CN107085716B - Cross-view gait recognition method based on multi-task generation countermeasure network - Google Patents
Cross-view gait recognition method based on multi-task generation countermeasure network Download PDFInfo
- Publication number
- CN107085716B CN107085716B CN201710373017.5A CN201710373017A CN107085716B CN 107085716 B CN107085716 B CN 107085716B CN 201710373017 A CN201710373017 A CN 201710373017A CN 107085716 B CN107085716 B CN 107085716B
- Authority
- CN
- China
- Prior art keywords
- gait
- template
- view
- loss
- encoder
- 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
- G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Abstract
The invention belongs to the field of computer vision and machine learning, and particularly relates to a cross-perspective gait recognition method for generating a confrontation network based on multiple tasks. The invention mainly solves the problem that the generalization performance of the model is reduced under the large visual angle change of gait recognition. For an original pedestrian video frame sequence, preprocessing each frame image, and extracting gait template features; then coding the gait hidden representation through a neural network, and carrying out angle transformation in a hidden space; then, generating a confrontation network through multiple tasks to reconstruct gait template characteristics of other visual angles; and finally, recognizing by using the gait implicit representation. Compared with classification-based or other reconstruction-based methods, the method has stronger interpretability and can improve the identification performance.
Description
Technical Field
The invention belongs to the technical field of computer vision and machine learning, and particularly relates to a cross-visual angle gait recognition method based on videos.
Background
The video-based cross-perspective gait recognition problem is one of the problems of computer vision and machine learning field research. When gait video frame sequences under different visual angles are given, whether main bodies of the gait frame sequences are the same object or not is judged according to computer vision or a machine learning algorithm. Currently, there are many predecessors working in this field, and the main methods can be divided into three categories: reconstruction-based methods, subspace-based methods, and deep learning-based methods. The following are some references to these three types of processes:
[1]W.Kusakunniran,Q.Wu,J.Zhang,and H.Li,“Support vector regression for multi-view gait recognition based on local motion feature selection,”in Conference on Computer Vision and Pattern Recognition,pp.974–981,2010.
[2]M.Hu,Y.Wang,Z.Zhang,J.J.Little,and D.Huang,“View-invariant discriminative projection for multi-view gait-based human identification,”IEEE Transactions on Information Forensics and Security,vol.8,no.12,pp.2034–2045,2013.
[3]W.Kusakunniran,Q.Wu,J.Zhang,H.Li,and L.Wang,“Recognizing gaits across views through correlated motion co-clustering,”IEEE Transactions on Image Processing,vol.23,no.2,pp.696–709,2014.
[4]S.Yu,H.Chen,Q.Wang,L.Shen,and Y.Huang,“Invariant feature extraction for gait recognition using only one uniform model,”Neurocomputing,vol.239,pp.81–93,2017.
[5]Z.Wu,Y.Huang,L.Wang,X.Wang,and T.Tan,“A comprehensive study on cross-view gait based human identification with deep CNNs,”IEEE Transactions on Pattern Analysis and Machine Intelligence,vol.39,no.2,pp.209–226,2017.
[6]Y.Feng,Y.Li,and J.Luo,“Learning effective gait features using LSTM,”in International Conference on Pattern Recognition,pp.320–325,2016.
[7]H.Iwama,M.Okumura,Y.Makihara,and Y.Yagi,“The OU-ISIR gait database comprising the large population dataset and performance evaluation of gait recognition,”IEEE Transactions on Information Forensics and Security,vol.7,no.5,pp.1511–1521,2012.
[8]Y.Makihara,R.Sagawa,Y.Mukaigawa,T.Echigo,and Y.Yagi,“Gait recognition using a view transformation model in the frequency domain,”in European Conference on Computer Vision,pp.151–163,Springer,2006.
[9]W.Kusakunniran,Q.Wu,H.Li,and J.Zhang,“Multiple views gait recognition using view transformation model based on optimized gait energy image,”in International Conference on Computer Vision,pp.1058–1064,2009.
[10]X.Xing,K.Wang,T.Yan,and Z.Lv,“Complete canonical correlation analysis with application to multi-view gait recognition,”Pattern Recognition,vol.50,pp.107–117,2016.。
the first reconstruction-based method is mainly based on a reconstructed gait template, such as a VTM model in [1,3,8,9], and predicts the value of each pixel point in template features under a target view angle respectively by training a plurality of models. To reduce computational overhead, each pixel of the template features is reconstructed simultaneously based on the method of the auto-encoder [4] and a view-invariant feature representation is extracted.
