Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A gait recognition system and method based on deep learning of self-attention mechanism

A technology of gait recognition and deep learning, which is applied in the field of gait recognition system of deep learning, can solve problems such as insufficient expressive ability and single information, and achieve the effect of enhancing expressive ability and reducing noise

Active Publication Date: 2022-05-20
HUAZHONG UNIV OF SCI & TECH
View PDF2 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] Aiming at the technical problems of the prior art gait recognition method based on deep learning with single feature information and insufficient expressive ability, the present invention provides a gait recognition system and method based on self-attention mechanism for deep learning. Attention mechanism, classification comparison and verification combination loss, and prior knowledge enhance deep features

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A gait recognition system and method based on deep learning of self-attention mechanism
  • A gait recognition system and method based on deep learning of self-attention mechanism
  • A gait recognition system and method based on deep learning of self-attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0074]The network model is trained using the stochastic gradient descent method, and the parameters are set as follows: batchsize is 64, baselearning rate is 0.01, weight decay is set to 0.0005, and the learning rate is reduced by 10 times every 10k times. Does not set any form of data augmentation, including cropping, flipping, etc. Initialize with the weights of a well-learned basic network, and the new layer is relearned and trained with fine-tuning.

[0075] The training specifically includes the following sub-steps:

[0076] (1) Each batch of training includes 64 images;

[0077] (2) Using the pre-trained model as the initial parameter of the network, after inputting the training samples into the network, forward propagation calculates the values ​​of each layer of the network;

[0078] (3) If the predetermined total number of iterations is not reached, then continue to step (4), otherwise end the training; wherein, the predetermined total number of iterations is 35000;...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a gait recognition system and method based on deep learning of a self-attention mechanism, belonging to the field of gait recognition. The present invention proposes to use the attention mechanism on the original feature map. By learning a matrix between 0 and 1 with the same size as the original feature, the original feature is denoised, and the salient features in the picture are selected to reduce the noise in the feature map. The noise of the classification loss and the comparison verification loss are organically combined to punish the features in combination with the loss function, which not only uses the identity information of the target, but also uses the different relationships between the targets to increase the discrimination between different features; The most important limb features in gait are added to the original depth features as prior knowledge. This combination can not only use the target global body shape features, but also correct the features that are not conducive to classification brought about by the clothes transformation learned by the global features, and enhance the original depth features. The expressive ability of deep features enhances the expressive ability of features from different dimensions.

Description

technical field [0001] The invention belongs to the field of gait recognition, and more specifically relates to a gait recognition system and method based on deep learning of a self-attention mechanism. Background technique [0002] Gait recognition refers to the identification and verification of a person's identity through their posture while walking. Gait is an effective dynamic biometric feature. As a unique feature of a person, it does not require the cooperation of the target in the recognition and does not require high distance. In addition, the gait will still not change with the change of appearance. . Therefore, the application prospect of gait in the future identity authentication is very considerable. [0003] The current way of representing gait is basically a grayscale image without background. The gait recognition method has eliminated the influence of color, background and long-term relative recognition long ago. Gait recognition is mainly affected by two...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06V10/764G06V40/20G06K9/62G06N3/08
CPCG06N3/08G06V40/25G06F18/2413
Inventor 凌贺飞吴佳李平
Owner HUAZHONG UNIV OF SCI & TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products