Person re-identification method, system and device, and computer-readable storage medium

Patent metadata

Field	Value
Publication number	US-12260675-B2
Application number	US-202218718411-A
Country	US
Kind code	B2
Filing date	Sep 19, 2022
Priority date	Apr 30, 2022
Publication date	Mar 25, 2025
Grant date	Mar 25, 2025

Abstract

The present application discloses a person re-identification method, system and device, and a computer-readable storage medium. The method includes: obtaining a first type of person image without a label; making label information for the first type of person image; training a target person re-identification network on the basis of the first type of person image and the label information to obtain a first trained target person re-identification network; extracting a region of interest in the first type of person image; and training, on the basis of the first type of person image and the region of interest, the first trained target person re-identification network to obtain a second trained target person re-identification network, and performing person re-identification on the basis of the target person re-identification network.

First claim

The invention claimed is:

1. A person re-identification method, comprising:
obtaining a first type of person image without a label;
making label information for the first type of person image;
training a target person re-identification network on a basis of the first type of person image and the label information to obtain a first trained target person re-identification network;
extracting a region of interest in the first type of person image; and
training, on a basis of the first type of person image and the region of interest, the first trained target person re-identification network to obtain a second trained target person re-identification network, and performing person re-identification on a basis of the target person re-identification network;
wherein the making label information for the first type of person image comprises:
determining body part boundary information in the first type of person image; and
taking the body part boundary information as the label information of the first type of person image;

wherein the extracting a region of interest in the first type of person image comprises:
extracting a feature map previous to an average pooling layer of the target person re-identification network;
superimposing all channels of the feature map to obtain Class Activation Mapping (CAM);
determining a real-time threshold value of the CAM on a basis of a hyper-parameter, and taking a pixel region in the first type of person image corresponding to the real-time threshold value greater than a preset threshold value as an initial region of interest;
searching a connected component in the initial region of interest, and determining the region of interest on a basis of the connected component; and
determining bounding box information of the region of interest.

2. The method according to claim 1 , wherein the determining body part boundary information in the first type of person image comprises:
determining the body part boundary information in the first type of person image on a basis of a template matching method.

3. The method according to claim 2 , wherein the determining the body part boundary information in the first type of person image on a basis of a template matching method comprises:
obtaining a preset human body part template;
determining a body part region corresponding to the preset human body part template in the first type of person image;
determining boundary coordinates of the body part region, the boundary coordinates comprising a height value of a boundary of the body part region in the first type of person image; and
taking a ratio of the boundary coordinates to a total height of the first type of person image as the body part boundary information.

4. The method according to claim 3 , wherein
the preset human body part template comprises a head template, a torso template, and a lower limb template;
the body part region comprises a head region, a torso region, and a lower limb region; and
the body part boundary information comprises starting boundary information of the head region, boundary information between the head region and the torso region, boundary information between the torso region and the lower limb region, and ending boundary information of the lower limb region.

5. The method according to claim 4 , wherein the determining a body part region corresponding to the preset human body part template in the first type of person image comprises:
segmenting temporary images with a same size as the preset human body part template from the first type of person image;
calculating a similarity value of each of the temporary images and the preset human body part template; and
selecting a temporary image with a maximum similarity value from the temporary images as the body part region corresponding to the preset human body part template in the first type of person image.

6. The method according to claim 5 , wherein the calculating a similarity value of each of the temporary images and the preset human body part template comprises:
calculating the similarity value of each of the temporary images and the preset human body part template on a basis of a similarity calculation formula;
the similarity calculation formula comprises:

c
=

∑

m
=
1

w

∑

n
=
1

h

S
ij

(

m
,
n

)

×

T
⁡
(

m
,
n

)

∑

m
=
1

w

∑

n
=
1

h

[

S
ij

(

m
,
n

)

]

2

⁢

∑

m
=
1

w

∑

n
=
1

Assignees

• Suzhou Metabrain Intelligent Technology Co Ltd

Inventors

• Zhao Yaqian
• Wang Li
• Fan Baoyu

Classifications

• G06T2207/20081

  Training; Learning · CPC title

• G06V10/751

  Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching · CPC title

• G06V10/771

  Feature selection, e.g. selecting representative features from a multi-dimensional feature space · CPC title

• G06V20/70 Primary

  Labelling scene content, e.g. deriving syntactic or semantic representations · CPC title

• G06V10/25

  Determination of region of interest [ROI] or a volume of interest [VOI] · CPC title