2016-04-05 | 2016-08-31 | 湖南大学 | Behavior identification method based on 3D point cloud and key bone nodes |
Also Published As
Publication number | Publication date |
---|---|
CN107423697A (en) | 2017-12-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Shapenet: A shapelet-neural network approach for multivariate time series classification | |
CN107423697B (en) | Behavior identification method based on nonlinear fusion depth 3D convolution descriptor | |
Kozerawski et al. | Clear: Cumulative learning for one-shot one-class image recognition | |
JP2015052832A (en) | Weight setting device and method | |
CN113326731A (en) | Cross-domain pedestrian re-identification algorithm based on momentum network guidance | |
CN108537137B (en) | Multi-modal biological characteristic fusion recognition method based on label identification correlation analysis | |
Reza et al. | ICA and PCA integrated feature extraction for classification | |
CN105023006B (en) | Face identification method based on enhanced nonparametric maximal margin criterion | |
CN108154156B (en) | Image set classification method and device based on neural topic model | |
CN104268507A (en) | Manual alphabet identification method based on RGB-D image | |
Chergui et al. | Kinship verification through facial images using cnn-based features | |
Najar et al. | A new hybrid discriminative/generative model using the full-covariance multivariate generalized Gaussian mixture models | |
CN109190471B (en) | Attention model method for video monitoring pedestrian search based on natural language description | |
Tamura et al. | Time series classification using macd-histogram-based sax and its performance evaluation | |
Hashida et al. | Multi-channel mhlf: Lstm-fcn using macd-histogram with multi-channel input for time series classification | |
CN114329031A (en) | Fine-grained bird image retrieval method based on graph neural network and deep hash | |
CN113222002A (en) | Zero sample classification method based on generative discriminative contrast optimization | |
Novakovic et al. | Classification accuracy of neural networks with pca in emotion recognition | |
Li et al. | Multiple instance discriminative dictionary learning for action recognition | |
CN113887509A (en) | Rapid multi-modal video face recognition method based on image set | |
CN107451537B (en) | Face recognition method based on deep learning multi-layer non-negative matrix decomposition | |
CN112465054A (en) | Multivariate time series data classification method based on FCN | |
CN112241922A (en) | Power grid asset comprehensive value evaluation method based on improved naive Bayes classification | |
Sahoo et al. | Vision-based static hand gesture recognition using dense-block features and svm classifier | |
CN111507243A (en) | Human behavior recognition method based on Grassmann manifold analysis |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |