Volume 37, Number 5, October 2019
|Page(s)||1070 - 1076|
|Published online||14 January 2020|
Person Re-Identification Net of Spindle Net Fusing Facial Feature
School of Computer Science and Technology, Changchun University of Science and Technology, Changchun 130022, China
2 School of Artificial Intelligence, Changchun University of Science and Technology, Changchun 130022, China
In the field of person re-identification, the extraction of pedestrian features is mainly focused on the extraction of features from the whole pedestrian or limb torso, and the facial features are less used. The facial features is integrated into the network to enhance pedestrian recognition accuracy rate. By introducing the MTCNN facial extraction network in the framework of person re-identification network Spindle Net, and improves the accuracy of person re-identification by improving the weight of facial features in the overall pedestrian characteristics. The experimental results show that the accuracy of Rank-1 on the CUHK01, CUHK03, VIPeR, PRID, i-LIDS, and 3DPeS data sets is 7% higher than that of Spindle Net.
目前在行人重识别 (person re-identification) 领域对行人特征的提取主要集中在整体行人或肢体躯干分别提取特征 较少使用面部特征。将面部特征融入到网络中以提高行人重识别的准确率。在行人重识别网络 Spindle Net 的框架中引入 MTCNN面部提取网络，通过提高面部特征在整体行人特征中的权重来提高行人重识别的准确率。实验结果表明，文中提出的网络相比于 Spindle Net 在CUHK01，CUHK03，VIPeR，PRID，i-LIDS，3DPeS数据集上Rank-1的准确率平均提升7%。
Key words: person re-identification / facial / convolutional neural network
关键字 : 行人重识别 / 面部 / 卷积神经网络
© 2019 Journal of Northwestern Polytechnical University. All rights reserved.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0 ), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.