The subspace-based approach [2,10] projects gait features from different perspectives into a common subspace and computes the similarity in the common subspace. However, the subspace method can only model the linear correlation among the features generally, neglects the nonlinear correlation, and has low recognition rate.
Recently, models based on deep neural networks have gained greater application in the field of computer vision. [5,6] the deep network is used for gait recognition, and the feature expression with unchanged visual angle is automatically extracted. Although the recognition accuracy is greatly improved compared with the past method, the depth model has no interpretability.
Disclosure of Invention
The invention aims to provide a video-based cross-view gait recognition method with high recognition rate and low calculation overhead.
The basic contents of the present invention will be described first.
One, periodic energy diagram
(a) In order to establish more effective gait template characteristics, the original pedestrian video sequence needs to be preprocessed. For each frame in a video sequence, firstly separating a front background, extracting a gait outline image, and then centering and aligning the gait outline image to the center of the image to obtain a standardized gait outline image;
(b) based on the normalized gait profile sequence, a normalized cycle position of each frame is first obtained using a cycle detection technique in gait recognition. Defining a gait cycleIs the time domain between two adjacent local minimum normalized period positions. The sequence of contour maps is then divided into one or more gait cycles and frames that do not constitute a complete gait cycle are discarded. Sequence B of gait profile according to centralizationtAnd corresponding normalized period position rtStructure of ncPeriodic energy map (PEI) template features for each channel. The value at the (x, y) coordinate of the kth channel in the periodic energy map PEI is calculated as follows:
wherein:
m represents the size of the time domain window covered by each channel, and t (k) determines the range of the time domain covered by each channel. In this way, the spatial information of the sequence of contour maps can be divided into n according to the periodic position of each framecAnd in a time domain, the time and space information of the gait is coded simultaneously.
Two, multitask generation countermeasure network
The multitask generation countermeasure network for gait recognition is mainly composed of an encoder, a view angle conversion layer, a generator and a discriminator as shown in figure 1. Wherein:
(a) an encoder E: to obtain a hidden representation of a particular view, a convolutional neural network is used as an encoder in the model. Input x of the encoderuThe PEI template with the visual angle u is provided by the invention. The input scale size is 64 × 64 × nc. Each channel of the PEI template is independently sent into an encoder, and then effective characteristic implicit expression z is extracted by using time pooling and passing through a full connection layeruAs the output of the encoder. For time pooling, the mode adopted is average pooling;
(b) view angle conversion layer V: assuming the distribution of gait data as a function of perspectiveOn a high-dimensional manifold. The movement of the samples in the manifold to a particular direction may enable a change in viewing angle while preserving identity information. Hidden representation z given a hidden spaceuThe view transformation can be described as:
wherein h isiRepresenting the transformation vector from view i-1 to view i. z is a radical ofvIs a hidden representation at view angle v;
(c) a generator G: by representing z implicitlyvAs an input, a gait template at the perspective v may be generated. The generator consists of five deconvolution layers and uses the ReLU as the excitation function. Because the direct reconstruction of the PEI template has certain difficulty, one channel of the PEI template is randomly selected for reconstruction, so that the information of all channels can be stored by hidden representation. Define the input of the generator as [ z ]v,c]And c is a one-hot encoded representation of the channel;
(d) a discriminator D: the structure of the discriminator is the same as that of the encoder except that time pooling operation is not adopted and the dimension of the output layer is different from that of the encoder. Defining the output dimension of the discriminator as nv+nc+ndWherein n isvIs the number of viewing angles, ncIs the number of PEI channels, ndIs the number of different identities in the training set. The output of the discriminator corresponds to nv+nc+ndAnd each sub-arbiter shares parameters except the last layer. Each arbiter is responsible for a different arbitration task to determine whether the generated samples belong to a particular distribution. For example, in the model, the first nv+ncThe sub-discriminators enable the generator to generate a distribution that satisfies a particular view and channel, then ndThe sub-discriminators enable the generator to generate samples corresponding to distributions belonging to a particular identity.
Three, loss function
The present invention uses two loss functions to train a multitask generative confrontation network.
(a) Pixel-by-pixel loss: to enhance the ability of the hidden representation to retain identity information, pixel-by-pixel loss between the generated template and the real template is first minimized. According to template characteristics xvAnd dummy features
The pixel-by-pixel loss is calculated as:
(b) multitask fighting loss: according to template characteristics xvAnd dummy features
Computing the multitask countermeasure loss:
where E represents the expectation for the corresponding sample set, | | · | | | ceiling1Representing the L1 norm, and vector s is the one-hot encoding of identity, channel, and angle information. D (-) represents the output of the discriminator, the dimension of the discriminator output is the same as the dimension of the vector s, and the non-zero element in the vector s determines the distribution to which the characteristics of the pseudo-template belong;
the final loss function is defined as:
L＝Lp+αLa (6)
alpha is used as a hyper-parameter to balance pixel-by-pixel loss and multi-task countermeasure loss, and after a final loss function is defined, parameters of an encoder, a view transformation layer, a generator and a discriminator are alternately updated by using a back propagation algorithm.
The invention provides a cross-view gait recognition method based on a multitask generation countermeasure network, which comprises the following specific steps:
(1) inputting pedestrian video frame sequences under different visual angles, and constructing gait template characteristics:
vector xiIs the gait template characteristic under the visual angle i, nvIs the number of all views;
(2) for any different view angles u, v, encoding the corresponding gait template features x by using a convolutional neural networkuTo the hidden space and obtaining a hidden representation in the hidden space as zu；
(3) Representing z implicitly in implicit spaceuCarrying out visual angle transformation to an angle v to obtain a hidden representation zv；
(4) Will implicitly represent zvAnd the channel one-hot code is used as input, and the pseudo-template characteristic under the output angle v is obtained after the generation network in the countermeasure network is generated through multiple tasks
(5) According to template characteristics xuAnd dummy features
Calculating the pixel-by-pixel loss Lp；
(6) Generating discriminative networks in countermeasure networks using multitasking, according to template features xuAnd dummy features
Computing a multitasking confrontation loss La；
(7) Weighting pixel-by-pixel loss and multi-task counterloss according to total loss L ═ Lp+αLaTraining the multitask generation countermeasure network, alpha is used as a hyper-parameter to balance pixel-by-pixel loss and multitask countermeasure loss.
In the invention, the gait template is constructed by the following steps:
(1) performing foreground and background separation on each frame of an original video frame sequence, and extracting a gait contour map; and translating and zooming the gait outline image to the image center to obtain a centralized gait outlineGraph sequence B1,B2,B3,…,Bn}；
(2) For each frame in the sequence of centered gait profiles, the normalized period position r is calculatedtA normalized cycle position of a t-th frame of the centered gait profile;
(3) sequence B of gait profile according to centralizationtAnd corresponding normalized period position rtStructure of ncThe periodic energy map template characteristics of each channel; the value at the (x, y) coordinate of the kth channel in the periodic energy map PEI is calculated as follows:
wherein:
m represents the size of the time domain window covered by each channel, and t (k) determines the range of the time domain covered by each channel.
In the invention, the visual angle transformation step is as follows:
implicit representation z at a given view angle uuThe calculation process for transforming to view v can be described as:
wherein h isiRepresenting the transformation vector from view i-1 to view i.
In the present invention, the loss L ispAnd LaThe calculation steps are as follows:
(1) according to template characteristics xvAnd dummy features
Calculate the pixel-by-pixel loss:
(2) according to template characteristics xvAnd dummy features
Computing the multitask countermeasure loss:
where E represents the expectation for the corresponding sample set, | | · | | | ceiling1Representing the L1 norm, and vector s is the one-hot encoding of identity, channel, and angle information. D (-) represents the output of the discriminator.
The method of the invention extracts the hidden representation with specific view angle by using the nonlinear modeling capability of the convolutional neural network. By carrying out view angle transformation in a hidden space, the calculation overhead is reduced; and based on the capability of generating the confrontation network to model the distribution, the characteristics with expression are extracted, and the identification efficiency can be greatly improved.
Drawings
FIG. 1: detailed model flow diagram of the invention.
FIG. 2: OU-ISIR, CASIA-B, USF data set sample presentation.
FIG. 3 average accuracy on CASIA-B data sets under different walking conditions.
Detailed Description
Having described the specific steps and models of the present invention, the following demonstrates the test effect of the present invention on several gait data sets.
The experiment used three data sets, including the OU-ISIR data set, the CASIA-B data set, and the USF data set. Fig. 2 shows some examples of these three data sets.
