CN109902605A - A kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling - Google Patents
A kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling Download PDFInfo
- Publication number
- CN109902605A CN109902605A CN201910126867.4A CN201910126867A CN109902605A CN 109902605 A CN109902605 A CN 109902605A CN 201910126867 A CN201910126867 A CN 201910126867A CN 109902605 A CN109902605 A CN 109902605A
- Authority
- CN
- China
- Prior art keywords
- gait
- gait recognition
- monoergic
- training
- adaptivenon
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention belongs to Gait Recognition fields, and in particular to a kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling.Including training process and identification process, training process includes that a whole picture gait energy diagram is divided into head, body and leg three parts, the garbage of removal left and right sides when wherein head zone is divided.The upper part of the body region to differ greatly is removed, the gait information of same people is made to be more nearly and concentrate, will only change at most most apparent leg area in walking process and be sent into training, the leg area of sample when training set only includes normal morphology at this time in Gait Recognition network;Leg area when identification process includes leg area when using backpack configuration respectively after sample decomposition and wears housing form after sample decomposition obtains recognition effect as test set.Method proposed by the present invention can cope with the identification after pedestrian's form changes well, have more practicability, can be widely applied to Gait Recognition field, effectively improve recognition effect.
Description
Technical field
The invention belongs to Gait Recognition fields, and in particular to a kind of Gait Recognition side based on monoergic figure adaptivenon-uniform sampling
Method.
Background technique
The extraction of moving target characteristic parameter is to realize an important link of individual segregation identification.Common feature extraction
There are two types of methods: 1) (the Human Identification from Freestyle Walks using of the method based on model
Posture-Based Gait Feature[J].IEEE Transactions on Information Forensics&
Security, 2017, PP (99): 1-1.) 2) method (the Investigating the use of Motion- based on appearance
based Features from Optical Flow for Gait Recognition[J].Neurocomputing,
2017.).Gait energy diagram is under the jurisdiction of the feature extracting method based on appearance.It believes personage in the walking of a gait cycle
Breath compressed, the processing such as frame difference, gets up to be formed the bigger gait information figure of information content for the information aggregate of dispersion.Gait energy
Figure is one of most important gait feature characterization mode in the Gait Recognition of view-based access control model, because its can show by force very much it is rudimentary
Feature by people in the industry to widely be praised highly.It, mostly will be whole in the research for carrying out Gait Recognition using gait energy diagram at present
Energy diagram is sent into Gait Recognition network.However for the Gait Recognition of cross-modality, the wheel of people's different parts in different shape
Exterior feature variation it is not consistent, such as: people is larger with upper part of the body profile difference when normal morphology in knapsack because above the waist than
The profile of packet is normally had more.At this point, difference between sample different shape is asked there may be diversity ratio class inherited in class is big
Topic, to influence recognition effect.
Summary of the invention
The purpose of the present invention is to provide a kind of gait recognition methods based on monoergic figure adaptivenon-uniform sampling, can solve
The certainly big problem of diversity ratio class inherited in class caused by sample different shape.
A kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling, including training process and identification process, specifically
Include the following steps:
Step 1, in the training process, neural network adjusts parameter, that is, weight and biasing by error back propagation
It is whole, to establish input i.e. gait energy diagram and export the mapping between personal part at once;
Step 2, in identification process, assess the training effect of established neural network and the validity of proposed method.
A kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling, step 1 specifically comprise the following steps:
Gait energy diagram is split by step 1.1, and a width figure is divided into head, body and leg three parts, wherein
The garbage of removal left and right sides when head zone is divided;
The upper part of the body region that step 1.2, removal differ greatly, makes the gait information of same people be more nearly and concentrate, only
At most most apparent leg area will be changed in walking process and be sent into training in Gait Recognition network, training set only includes just at this time
The leg area of sample when normal form.
A kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling, Gait Recognition network be AlexNet and
GoogLeNet network carries out two networks since this experimental data set picture size is different from the initial setting of two kinds of networks
Following adjustment: the first layer convolution kernel size of AlexNet is changed to 3*3, step-length is set as 2, and convolution nuclear volume is set as 96, deletes
The pond layer of first layer exports the image of 96 27*27 after first layer convolution, is connected with second layer convolution;It will
The first layer of GoogLeNet is deleted, directly by the second layer convolution pond of image input GoogLeNet.
A kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling, step 2 specifically comprise the following steps:
Step 2.1, leg area when using backpack configuration after sample decomposition are as test set;
Step 2.2 uses leg area when wearing housing form after sample decomposition as test set.
The beneficial effects of the present invention are:
Gait energy diagram is split by the present invention, and removal leads to the core part of differences between samples, by sample different shape
Between change the smallest leg area for identification, thus effectively solve sample different shape between difference there may be differences in class
The problem bigger than class inherited, more rationally.Carry out cross-modality Gait Recognition when, use integral energy figure as input and
The discrimination of the mentioned method of this patent compares such as Fig. 4, and as can be seen from the figure the recognition performance of the mentioned method of the present invention is better than biography
System method, discrimination have larger promotion, and the loss decline for verifying collection also becomes apparent from.It can be seen that the mentioned method of the present invention
There is preferable superiority in terms of recognition performance, can be widely applied to Gait Recognition field, effectively improve recognition effect.
