Processing

Please wait...

Settings

Settings

Goto Application

1. WO2022001137 - PEDESTRIAN RE-IDENTIFICATION METHOD, APPARATUS AND DEVICE AND MEDIUM

Publication Number WO/2022/001137
Publication Date 06.01.2022
International Application No. PCT/CN2021/076976
International Filing Date 20.02.2021
IPC
G06K 9/00 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
G06K 9/62 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
CPC
G06K 9/6223
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
6218Clustering techniques
622Non-hierarchical partitioning techniques
6221based on statistics
6223with a fixed number of clusters, e.g. K-means clustering
G06K 9/6256
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
6256Obtaining sets of training patterns; Bootstrap methods, e.g. bagging, boosting
G06V 40/10
Applicants
  • 浪潮电子信息产业股份有限公司 INSPUR ELECTRONIC INFORMATION INDUSTRY CO., LTD. [CN]/[CN]
Inventors
  • 金良 JIN, Liang
  • 尹云峰 YIN, Yunfeng
  • 范宝余 FAN, Baoyu
  • 张润泽 ZHANG, Runze
  • 郭振华 GUO, Zhenhua
  • 姜金哲 JIANG, Jinzhe
  • 梁玲燕 LIANG, Lingyan
Agents
  • 北京集佳知识产权代理有限公司 UNITALEN ATTORNEYS AT LAW
Priority Data
202010605966.329.06.2020CN
Publication Language Chinese (zh)
Filing Language Chinese (ZH)
Designated States
Title
(EN) PEDESTRIAN RE-IDENTIFICATION METHOD, APPARATUS AND DEVICE AND MEDIUM
(FR) PROCÉDÉ, APPAREIL ET DISPOSITIF DE RÉIDENTIFICATION DE PIÉTON, ET SUPPORT
(ZH) 一种行人重识别方法、装置、设备及介质
Abstract
(EN) A pedestrian re-identification method, apparatus and device and a medium. The method comprises: extracting features of an original training set by using a first pedestrian re-identification model, wherein the original training set comprises pedestrian sample images and corresponding label information; performing clustering according to feature spatial distribution characteristics of an original data set; screening out difficult samples according to a clustering result; adding the difficult samples to the original training set to obtain a target training set; training the first pedestrian re-identification model by using the target training set to obtain a second pedestrian re-identification model; and when a pedestrian image to be identified is acquired, outputting a corresponding identification result by using the second pedestrian re-identification model. In this way, the difficult samples are mined, and the sample spatial distribution in the original data set is changed by means of the mined difficult samples, and the attention to the difficult samples can be increased, thereby improving the accuracy of pedestrian re-identification.
(FR) L'invention concerne un procédé, un appareil et un dispositif de réidentification de piéton, et un support. Le procédé comprend les étapes consistant à : extraire des attributs d'un ensemble d'entraînement d'origine à l'aide d'un premier modèle de réidentification de piéton, l'ensemble d'entraînement d'origine comprenant des images d'échantillon de piéton et des informations d'étiquette correspondantes ; effectuer un regroupement en fonction de caractéristiques de distribution spatiale d'attributs d'un ensemble de données d'origine ; cribler des échantillons difficiles selon un résultat de regroupement ; ajouter les échantillons difficiles à l'ensemble d'entraînement d'origine pour obtenir un ensemble d'entraînement cible ; entraîner le premier modèle de réidentification de piéton en utilisant l'ensemble d'entraînement cible pour obtenir un second modèle de réidentification de piéton ; et lorsqu'une image de piéton à identifier est acquise, délivrer en sortie un résultat d'identification correspondant en utilisant le second modèle de réidentification de piéton. De cette manière, les échantillons difficiles sont analysés, et la distribution spatiale d'échantillon dans l'ensemble de données d'origine est modifiée au moyen des échantillons difficiles analysés, et l'attention portée aux échantillons difficiles peut être augmentée, ce qui permet d'améliorer la précision de la réidentification des piétons.
(ZH) 一种行人重识别方法、装置、设备及介质,包括:利用第一行人重识别模型提取原始训练集的特征;其中,所述原始训练集包括行人样本图像和对应的标签信息;根据所述原始数据集的特征空间分布特性,进行聚类;根据聚类结果筛选出困难样本;将所述困难样本添加至所述原始训练集,得到目标训练集;利用所述目标训练集对所述第一行人重识别模型进行训练,得到第二行人重识别模型;当获取到待识别行人图像,则利用所述第二行人重识别模型输出对应的识别结果。这样,挖掘出困难样本,通过挖掘出的困难样本改变原始数据集中的样本空间分布,能够增加对困难样本的关注度,从而提升行人重识别的准确度。
Related patent documents
Latest bibliographic data on file with the International Bureau