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

Pedestrian re-identification method integrated with position awareness attention

A pedestrian re-recognition and attention technology, applied in the field of computer vision, can solve the problems of excessive parameters and ignoring useless information, so as to achieve good recognition effect and improve expression ability.

Pending Publication Date: 2022-06-24
NANJING UNIV OF POSTS & TELECOMM
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Wang et al. (Wang, Xiaolong, et al.Non-local neural networks. / / Proceedings of the IEEEconference on computer vision and pattern recognition.2018.) proposed to insert the non-local attention module into the network model, and use the attention mechanism to make The model focuses on task-related features and ignores a lot of useless information; Dosovitskiy et al. (Dosovitskiy A, Beyer L, Kolesnikov A, et al.An image isworth16×16words: Transformers for image recognition at sacle[J].arXiv:2010.11929,2020 .) The proposed algorithm (Vision Transformer, ViT) can efficiently complete the image classification task by adding position coding and making full use of the position where the feature appears as prior knowledge to improve the representativeness of the feature; as a typical application of position coding method, the ViT algorithm has been proven to have a significant effect on computer vision tasks, but the position code in ViT is directly added to the input image, and the number of parameters is too large, and the network may encounter difficulties in learning the corresponding 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
  • Pedestrian re-identification method integrated with position awareness attention
  • Pedestrian re-identification method integrated with position awareness attention
  • Pedestrian re-identification method integrated with position awareness attention

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0042] see figure 1 , a pedestrian re-identification method incorporating position-aware attention provided by the present invention mainly includes inputting the output feature map obtained from the original image through the first two layers of the ResNet50 network into the position-aware attention module for processing and converting the position-aware attention The module is integrated into the ResNet50 network for training and testing.

[0043] The output feature map obtained from the original image through the first two layers of the ResNet50 network is input into the position-aware attention module for processing, including:

[0044] S1: Obtain the input feature map, extract three different feature maps through the convolution filter, perform the pooling operation on two of the feature maps to obtain the feature maps φ and g, and the feature map θ remains unchanged; then the above three-dimensional feature Figures θ, φ, and g are flattened and straightened into a two-d...

Embodiment 2

[0051] The inventor found that in the ResNet50 network person re-identification method, the model considers that the importance of each sub-feature in the feature map is the same, and all features need to be considered, resulting in slow training speed and inability to efficiently extract useful tasks. key features. In order to solve the above problems, in a preferred embodiment of the present invention, a non-local attention module is provided, the basic structure of which is as follows figure 2 shown, the specific steps are as follows:

[0052] S1.1 will input feature map X ∈ R b×c×h×w Through three different weight coefficients and the number of output channels is the number of input channels The 1 × 1 convolution filter of the Width and channel number reduction factor;

[0053] S1.2 Select feature maps φ and g from three different feature maps for pooling operation to obtain feature maps and The feature map without pooling operation is denoted as

[0054] S1.3...

Embodiment 3

[0079] The present invention also provides an example for showing a specific experimental process of the method provided by the present invention.

[0080] In this embodiment, three data sets of Market1501, DukeMTMC-ReID, and CUHK03 are used for training and testing. Market1501 was collected from the campus of Tsinghua University in the summer of 2015, including 1,501 pedestrian IDs, and a total of 32,668 images were collected by 6 cameras. The training set contains 751 pedestrian IDs and a total of 12,936 images. The test set contains the remaining 750 IDs, 3368 There are 15,913 images to be retrieved; DukeMTMC-reID was collected from the Duke University campus in the winter of 2015, including 1,812 pedestrian IDs, and a total of 36,411 images. The training set contains 702 pedestrian IDs and a total of 16,522 images. The set contains the remaining 702 pedestrian ID pictures. The CUHK03 dataset contains 14,096 images labeled by humans, and 14,097 images labeled by detection,...

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 provides a pedestrian re-identification method integrated with position awareness attention, which comprises the following steps: introducing a position awareness attention module into a ResNet50 network, the module is an effective improvement of a non-local attention module, and embedding position information into the non-local attention module which captures a long-range feature dependency relationship, so as to identify pedestrian re-identification. And the expression ability of the extracted features is effectively improved. The position awareness attention module provided by the invention belongs to a lightweight structure, and the module is integrated into the ResNet50 network, so that distinguishable features of pedestrians can be effectively extracted, and features with small association degree with a pedestrian recognition task are inhibited at the same time; compared with a traditional network model and other related methods, a better recognition effect is achieved on a plurality of popular pedestrian re-recognition standard data sets.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a pedestrian re-identification method incorporating position-aware attention. Background technique [0002] Person re-identification refers to retrieving a pedestrian image with the same identity as a given query image from a pedestrian image database in a scene with multiple non-overlapping cameras. Pedestrian re-identification can be widely used in intelligent security and video surveillance and other fields. [0003] Person re-identification can be thought of as a feature-embedding problem, ideally the intra-class distance (different pictures of the same person) should be smaller than the inter-class distance (pictures of different people), unfortunately most existing Feature embedding solutions require grouping samples in a pairwise fashion, which is often computationally intensive. In practice, classification methods are often used as feature embedding solutions du...

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 Applications(China)
IPC IPC(8): G06V40/20G06V40/10G06N3/04G06T7/73G06K9/62G06V10/774G06V10/82G06V10/80
CPCG06T7/73G06N3/045G06F18/25G06F18/214
Inventor 吴晓富陈江萍张索非颜俊
Owner NANJING UNIV OF POSTS & TELECOMM
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

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com