Detailed description of the invention
Fig. 1 is the flow chart for the Gait Recognition divided based on monoergic figure;
Fig. 2 is that monoergic figure divides schematic diagram;
Fig. 3 is with the gait energy diagram under a group traveling together's different shape;
Figure compared with Fig. 4 mentions the discrimination of method with the present invention by conventional method.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings:
As shown in Fig. 1, the flow chart of the Gait Recognition to be divided based on monoergic figure (is carried out whole gait energy diagram
Segmentation, takes the gait energy diagram of leg area as the input of neural network, and the output of network is the identity information of pedestrian.This reality
The leg energy diagram for choosing normal morphology in CASIA (B) gait data library is tested as training set, backpack configuration and wears housing form
Leg energy diagram respectively as test set, Classification and Identification is carried out using softmax classifier.Know in the gait for carrying out cross-modality
Not Shi Yan in, only with the gait energy diagram of single visual angle, 90 ° of gait energy diagram is chosen in this experiment.The reason is that side visual angle
Human body contour outline contain more valuable information).
As shown in Fig. 2, divide schematic diagram for monoergic figure (to be split gait energy diagram, a width figure is divided into
Head, middle part (body) and leg three parts also remove the garbage of the left and right sides when wherein head zone is divided).
As shown in Fig. 3, for the gait energy diagram under a group traveling together's different shape (from left to right successively are as follows: when knapsack
Form, form when wearing housing, normal morphology).
As shown in Fig. 4, (leg area is tested for figure compared with mentioning the discrimination (%) of method with the present invention by conventional method
Card discrimination is higher than the verifying discrimination of integral energy figure, and the loss decline for verifying collection also becomes apparent from).
Technical solution:
Using gait energy diagram dividing method CASIA provided by Institute of Automation, CAS (B) step proposed by the present invention
It is tested on state database.CASIA (B) database include 124 people 0 °, 18 °, 36 °, 54 °, 72 °, 108 °, 126 °,
144 °, 162 °, the video sequence walked under 180 ° of 11 visual angles, wherein everyone includes 6 normal walking video sequences, and 2
Knapsack video sequence, 2 are worn housing video sequence.Here gait energy diagram feature is extracted to video sequence, obtains different shape
Gait energy diagram, such as Fig. 3.
When carrying out the Gait Recognition of cross-modality, conventional method is compared with the discrimination of the mentioned method of the present invention such as Fig. 4 institute
Show, it can be seen that being compared with the traditional method, the recognition performance of the mentioned method of the present invention has larger promotion better than discrimination,
And the loss decline of verifying collection also becomes apparent from, this is because removal leads to the core part of differences between samples, by sample different shape
Between change the smallest leg area for identification, there may be differences in class for the difference that can effectively solve between sample different shape
The problem bigger than class inherited.It can be seen that the mentioned method of the present invention has preferable superiority in terms of recognition performance.Although
It only uses leg energy diagram and is trained identification, have ignored some other information for helping to identify identity, such as trunk information,
But it will lead to the problem that diversity ratio class inherited is big in class since trunk information change is larger between different shape, so tradeoff benefit
Disadvantage, only selection leg energy diagram progress Classification and Identification is more conducive to avoid risk in scheme, and experiment also indicates that, leg is used alone
The recognition effect of portion's energy diagram is better than the recognition effect using integral energy figure, this is because between same a group traveling together's different shape
Energy diagram variation in leg is minimum.
Since in practical applications, the form of pedestrian is changeable, method proposed by the present invention can cope with pedestrian's shape well
State change after identification, have more practicability.It the composite can be widely applied to Gait Recognition field, effectively improve identification
Effect.
Claims (4)
1. a kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling, which is characterized in that including training process and identification
Process specifically comprises the following steps:
Step 1, in the training process, neural network is adjusted parameter, that is, weight and biasing by error back propagation, from
And it establishes input i.e. gait energy diagram and exports the mapping between personal part at once;
Step 2, in identification process, assess the training effect of established neural network and the validity of proposed method.
2. a kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling according to claim 1, which is characterized in that institute
Step 1 is stated to specifically comprise the following steps:
Gait energy diagram is split by step 1.1, a width figure is divided into head, body and leg three parts, wherein head
The garbage of removal left and right sides when region segmentation;
The big upper part of the body region of step 1.2, removal difference, makes the gait information of same people be more nearly and concentrate, only will walking
The at most most apparent leg area of variation is sent into training in Gait Recognition network in the process, and training set only includes normal morphology at this time
When sample leg area.
3. a kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling according to claim 1 or claim 2, feature exist
In the Gait Recognition network is AlexNet and GoogLeNet network, due to this experimental data set picture size and two kinds of nets
The initial setting of network is different, and two networks have been carried out following adjustment: the first layer convolution kernel size of AlexNet is changed to 3*3,
Step-length is set as 2, and convolution nuclear volume is set as 96, deletes the pond layer of first layer, exports 96 27*27's after first layer convolution
Image is connected with second layer convolution;The first layer of GoogLeNet is deleted, directly by the second of image input GoogLeNet
Layer convolution pond.
4. a kind of gait recognition method based on monoergic figure adaptivenon-uniform sampling according to claim 1, which is characterized in that institute
Step 2 is stated to specifically comprise the following steps:
Step 2.1, leg area when using backpack configuration after sample decomposition are as test set;
Step 2.2 uses leg area when wearing housing form after sample decomposition as test set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910126867.4A CN109902605B (en) | 2019-02-20 | 2019-02-20 | Gait recognition method based on single energy map adaptive segmentation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910126867.4A CN109902605B (en) | 2019-02-20 | 2019-02-20 | Gait recognition method based on single energy map adaptive segmentation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109902605A true CN109902605A (en) | 2019-06-18 |
CN109902605B CN109902605B (en) | 2023-04-07 |
Family
ID=66945029
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910126867.4A Active CN109902605B (en) | 2019-02-20 | 2019-02-20 | Gait recognition method based on single energy map adaptive segmentation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109902605B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016065534A1 (en) * | 2014-10-28 | 2016-05-06 | 中国科学院自动化研究所 | Deep learning-based gait recognition method |
CN107292250A (en) * | 2017-05-31 | 2017-10-24 | 西安科技大学 | A kind of gait recognition method based on deep neural network |
CN108460340A (en) * | 2018-02-05 | 2018-08-28 | 北京工业大学 | A kind of gait recognition method based on the dense convolutional neural networks of 3D |
-
2019
- 2019-02-20 CN CN201910126867.4A patent/CN109902605B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016065534A1 (en) * | 2014-10-28 | 2016-05-06 | 中国科学院自动化研究所 | Deep learning-based gait recognition method |
CN107292250A (en) * | 2017-05-31 | 2017-10-24 | 西安科技大学 | A kind of gait recognition method based on deep neural network |
CN108460340A (en) * | 2018-02-05 | 2018-08-28 | 北京工业大学 | A kind of gait recognition method based on the dense convolutional neural networks of 3D |
Non-Patent Citations (3)
Title |
---|
刘晓芳等: "基于视觉的步态识别研究综述", 《小型微型计算机系统》 * |
李凯等: "融合腿部局部特征的步态识别方法", 《计算机工程与设计》 * |
王修晖等: "基于连续密度隐马尔可夫模型的人体步态识别", 《模式识别与人工智能》 * |
Also Published As
Publication number | Publication date |
---|---|
CN109902605B (en) | 2023-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yuan et al. | Fingerprint liveness detection using an improved CNN with image scale equalization | |
CN107194341B (en) | Face recognition method and system based on fusion of Maxout multi-convolution neural network | |
CN104881637B (en) | Multimodal information system and its fusion method based on heat transfer agent and target tracking | |
CN106529477B (en) | Video human Activity recognition method based on significant track and temporal-spatial evolution information | |
CN105868694B (en) | The bimodal emotion recognition method and system acted based on facial expression and eyeball | |
CN105913038B (en) | A kind of micro- expression recognition method of dynamic based on video | |
CN107590452A (en) | A kind of personal identification method and device based on gait and face fusion | |
CN107145842A (en) | With reference to LBP characteristic patterns and the face identification method of convolutional neural networks | |
Wu et al. | End-to-end chromosome Karyotyping with data augmentation using GAN | |
CN103714181B (en) | A kind of hierarchical particular persons search method | |
CN105956546A (en) | Emotion recognition method based on EEG signals | |
CN102915436A (en) | Sparse representation face recognition method based on intra-class variation dictionary and training image | |
Guo et al. | Extended local binary patterns for efficient and robust spontaneous facial micro-expression recognition | |
CN103632147A (en) | System and method for implementing standardized semantic description of facial features | |
CN103077378B (en) | Contactless face recognition algorithms based on extension eight neighborhood Local textural feature and system of registering | |
CN106548149A (en) | The recognition methods of the micro- facial expression image sequence of face in monitor video sequence | |
Zhang et al. | Gait-based age estimation with deep convolutional neural network | |
CN104008364B (en) | Face identification method | |
CN107545243A (en) | Yellow race's face identification method based on depth convolution model | |
CN108898123A (en) | A kind of face identification method based on signature analysis | |
CN105138995A (en) | Time-invariant and view-invariant human action identification method based on skeleton information | |
CN109544523A (en) | Quality of human face image evaluation method and device based on more attribute face alignments | |
CN109726619A (en) | A kind of convolutional neural networks face identification method and system based on parameter sharing | |
Wang et al. | Pyramid-based multi-scale lbp features for face recognition | |
CN108537143B (en) | A kind of face identification method and system based on key area aspect ratio pair |
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 |