摘要
随着人们对于公共安全的要求越来越重视,视频监控设备的安装已经变得非常普遍,行人再识别作为针对监控视频中行人进行分析的技术也受到更多人的关注。基于现有的深度学习网络提出了一种以最小化三元组损失为训练目标的非监督行人再识别算法。该设计主要通过预训练模型对数据进行特征提取,然后通过k-means聚类,最后对聚类后的数据进行三元组配对进行网络训练优化。通过在相关数据集上的测试结果可以看出,该设计在处理非标签数据行人再识别方面具有一定的有效性。
Nowadays,installing video surveillance facilities in public areas has become very common.The person re-identification technique for person analysis in surveillance videos has also attracted more attention.This paper proposes an unsupervised person re-recognition algorithm based on the existing deep learning network that aims at minimize triple loss in model training.This design mainly uses the pre-training model to extract features from the raw data,and then uses k-means clustering.Finally,the clustered data is triple-paired for network training optimization.As can be seen from the test results on the relevant data sets,this design has certain effectiveness in processing the non-tag data for person re-identification.
作者
王兴柱
王儒敬
WANG Xing-zhu;WANG Ru-jing(Hefei Institute of Intelligent Machines,Chinese Academy of Science,Hefei 230031,China;University of Science and Technology of China,Hefei 230026,China)
出处
《仪表技术》
2018年第12期19-21,32,共4页
Instrumentation Technology
关键词
行人再识别
三元组
非监督
深度学习
person re-identification
triplet
unsupervised
deep learning