Object re-identification apparatus and method thereof

Patent metadata

Field	Value
Publication number	US-12561969-B2
Application number	US-202318134352-A
Country	US
Kind code	B2
Filing date	Apr 13, 2023
Priority date	Aug 26, 2022
Publication date	Feb 24, 2026
Grant date	Feb 24, 2026

Abstract

In an embodiment an apparatus includes a processor configured to generate a feature extraction module using a dataset in which an attribute for each object is defined, receive an image obtained by a camera, extract an attribute of an object of interest from the image using the learned feature extraction module, identify an object re-identification candidate group based on the extracted attribute of the object of interest and re-identify the object of interest based on the identified object re-identification candidate group.

First claim

What is claimed is:

1 . An apparatus comprising:
a processor configured to:
generate a feature extraction module using a dataset in which an attribute for each object is defined;
receive an image obtained by a camera;
extract an attribute of an object of interest from the image using the learned feature extraction module;
identify an object re-identification candidate group based on the extracted attribute of the object of interest; and
re-identify the object of interest based on the identified object re-identification candidate group,
wherein the feature extraction module comprises:
a base network configured to generate a feature map including an overall feature of the object of interest in the image,
a first branch network configured to extract the attribute of the object of interest from the feature map, and
a second branch network configured to extract an object representation of the object of interest from the feature map, and
wherein the first branch network and the second branch network are configured to filter unnecessary information from the feature map using a spatial mask.

2 . The apparatus of claim 1 , wherein the dataset includes an object ID and labeling information in which an attribute of an object is defined for each image.

3 . The apparatus of claim 2 , wherein the attribute is defined as being “complex” when the object has two or more colors, and is defined as being “unknown” when there is no portion of the object in the image or is obscured by another object.

4 . The apparatus of claim 1 , wherein the first branch network is generated using a binary crossentropy loss function.

5 . The apparatus of claim 1 , wherein the second branch network is generated using an ID classification loss function and a triplet loss function.

6 . The apparatus of claim 1 , wherein the processor is configured to:
configure candidate objects registered with an object DB as a tree with respect to an object attribute; determine whether each node condition of the tree is met based on the attribute of the object of interest; and add candidate objects according to the determined result to the object re-identification candidate group.

7 . The apparatus of claim 6 , wherein the processor is configured to add one of a time condition, a distance condition, or a combination thereof to the tree.

8 . The apparatus of claim 1 , wherein the processor is configured to:
calculate a similarity between an object representation included in the object re-identification candidate group and the object representation of the object of interest; and search a candidate object having the highest similarity.

9 . A method comprising:
generating, by a processor, a feature extraction module using a dataset in which an attribute for each object is defined; receiving, by the processor, an image obtained by a camera; extracting, by the processor, an attribute of an object of interest from the image using the learned feature extraction module; identifying, by the processor, an object re-identification candidate group based on the extracted attribute of the object of interest; and re-identifying, by the processor, the object of interest based on the identified object re-identification candidate group, wherein extracting the attribute of the object of interest comprises:
generating, by the processor, a feature map including an overall feature of the object of interest in the image using a base network included in the feature extraction module,
extracting, by the processor, the attribute of the object of interest from the feature map using a first branch network included in the feature extraction module,
extracting, by the processor, an object representation of the object of interest from the feature map using a second branch network included in the feature extraction module, and
wherein the first branch network and the second branch network filter unnecessary information from the feature map using a spatial mask.

10 . The method of claim 9 , wherein the dataset includes an object ID and labeling information in which an attribute of an object is defined for each image.

11 . The method of claim 10 , wherein the attribute is defined as being “complex” when the object has two or more colors, and is defined as being “unknown” when there is no portion of the object in the image or is obscured by another object.

12 . The method of claim 9 , wherein the first branch network is learned using a binary crossentropy loss function.

13 . The method of claim 9 , wherein the second branch network is learned using an ID classification loss function and a triplet loss function.

14 . The method of claim 9 , wherein identifying the object re-identification candidate group comprises:
configuring, by the processor, candidate objects registered with an object DB as a tree with respect to an object attribute; determining, by the processor, whether each node condition of the tree is met based on the attribute of the object of interest; and adding, by the processor, candidate objects according to the determined result to the object re-identification candidate group.

15 . The method of claim 9 , wherein re-identifying the object of interest comprises:
calculating, by the processor, a similarity between an object representation included in the object re-identification candidate group and the object representation of the object of interest; and searching, by the processor, a candidate object having the highest similarity.

16 . An apparatus comprising:
a processor configured to:
generate a feature extraction module using a dataset in which an attribute for each object is defined;
receive an image obtained by a camera;
extract an attribute of an object of interest from the image using the learned feature extraction module;
identify an object re-identification candidate group based on the extracted attribute of the object of interest; and
re-identify the object of interest based on the identified object re-identification candidate group,
wherein the feature extraction module comprises:
a base network configured to generate a feature map including an overall feature of the object of interest in the image,
a first branch network configured to extract the attribute of the object of interest from the feature map, and
a second branch network configured to extract an object representation of the object of interest from the feature map, and
wherein the first branch network is generated using a binary crossentropy loss function, or wherein the second branch network is generated using an ID classification loss function and a triplet loss function.

17 . The apparatus of claim 16 , wherein the dataset includes an object ID and labeling information in which an attribute of an object is defined for each image.

18 . The apparatus of claim 17 , wherein the attribute is defined as being “complex” when the object has two or more colors, and is defined as being “unknown” when there is no portion of the object in the image or is obscured by another object.

19 . The apparatus of claim 16 , wherein the processor is configured to:
configure candidate objects registered with an object DB as a tree with respect to an object attribute; determine whether each node condition of the tree is met based on the attribute of the object of interest; and add candidate objects according to the determined result to the object re-identification candidate group.

20 . The apparatus of claim 16 , wherein the proce

Assignees

• Hyundai Motor Co Ltd
• Kia Corp
• Uif Univ Industry Foundation Yonsei Univ

Inventors

• Jung Hyeon Seok
• Cho Kyung Hwan
• Jin Moon Sub
• Choi Kyu Sang
• Ham Bum Sub
• Eom Chan Ho
• Lee Geon

Classifications

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06V10/761

  Proximity, similarity or dissimilarity measures · CPC title

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title

• G06V10/56

  relating to colour · CPC title

• G06V10/95

  structured as a network, e.g. client-server architectures · CPC title