CN109902605B - Gait recognition method based on single energy map adaptive segmentation - Google Patents
Gait recognition method based on single energy map adaptive segmentation Download PDFInfo
- Publication number
- CN109902605B CN109902605B CN201910126867.4A CN201910126867A CN109902605B CN 109902605 B CN109902605 B CN 109902605B CN 201910126867 A CN201910126867 A CN 201910126867A CN 109902605 B CN109902605 B CN 109902605B
- Authority
- CN
- China
- Prior art keywords
- gait
- recognition
- training
- leg
- sample
- 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
Abstract
The invention belongs to the field of gait recognition, and particularly relates to a gait recognition method based on single energy map self-adaptive segmentation. The gait energy image recognition method comprises a training process and a recognition process, wherein the training process comprises the step of dividing a whole gait energy image into a head part, a body part and a leg part, and useless information on the left side and the right side is removed when a head area is divided. Removing the upper half body area with larger difference, enabling the gait information of the same person to be closer and more concentrated, and only sending the leg area which changes most obviously in the walking process into a gait recognition network for training, wherein the training set only comprises the leg area of the sample in the normal form; the identification process comprises the steps of respectively using the leg area obtained after the sample is divided in the backpack form and the leg area obtained after the sample is divided in the jacket-wearing form as a test set, and obtaining the identification effect. The method provided by the invention can be used for well recognizing the pedestrian with changed shape, has higher practicability, can be widely applied to the field of gait recognition, and effectively improves the recognition effect.
Description
Technical Field
The invention belongs to the field of gait recognition, and particularly relates to a gait recognition method based on single energy map self-adaptive segmentation.
Background
The extraction of the characteristic parameters of the moving target is an important link for realizing individual classification and identification. There are two common methods of feature extraction: 1) model-Based methods (Human Identification from free weights walk using Posture-Based gap Feature [ J ]. IEEE Transactions on Information forms & Security,2017, PP (99): 1-1.) 2) appearance-Based methods (Investigating the use of Motion-Based gaps from Optical Flow for gap registration [ J ]. Neurocoputing, 2017.). The gait energy map belongs to an appearance-based feature extraction method. The walking information of a person in a gait cycle is processed by compression, frame difference and the like, and the dispersed information is gathered to form a gait information image with larger information quantity. The gait energy map is one of the most important gait feature characterization methods in vision-based gait recognition, and is widely appreciated by those in the industry because it can strongly represent low-level features. In the current research of using a gait energy map to carry out gait recognition, the whole energy map is mostly sent into a gait recognition network. However, in cross-modality gait recognition, the contour of different parts of a human varies in different modalities and is not uniform, for example: when a person is carrying the backpack, the upper half body contour has a larger difference with the normal shape because the upper half body has a larger contour than the normal one. At this time, the difference between different forms of the sample may cause a problem that the intra-class difference is larger than the inter-class difference, thereby affecting the recognition effect.
Disclosure of Invention
The invention aims to provide a gait recognition method based on single energy map adaptive segmentation, which can solve the problem that intra-class differences are larger than inter-class differences caused by different forms of a sample.
A gait recognition method based on single energy map adaptive segmentation comprises a training process and a recognition process, and specifically comprises the following steps:
and 2, evaluating the training effect of the established neural network and the effectiveness of the method in the identification process.
The gait recognition method based on the single energy map self-adaptive segmentation specifically comprises the following steps in step 1:
1.1, segmenting a gait energy image, and segmenting one image into a head part, a body part and a leg part, wherein useless information on the left side and the right side is removed when a head area is segmented;
and 1.2, removing the upper half body area with larger difference, enabling the gait information of the same person to be closer and more concentrated, and only sending the leg area which changes most obviously in the walking process into a gait recognition network for training, wherein the training set only comprises the leg area of the sample in a normal form.
The gait recognition network is AlexNet and GoogLeNet, and because the size of the image in the experimental data set is different from the initial settings of the two networks, the two networks are adjusted as follows: changing the size of a first layer of convolution kernels of AlexNet into 3*3, setting the step length to be 2, setting the number of the convolution kernels to be 96, deleting the pooling layer of the first layer, outputting 96 images of 27 x 27 after the first layer of convolution, and connecting the images with a second layer of convolution; the first layer of google lenet is removed and the image input is pooled directly into the second layer convolution of google lenet.
The gait recognition method based on the single energy map self-adaptive segmentation specifically comprises the following steps in step 2:
step 2.1, using the leg region after sample segmentation in the backpack form as a test set;
and 2.2, taking the leg region obtained after the sample is segmented in the coat-wearing form as a test set.
The invention has the beneficial effects that:
according to the gait energy image recognition method, the gait energy image is divided, the core part causing the sample difference is removed, and the leg part area with the minimum change among different forms of the sample is used for recognition, so that the problem that the intra-class difference is larger than the inter-class difference possibly caused by the difference among the different forms of the sample is effectively solved, and the gait energy image recognition method is more reasonable. When the trans-form gait recognition is carried out, the whole energy graph is used as input and compared with the recognition rate of the method provided by the patent as shown in figure 4, and the recognition performance of the method provided by the invention is superior to that of the traditional method, the recognition rate is greatly improved, and the loss reduction of a verification set is more obvious. Therefore, the method provided by the invention has better superiority in the aspect of identification performance, can be widely applied to the field of gait identification, and effectively improves the identification effect.
Drawings
FIG. 1 is a flow chart of gait recognition based on single energy map segmentation;
FIG. 2 is a schematic diagram of a single energy map segmentation;
FIG. 3 is a diagram of gait energy of the same pedestrian in different forms;
FIG. 4 is a graph comparing the recognition rates of the conventional method and the method of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings:
as shown in the attached figure 1, the gait recognition flow chart based on single energy map segmentation is shown (the whole gait energy map is segmented, the gait energy map of a leg area is taken as the input of a neural network, the output of the network is the identity information of a pedestrian).
As shown in fig. 2, it is a schematic diagram of single energy map segmentation (a gait energy map is segmented, and one map is segmented into three parts of a head, a middle part (a body) and a leg, wherein useless information on the left and right sides is also removed when the head region is segmented).
As shown in the attached figure 3, the gait energy diagram is the gait energy diagram of the same pedestrian under different forms (from left to right, the form during knapsack, the form during coat wearing and the normal form).
As shown in fig. 4, the graph is a comparison of the recognition rate (%) of the conventional method and the method of the present invention (the verified recognition rate of the leg region is higher than that of the overall energy map, and the loss of the verified set is more obvious).
The technical scheme is as follows:
the gait energy graph segmentation method provided by the invention is applied to carry out experiments on a CASIA (B) gait database provided by the automation of Chinese academy of sciences. The CASIA (B) database contains 124 video sequences of a person walking at 11 views of 0 °, 18 °, 36 °, 54 °, 72 °, 108 °, 126 °, 144 °, 162 °, 180 °, wherein each person contains 6 normal walking video sequences, 2 backpack video sequences, 2 coat-wearing video sequences. Here, gait energy map features are extracted from the video sequence to obtain gait energy maps of different forms, such as fig. 3.
When cross-form gait recognition is carried out, the recognition rate comparison of the traditional method and the method provided by the invention is shown in fig. 4, and it can be known from the figure that compared with the traditional method, the recognition performance of the method provided by the invention is greatly improved better than the recognition rate, and the loss reduction of a verification set is more obvious, because the core part causing the sample difference is removed, the leg part area with the minimum change among different forms of the sample is used for recognition, and the problem that the intra-class difference is possibly generated by the difference among the different forms of the sample and is larger than the inter-class difference can be effectively solved. Therefore, the method provided by the invention has better superiority in the aspect of identification performance. Although only the leg energy diagram is used for training and recognition, and some other information which is helpful for identity recognition, such as trunk information and the like, is ignored, the problem that the trunk information changes greatly among different forms to cause the difference in the class to be larger than the difference among the classes is solved, so that the advantages and disadvantages are balanced, and the scheme that only the leg energy diagram is selected for classification and recognition is more helpful for avoiding risks.
In practical application, the pedestrian shape is changeable, so that the method provided by the invention can be used for well recognizing the pedestrian shape after being changed, and has higher practicability. The invention can be widely applied to the field of gait recognition, and effectively improves the recognition effect.
