Journal of Computer Applications
Next Articles
Received:
Revised:
Online:
Published:
Contact:
王翠1,邓淼磊2,张德贤3,李磊1,杨晓艳1
通讯作者:
基金资助:
Abstract: Abstract: Person search is one of the important research directions in the field of computer vision. Its research goal is to detect and identify characters in unarmed image libraries. Although there are already a large number of algorithm research on person search algorithms, summary research is still insufficient. In order to deeply understand the person search algorithm, a large number of related literature summarized and analyzed. First of all, according to the different network structure, the person search is divided into two categories: one is a two -step method, and the other is based on the end -to -end step method. The key technologies of the one -step method are analyzed and the characteristic learning and measurement learning Introduction; further introduce the data sets and evaluation indicators in the field of person search, and compare the performance and analysis of the mainstream algorithm; although the experimental results have achieved good performance, most of the methods have a high calculation cost, and It takes a long time, and the one -step method can solve the two sub -tasks in the more efficient learning framework, which is better. Finally, the person search algorithm is summarized and discussed the future development direction.
Key words: person search, one-step, end-to-end, computer vision, transformer
摘要: 行人搜索是计算机视觉领域中重要的研究方向之一,其研究目的是在未剪裁的图像库中检测和识别人物。对于行人搜索算法,尽管已有大量算法研究,但总结性研究尚有不足。为深入了解行人搜索算法,对大量相关文献进行了总结与分析。首先根据网络结构的不同,将行人搜索分为两类:一类是传统的两步法,一类是基于端到端的一步法,对一步法的关键技术:特征学习和度量学习进行重点分析和介绍;进一步介绍了行人搜索领域的数据集和评价指标,对主流算法进行性能比较与分析;实验结果表明,两步法虽然实现了很好的性能,但大多数的方法计算成本很高,且耗时较长,而一步法可以在更高效的学习框架中共同解决两个子任务,效果更好;最后对行人搜索算法进行总结,并讨论了未来的发展方向。
关键词: 行人搜索, 一步法, 端到端, 计算机视觉, transformer
CLC Number:
TP391
王翠 邓淼磊 张德贤 李磊 杨晓艳. 基于图像的端到端行人搜索算法综述[J]. 《计算机应用》唯一官方网站, DOI: 10.11772/j.issn.1001-9081.2023081195.
0 / Recommend
Add to citation manager EndNote|Ris|BibTeX
URL: https://www.joca.cn/EN/10.11772/j.issn.1001-9081.2023081195