The OU-ISIR dataset has a total of 4007 different people, with 2135 males and 1872 females, with an age distribution ranging from 1 year to 94. The data has 4 different angles. Respectively 55 deg., 65 deg., 75 deg., 85 deg.. The method is used for extracting the PEI template, setting the number of channels to be 3, interpolating the PEI to the picture with the size of 64 multiplied by 64 pixels, and putting the picture into a multitask generation countermeasure network for training.
The CASIA-B dataset has a total of 124 different people, 11 different perspectives. Wherein, each person has 6 groups of gait sequences of normal walking, 2 groups of gait sequences of walking with the bag, and 2 groups of gait sequences of wearing the coat at each visual angle. CASIA-B has a larger viewing angle range than OU-ISIR, but the number of people is relatively small. The number of PEI channels is set to be 10, and gait recognition accuracy rates under different walking states are tested respectively.
The USF is another common gait data set, with a total of 122 different people, each with 5 different forms of gait sequences, close to the real scene. In the experiments, we only used gait sequences at different perspectives for testing. We set the number of channels for PEI to 10.
The experiment uses Rank-1 recognition accuracy as a performance index. The identification is performed by using a nearest neighbor classifier in a hidden space of a view transformation layer.
Experimental example 1: recognition performance of multitask generation countermeasure networks
This part of the experiment shows different models, identifying accuracy across viewing angles. As a comparison we have chosen an autoencoder, canonical correlation analysis, linear discriminant analysis, convolutional neural network, and local tensor discriminant model. Table 1 shows a comparison of the method of the present invention with other methods on three data sets. It can be seen that the present invention is greatly enhanced over other methods.
Experimental example 2: effect of different loss functions on model Performance
Table 2 shows the performance variation of the model over the CASIA-B dataset when different penalty functions are used. It can be seen that the recognition performance of the model can be improved by combining the multitask pair loss and the pixel-by-pixel loss; and the performance of the model is degraded when different loss functions are used alone.
Experimental example 3: effect of different Walking states on model Performance
Fig. 3 shows the cross-view recognition accuracy on the CASIA-B under different walking states, there are three different walking states in total: normal walking, walking with the bag, and walking with the coat. It can be seen that the accuracy is highest during normal walking, and the decrease in model performance is more pronounced with the coat-worn gait sequence than with the bag-worn gait sequence.
Experimental example 4: influence of different templates and PEI (polyetherimide) channel numbers on identification accuracy
Table 3 shows the effect of using different templates and the number of PEI channels on the recognition accuracy. We used a gait energy map (GEI) and a time gait template (CGI) and analyzed the average recognition accuracy for PEI channel numbers of 3, 5, and 10, respectively. It can be seen that the PEI templates of our invention have higher recognition accuracy than 6EI and C6I, and as the number of PEI templates increases, the recognition accuracy can be further improved.
Table 1: recognition accuracy (%) under different methods
Table 2: model identification accuracy (%) under different loss functions
| 54° | 90° | 126° | |
| Loss per pixel + multi-tasking penalty | 82.4 | 73.1 | 81.3 |
| Pixel by pixel loss | 81.5 | 71.7 | 83.5 |
| Multitasking countermeasure loss | 74.6 | 68.6 | 75.4 |
TABLE 3 influence of different templates and the number of PEI channels on the recognition accuracy
Claims (3)
1. The cross-perspective gait recognition method for generating the confrontation network based on multiple tasks is characterized by comprising the following specific steps:
(1) inputting pedestrian video frame sequences under different visual angles, and constructing gait template characteristics:
vector xiIs the gait template characteristic under the visual angle i, nvIs the number of all views;
(2) for any different view angles u, v, encoding the corresponding gait template features x by using a convolutional neural networkuTo the hidden space and obtaining a hidden representation in the hidden space as zu；
