Method for re-identifying persons on basis of deep learning encoding models

A coding model and deep learning technology, applied in the field of person re-identification based on the deep learning coding model, can solve problems such as poor effect, high computational complexity of the classifier, and poor quality

Inactive Publication Date: 2017-05-31
张烜
View PDF2 Cites 32 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to propose a person re-identification method based on a deep learning coding model, which effectively solves the problems of poor effect and weak robustness caused by poor monitoring video quality, viewing angle and illumination differe

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
  • Method for re-identifying persons on basis of deep learning encoding models
  • Method for re-identifying persons on basis of deep learning encoding models
  • Method for re-identifying persons on basis of deep learning encoding models

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0056] Embodiment 1: The person re-identification method based on the deep learning coding model in this embodiment has serious quantization errors for vector quantization coding, and sparse coding is only a shallow learning model, which easily leads to the lack of selectivity of visual dictionaries for image features. First, a deep learning network—unsupervised Restricted Boltzmann Machine (RBM) is used to replace the traditional K-Means clustering and sparse coding methods to encode and learn the SIFT feature library to generate a visual dictionary; secondly, according to the learned The dictionary, get the sparse vector corresponding to each SIFT feature, and fuse it to get the deep learning representation vector of the image, and use it to train the SVM classifier; then, use the category label information of the training data to supervise the RBM network learning fine-tuning, and use the SVM classifier to complete the classification and recognition of pedestrians.

[0057]...

Embodiment 2

[0058] Embodiment two: see figure 2 , image 3 The person re-identification method based on the deep learning coding model of this embodiment adopts the following steps to generate a visual dictionary with both sparsity and selectivity:

[0059] First, extract the SIFT features of the training image library; extract the SIFT features; secondly, combine the spatial information of the SIFT features, use the adjacent SIFT features as the input of the RBM, train the RBM through the CD fast algorithm, and obtain the hidden layer features; then the adjacent hidden layer The features are used as the input of the next layer of RBM to get the output dictionary. Among them, ω 1 and ω 2 is the connection weight of RBM. RBM has an obvious layer and a hidden layer, but in RBM, there is no connection between neurons in the same layer, so the learning process is simpler.

[0060] During the training process of the network, the hidden layer and the visible layer of RBM are related by the...

Embodiment 3

[0076] Embodiment three: see Figure 4 , in order to express the image content more accurately in this embodiment, a regular term h(z) is added to the RBM objective optimization function, and the objective function Adjust as follows:

[0077]

[0078] Among them, λ is the weighting coefficient of the regular term. Deep learning coding can make the learned visual dictionary more selective, and make the image expression vector have better sparsity.

[0079] The sparsity and selectivity can be quantitatively analyzed by using the mean value of the visual dictionary’s response to each dimension feature, namely:

[0080]

[0081] in, is the expected value of the average activation probability of each word for K features, word z j for feature x k The expected value of the response probability can be denoted as p jk ∈(0,1), then the expected response value of the entire dictionary to K input features can be written as a matrix Each row element p in the matrix j· repr...

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 relates to a method for re-identifying persons on the basis of deep learning encoding models. The method includes steps of firstly, encoding initial SIFT features in bottom-up modes by the aid of unsupervised RBM (restricted Boltzmann machine) networks to obtain visual dictionaries; secondly, carrying out supervised fine adjustment on integral network parameters in top-down modes; thirdly, carrying out supervised fine adjustment on the initial visual dictionaries by the aid of error back propagation and acquiring new image expression modes, namely, image deep learning representation vectors, of video images; fourthly, training linear SVM (support vector machine) classifiers by the aid of the image deep learning representation vectors so as to classify and identify pedestrians. The method has the advantages that the problems of poor effects and low robustness due to poor surveillance video quality and viewing angle and illumination difference of the traditional technologies for extracting features and the problem of high computational complexity of the traditional classifiers can be effectively solved by the aid of the method; the person target detection accuracy and the feature expression performance can be effectively improved, and the pedestrians in surveillance video can be efficiently identified.

Description

technical field [0001] The invention relates to a person re-identification method based on a deep learning coding model. Background technique [0002] In recent years, with the extensive construction and application of video surveillance systems, it has played an increasingly important role in combating crime and maintaining stability. Most of the current monitoring systems use real-time shooting and manual monitoring, which requires the monitoring personnel to always pay attention to the monitoring screen and carefully distinguish the events in the video, which is obviously unrealistic, not to mention that there are a lot of omissions and subjective errors in the way of manual viewing . Considering the increasing scale of surveillance video, the labor cost required by this method will also be unaffordable and inefficient. Therefore, there is an urgent need for a convenient and quick method to replace the existing manual-led monitoring system. The strong realistic demand ...

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
IPC IPC(8): G06K9/62G06K9/46G06N3/08
CPCG06N3/08G06V10/462G06F18/217G06F18/2411
Inventor 赵永威谭佩耀胡畏李博
Owner 张烜
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap