视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第1张图片

论文作者:Xuehu Liu,Pingping Zhang,Huchuan Lu

作者单位:Dalian University of Technology;Ningbo Institute

论文链接:https://arxiv.org/pdf/2308.03703v1.pdf

内容简介:

1)方向:视频人员重识别(V-ReID)

2)应用:视频人员重识别

3)背景:V-ReID旨在从非重叠摄像头捕获的原始视频中检索特定的人物。由于人物和场景的变化,高性能仍然面临许多障碍。

4)方法:本文提出了一种名为Long Short-Term Representation Learning (LSTRL)的新型深度学习框架,用于有效的V-ReID。为了提取长期表示,提出了Multi-granularity Appearance Extractor(MAE),可以在多个帧中有效地捕获四个粒度的外观。同时,为了提取短期表示,提出了Bi-direction Motion Estimator(BME),可以从连续的帧中高效地提取互补的运动信息。MAE和BME可以轻松插入现有网络进行高效的特征学习,从而显著提高V-ReID的特征表示能力。

5)结果:在三个广泛使用的基准测试中进行了大量实验证明,所提出的方法可以比大多数最先进的方法提供更好的性能。

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第2张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第3张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第4张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第5张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第6张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第7张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第8张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第9张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第10张图片

视频人员重识别(Video-based Person Re-identification with Long Short-Term Representation Learning)_第11张图片

你可能感兴趣的:(行人冲识别)