(3) Representing z implicitly in implicit spaceuCarrying out visual angle transformation to an angle v to obtain a hidden representation zv；
(4) Will implicitly represent zvAnd the channel one-hot code is used as input, and the pseudo-template characteristic under the output angle v is obtained after the generation network in the countermeasure network is generated through multiple tasks
(5) According to template characteristics xuAnd dummy features
Calculating the pixel-by-pixel loss Lp；
(6) Generating discriminative networks in countermeasure networks using multitasking, according to template features xuAnd dummy features
Computing a multitasking confrontation loss La；
(7) Weighting pixel-by-pixel loss and multi-task counterloss according to total loss L ═ Lp+αLaTraining a multitask to generate an antagonistic network;
alpha is used as a hyper-parameter to balance pixel-by-pixel loss and multi-task countermeasure loss, and after a final loss function is defined, parameters of an encoder, a view transformation layer, a generator and a discriminator are alternately updated by using a back propagation algorithm;
the gait template is constructed by the following steps:
(1) performing foreground and background separation on each frame of an original video frame sequence, and extracting a gait contour map; and translating and zooming the gait outline image to the image center to obtain a centralized gait outline image sequence B1，B2，B3，...，Bn}；
(2) For each frame in the sequence of centered gait profiles, the normalized period position r is calculatedtA normalized cycle position of a t-th frame of the centered gait profile;
(3) sequence B of gait profile according to centralizationtAnd corresponding normalized period position rtStructure of ncPer channel energy cycle map PEITemplate characteristics; the value at the (x, y) coordinate of the kth channel in the periodic energy plot is calculated as:
wherein:
m represents the size of the time domain window covered by each channel, and T (k) determines the range of the time domain covered by each channel;
the multi-task generation countermeasure network mainly comprises four parts, namely an encoder, a visual angle conversion layer, a generator and a discriminator; wherein:
(a) an encoder: using a convolutional neural network as an encoder in a model, the input x of the encoderuIs a PEI template with a viewing angle u; the input scale size is 64 × 64 × nc(ii) a Each channel of the PEI template is independently sent into an encoder, and then effective characteristic implicit expression z is extracted by using time pooling and passing through a full connection layeruAs the output of the encoder;
(b) viewing angle conversion layer: assuming that the gait data is distributed on a high-dimensional manifold along with the change of the view angle, the change of the view angle is realized under the condition that samples in the manifold move towards a specific direction to keep identity information; hidden representation z given a hidden spaceuThe view transformation can be described as:
wherein h isiRepresenting the transformation vector from view i-1 to view i, zvIs a hidden representation at view angle v;
(c) a generator: by representing z implicitlyvGenerating a gait template under a visual angle v as input; the generator consists of five deconvolution layersComposing and using the ReLU as an excitation function; randomly selecting one channel of the PEI template for reconstruction so as to ensure that the hidden representation can store the information of all the channels; define the input of the generator as [ z ]v，c]And c is a one-hot encoded representation of the channel;
(d) a discriminator: the structure of the discriminator is the same as that of the encoder except that time pooling operation is not adopted and the dimension of an output layer is different from that of the encoder; defining the output dimension of the discriminator as nv+nc+ndWherein n isvIs the number of viewing angles, ncIs the number of PEI channels, ndIs the number of different identities in the training set; the output of the discriminator corresponds to nv+nc+ndSub-discriminators sharing parameters except the last layer among the sub-discriminators; each arbiter is responsible for a different arbitration task to determine whether the generated samples belong to a particular distribution.
2. The gait recognition method according to claim 1, characterized in that the perspective transformation step is:
implicit representation z at a given view angle uuImplicit representation z transformed to view angle vvThe calculation process of (2) is as follows:
wherein h isiRepresenting the transformation vector from view i-1 to view i.
