摘要
针对行人重识别中行人检测误差引起的空间错位,基于局部的深度网络模型仅学习相邻局部关系,导致远距离局部相关性缺失,因此,提出了一种结合一阶和二阶空间信息的行人重识别算法。在主干网络上,学习一阶空间掩模对输入图像的空间权值进行微调,以减少背景干扰;通过二阶空间掩模对远距离的依赖关系进行建模,并将局部特征集成到依赖模型中,以获取全局特征表示。局部分支引入DropBlock对抽取的行人特征进行正则化,避免了网络模型过于依赖特定部位特征。训练阶段用标签平滑分类损失和引入正样本中心的三元组损失联合优化整个网络。在Market-1501和DukeMTMC-reID数据集上的实验结果表明,相比其他主流算法,本算法的行人重识别精度更高,且提取的行人特征判别性和鲁棒性更好。
In order to solve the problem of the spatial dislocation caused by person detection error in person re-identification,the local-based deep neural networks model only learn the adjacent local relationship,resulting in lack of long-distance local correlation.This paper proposes a person re-identification algorithm based on first-order and second-order spatial information.On the backbone network,first-order spatial mask is learned to fine-tune the spatial weight of the input image to reduce the background interference.The second-order spatial mask is used to model the long-distance dependency relationship,and local features are integrated into the dependency model to obtain the global feature representation.In the local branch,DropBlock is introduced to regularize the pedestrian features to avoid the network model relying too much on specific part features.In the training stage,the whole network is optimized by the label-smoothed cross-entropy loss and the triple loss with positive samples’center.Experimental results based on Market-1501 and DukeMTMC-reID data sets show that compared with other mainstream algorithms,the person re-identification accuracy of the algorithm is higher,and the extracted pedestrian features are more discriminative and robust.
作者
刘莎
党建武
王松
王阳萍
Liu Sha;Dang Jianwu;Wang Song;Wang Yangping(School of Electronics and Information Engineering,Lanzhou Jiaotong University,Lanzhou,Gansu 730070,China;Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphic&Image Processing,Lanzhou,Gansu 730070,China;National Experimental Teaching Demonstration Center on Computer Science and Technology,Lanzhou Jiaotong University,Lanzhou,Gansu 730070,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2021年第2期299-307,共9页
Laser & Optoelectronics Progress
基金
甘肃省科技计划项目(18JR3RA104)
甘肃省教育厅科技项目(2017D-08)。
关键词
机器视觉
行人重识别
一阶空间掩模
二阶空间掩模
中心三元组损失
machine vision
person re-identification
first-order spatial mask
second-order spatial mask
center triplet loss
