摘 要：一般人流量密度大的场所容易发生行人走失的情况，如：车站、游乐园、广场。针对此类人流量大的密集场所中发生的行人走失情况，提出一种基于 TransReID 的智能寻人系统，结合跨镜技术在场所内对行人目标进行检测搜寻。TransReID 在 Transformer 的基础上做了网络结构层的改进，提高了鲁棒特征提取效率。对比 CNN 网络结构有了很大的提升，TransReID 在对行人目标重识别上也有着更好的综合性能。
关键词：目标检测；重识别；行人；TransReID
DOI:10.19850/j.cnki.2096-4706.2021.14.022
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2021）14-0083-03
The Application of Intelligent Human Search Based on TransReID
PANG Zunyi, LI Dan
(Jincheng College of Sichuan University, Chengdu 610000, China)
Abstract: Lost pedestrians are easy to happen in places with high pedestrian flow density, such as: station, amusement and square. In view of the lost pedestrians situation in such places with high pedestrian flow density, the intelligent human search system based on TransReID is proposed, it combined the cross-mirror technology to search pedestrians targets in the sites. TransReID has improved the network structure layer and the robust feature extraction efficiency on the basis of Transformer. Compared with CNN, the network structure has been greatly improved, TransReID also has better comprehensive performance in re-recognition of pedestrian targets.
Keywords: target detection; re-indentification; pedestrians; TransReID
参考文献：
[1] BEAL J，KIM E，TZENG E，et al. Toward TransformerBased Object Detection [J/OL].arXiv：2012.09958 [cs.CV].（2020-12-17）. https://arxiv.org/abs/2012.09958v1.
[2] HE S T，LUO H，WANG P C，et al. TransReID： Transformer-based Object Re-Identification [J/OL].arXiv：2102.04378 [cs. CV].（2021-02-08）. https://arxiv.org/abs/2102.04378v1.
[3] MENG D C，LI L，LIU X J，et al. Parsing-Based ViewAware Embedding Network for Vehicle Re-Identification [C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR），Seattle：IEEE，2020：7101-7110.
[4] LUO H，GU Y Z，LIAO X Y，et al. Bag of tricks and a strong baseline for deep person re-identifification [J/OL].arXiv： 1903.07071 [cs.CV].（2019-05-19）.https：//arxiv.org/abs/1903.07071v3.
[5] 郝翠翠 . 基于人脸识别的寻人系统设计与实现 [D]. 大连：大连理工大学，2015.
作者简介：庞遵毅（2000.02—），男，汉族，四川内江人，本科在读，研究方向：人工智能。