3. The gait recognition method according to claim 1, characterized in that the loss L ispAnd LaThe calculation steps are as follows:
(1) according to template characteristics xvAnd dummy features
Calculate the pixel-by-pixel loss:
(2) according to template characteristics xvAnd dummy features
Computing the multitask countermeasure loss:
where E represents the expectation for the corresponding sample set, | | · | | | ceiling1Representing the norm of L1, and the vector s is the one-hot code of identity, channel and angle information; d (-) represents the output of the discriminator.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710373017.5A CN107085716B (en) | 2017-05-24 | 2017-05-24 | Cross-view gait recognition method based on multi-task generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710373017.5A CN107085716B (en) | 2017-05-24 | 2017-05-24 | Cross-view gait recognition method based on multi-task generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107085716A CN107085716A (en) | 2017-08-22 |
| CN107085716B true CN107085716B (en) | 2021-06-04 |
Family
ID=59607349
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710373017.5A Active CN107085716B (en) | 2017-05-24 | 2017-05-24 | Cross-view gait recognition method based on multi-task generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107085716B (en) |
Families Citing this family (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563428B (en) * | 2017-08-25 | 2019-07-02 | 西安电子科技大学 | Based on the Classification of Polarimetric SAR Image method for generating confrontation network |
| CN107609587B (en) * | 2017-09-11 | 2020-08-18 | 浙江工业大学 | Multi-class multi-view data generation method for generating countermeasure network based on deep convolution |
| WO2019056257A1 (en) * | 2017-09-21 | 2019-03-28 | Nokia Technologies Oy | Apparatus, method and computer program product for biometric recognition |
| CN109697389B (en) * | 2017-10-23 | 2021-10-01 | 北京京东尚科信息技术有限公司 | Identity recognition method and device |
| CN108009568A (en) * | 2017-11-14 | 2018-05-08 | 华南理工大学 | A kind of pedestrian detection method based on WGAN models |
| CN107993210A (en) * | 2017-11-30 | 2018-05-04 | 北京小米移动软件有限公司 | Image repair method, device and computer-readable recording medium |
| CN108256627A (en) * | 2017-12-29 | 2018-07-06 | 中国科学院自动化研究所 | The mutual generating apparatus of audio-visual information and its training system that generation network is fought based on cycle |
| CN108171701B (en) * | 2018-01-15 | 2021-06-22 | 复旦大学 | Significance detection method based on U network and counterstudy |
| CN108491380B (en) * | 2018-03-12 | 2021-11-23 | 思必驰科技股份有限公司 | Anti-multitask training method for spoken language understanding |
| CN108596026B (en) * | 2018-03-16 | 2020-06-30 | 中国科学院自动化研究所 | Cross-view gait recognition device and training method based on double-flow generation countermeasure network |
| CN108629823B (en) * | 2018-04-10 | 2022-09-06 | 北京京东尚科信息技术有限公司 | Method and device for generating multi-view image |
| CN110443232B (en) | 2018-04-12 | 2022-03-25 | 腾讯科技（深圳）有限公司 | Video processing method and related device, image processing method and related device |
| CN108846355B (en) * | 2018-06-11 | 2020-04-28 | 腾讯科技（深圳）有限公司 | Image processing method, face recognition device and computer equipment |
| CN109583298B (en) * | 2018-10-26 | 2023-05-02 | 复旦大学 | Cross-view gait recognition method based on set |
| CN109543546B (en) * | 2018-10-26 | 2022-12-20 | 复旦大学 | Gait age estimation method based on depth sequence distribution regression |
| CN111144165B (en) * | 2018-11-02 | 2024-04-12 | 银河水滴科技(宁波)有限公司 | Gait information identification method, system and storage medium |
| CN111144171A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Abnormal crowd information identification method, system and storage medium |
| CN111144167A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Gait information identification optimization method, system and storage medium |
| CN111144170A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Gait information registration method, system and storage medium |
| CN109726654A (en) * | 2018-12-19 | 2019-05-07 | 河海大学 | A kind of gait recognition method based on generation confrontation network |
| EP3888017A4 (en) * | 2019-01-04 | 2022-08-03 | Sony Corporation of America | Multi-forecast networks |
| CN110119780B (en) * | 2019-05-10 | 2020-11-27 | 西北工业大学 | Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network |
| CN110381268B (en) * | 2019-06-25 | 2021-10-01 | 达闼机器人有限公司 | Method, device, storage medium and electronic equipment for generating video |
| CN110516525B (en) * | 2019-07-01 | 2021-10-08 | 杭州电子科技大学 | SAR image target recognition method based on GAN and SVM |
| CN110659586B (en) * | 2019-08-31 | 2022-03-15 | 电子科技大学 | Gait recognition method based on identity-preserving cyclic generation type confrontation network |
| CN112488984A (en) * | 2019-09-11 | 2021-03-12 | 中信戴卡股份有限公司 | Method and device for acquiring defect picture generation network and defect picture generation method |