Claims (1)
1. A gait recognition method based on single energy diagram self-adaptive segmentation is characterized by comprising a training process and a recognition process, and specifically comprises the following steps:
step 1, in the training process, the neural network adjusts parameters, namely weight and bias, through error back propagation, and accordingly a mapping between an input gait energy map and an output pedestrian identity is established;
step 2, in the identification process, evaluating the training effect of the established neural network and the effectiveness of the method;
the step 1 specifically comprises the following steps:
1.1, segmenting a gait energy image, and segmenting one image into a head part, a body part and a leg part, wherein useless information on the left side and the right side is removed when a head area is segmented;
step 1.2, removing the upper half body area with large difference, enabling the gait information of the same person to be closer and more concentrated, and only sending the leg area which changes most obviously in the walking process into a gait recognition network for training, wherein the training set only comprises the leg area of the sample in a normal form;
the gait recognition networks are AlexNet and GoogLeNet networks, and because the size of the image in the experimental data set is different from the initial settings of the two networks, the two networks are adjusted as follows: changing the size of a first layer of convolution kernels of AlexNet into 3*3, setting the step length to be 2, setting the number of the convolution kernels to be 96, deleting the pooling layer of the first layer, outputting 96 images of 27 x 27 after the first layer of convolution, and connecting the images with a second layer of convolution; deleting the first layer of GoogLeNet, and directly inputting the image into a second layer of convolution pool of GoogLeNet;
the step 2 specifically comprises the following steps:
step 2.1, taking the leg area obtained after the sample is divided in the backpack form as a test set;
and 2.2, taking the leg region obtained after the sample is segmented in the coat-wearing form as a 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 CN109902605A (en) | 2019-06-18 |
| CN109902605B true 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 |
|---|
| 基于视觉的步态识别研究综述;刘晓芳等;《小型微型计算机系统》;20180815(第08期);全文 * |
| 基于连续密度隐马尔可夫模型的人体步态识别;王修晖等;《模式识别与人工智能》;20160815(第08期);全文 * |
| 融合腿部局部特征的步态识别方法;李凯等;《计算机工程与设计》;20160516(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109902605A (en) | 2019-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711281B (en) | Pedestrian re-recognition and feature recognition fusion method based on deep learning | |
| CN107633207B (en) | AU characteristic recognition methods, device and storage medium | |
| CN108182441B (en) | Parallel multichannel convolutional neural network, construction method and image feature extraction method | |
| CN107220611B (en) | Space-time feature extraction method based on deep neural network | |
| CN111339847B (en) | Face emotion recognition method based on graph convolution neural network | |
| CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
| CN112184752A (en) | Video target tracking method based on pyramid convolution | |
| CN106845415B (en) | Pedestrian fine identification method and device based on deep learning | |
| CN106778584A (en) | A kind of face age estimation method based on further feature Yu shallow-layer Fusion Features | |
| CN110033007B (en) | Pedestrian clothing attribute identification method based on depth attitude estimation and multi-feature fusion | |
| CN107145842A (en) | With reference to LBP characteristic patterns and the face identification method of convolutional neural networks | |
| CN107463917B (en) | Improved LTP and two-dimensional bidirectional PCA fusion-based face feature extraction method | |
| CN105975932B (en) | Gait Recognition classification method based on time series shapelet | |
| CN110378208B (en) | Behavior identification method based on deep residual error network | |
| CN109902565B (en) | Multi-feature fusion human behavior recognition method | |
| CN110728179A (en) | Pig face identification method adopting multi-path convolutional neural network | |
| CN107944398A (en) | Based on depth characteristic association list diagram image set face identification method, device and medium | |
| CN109726619A (en) | A kind of convolutional neural networks face identification method and system based on parameter sharing | |
| CN106611156B (en) | Pedestrian identification method and system based on self-adaptive depth space characteristics | |
| WO2019119919A1 (en) | Image recognition method and electronic device | |
| CN112070010B (en) | Pedestrian re-recognition method for enhancing local feature learning by combining multiple-loss dynamic training strategies | |
| CN112464730A (en) | Pedestrian re-identification method based on domain-independent foreground feature learning | |
| CN108320281B (en) | Image significance detection method and terminal based on multi-feature diffusion | |
| CN113177464A (en) | End-to-end multi-modal gait recognition method based on deep learning | |
| CN116311377A (en) | Method and system for re-identifying clothing changing pedestrians based on relationship between images |
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 |