| CN111639580B (en) * | 2020-05-25 | 2023-07-18 | 浙江工商大学 | Gait recognition method combining feature separation model and visual angle conversion model |
| CN112001254B (en) * | 2020-07-23 | 2021-07-13 | 浙江大华技术股份有限公司 | Pedestrian identification method and related device |
| CN112733704B (en) * | 2021-01-07 | 2023-04-07 | 浙江大学 | Image processing method, electronic device, and computer-readable storage medium |
| CN112975968B (en) * | 2021-02-26 | 2022-06-28 | 同济大学 | Mechanical arm imitation learning method based on third visual angle variable main body demonstration video |
| CN113112572B (en) * | 2021-04-13 | 2022-09-06 | 复旦大学 | Hidden space search-based image editing method guided by hand-drawn sketch |
| CN113111797B (en) * | 2021-04-19 | 2024-02-13 | 杭州电子科技大学 | Cross-view gait recognition method combining self-encoder and view transformation model |
| CN113657463B (en) * | 2021-07-28 | 2023-04-07 | 浙江大华技术股份有限公司 | Gait recognition model training method, gait recognition method and related device |
| CN114120076B (en) * | 2022-01-24 | 2022-04-29 | 武汉大学 | Cross-view video gait recognition method based on gait motion estimation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102426645A (en) * | 2011-08-30 | 2012-04-25 | 北京航空航天大学 | Multi-view and multi-state gait recognition method |
| CN104299012A (en) * | 2014-10-28 | 2015-01-21 | 中国科学院自动化研究所 | Gait recognition method based on deep learning |
| CN105574510A (en) * | 2015-12-18 | 2016-05-11 | 北京邮电大学 | Gait identification method and device |
| CN106096532A (en) * | 2016-06-03 | 2016-11-09 | 山东大学 | A kind of based on tensor simultaneous discriminant analysis across visual angle gait recognition method |
-
2017
- 2017-05-24 CN CN201710373017.5A patent/CN107085716B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102426645A (en) * | 2011-08-30 | 2012-04-25 | 北京航空航天大学 | Multi-view and multi-state gait recognition method |
| CN104299012A (en) * | 2014-10-28 | 2015-01-21 | 中国科学院自动化研究所 | Gait recognition method based on deep learning |
| CN105574510A (en) * | 2015-12-18 | 2016-05-11 | 北京邮电大学 | Gait identification method and device |
| CN106096532A (en) * | 2016-06-03 | 2016-11-09 | 山东大学 | A kind of based on tensor simultaneous discriminant analysis across visual angle gait recognition method |
Non-Patent Citations (2)
| Title |
|---|
| A Comprehensive Study on Cross-View Gait Based Human Identification with Deep CNNs;Zifeng Wu et al;《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》;20170228;全文 * |
| 基于深度学习的多视角步态识别算法设计与研究;汪青;《中国优秀硕士学位论文全文数据库 信息科技辑》;20170515;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107085716A (en) | 2017-08-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107085716B (en) | Cross-view gait recognition method based on multi-task generation countermeasure network | |
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| CN110084156B (en) | Gait feature extraction method and pedestrian identity recognition method based on gait features | |
| Hu et al. | Incremental tensor subspace learning and its applications to foreground segmentation and tracking | |
| CN107169117B (en) | Hand-drawn human motion retrieval method based on automatic encoder and DTW | |
| Guo et al. | JointPruning: Pruning networks along multiple dimensions for efficient point cloud processing | |
| Javed et al. | OR-PCA with dynamic feature selection for robust background subtraction | |
| CN112488205A (en) | Neural network image classification and identification method based on optimized KPCA algorithm | |
| Khan et al. | A customized Gabor filter for unsupervised color image segmentation | |
| Wang et al. | Incremental MPCA for color object tracking | |
| CN113095149A (en) | Full-head texture network structure based on single face image and generation method | |
| CN112418041A (en) | Multi-pose face recognition method based on face orthogonalization | |
| CN104077742B (en) | Human face sketch synthetic method and system based on Gabor characteristic | |
| Basavaiah et al. | Robust Feature Extraction and Classification Based Automated Human Action Recognition System for Multiple Datasets. | |
| Pini et al. | Learning to generate facial depth maps | |
| Tin et al. | Gender and age estimation based on facial images | |
| CN113887509B (en) | Rapid multi-modal video face recognition method based on image set | |
| Pakulich et al. | Age recognition from facial images using convolutional neural networks | |
| Langs et al. | Modeling the structure of multivariate manifolds: Shape maps | |
| Ali et al. | Deep multi view spatio temporal spectral feature embedding on skeletal sign language videos for recognition | |
| Chen et al. | Video foreground detection algorithm based on fast principal component pursuit and motion saliency | |
| Rajapakse et al. | NMF vs ICA for face recognition | |
| Maghari et al. | Quantitative analysis on PCA-based statistical 3D face shape modeling. | |
| CN114360058A (en) | Cross-visual angle gait recognition method based on walking visual angle prediction | |
| CN116543419B (en) | Hotel health personnel wearing detection method and system based on embedded platform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |