| 基于PCB特征加权的行人重识别算法 |
| |
| 引用本文： | 赵师亮,吴晓富,张索非. 基于PCB特征加权的行人重识别算法[J]. 信号处理, 2020, 36(8): 1300-1307. DOI: 10.16798/j.issn.1003-0530.2020.08.013 |
| |
| 作者姓名： | 赵师亮 吴晓富 张索非 |
| |
| 作者单位： | 南京邮电大学通信与信息工程学院 |
| |
| 基金项目： | 国家自然科学基金(61372123，61701252) |
| |
| 摘 要： | 为充分挖掘行人重识别特征，最近流行的PCB算法给出了一种特征均匀分块并通过RPP网络对齐特征的方法。PCB算法充分发挥了局部特征的作用，有效提高了行人重识别的准确率。为进一步提高行人重识别的性能，本文基于全局特征与局部特征对网络性能的影响差异提出了一种特征加权的PCB行人重识别算法。在典型的行人识别数据库Market1501、DukeMTMC-Reid上的实验结果表明:所提算法具有更好的首中准确率（Rank1）和平均准确率（mAP）；相比与经典的PCB+RPP算法，所提算法在Market1501数据集上Rank1提高了0.8%，mAP提高了4.5%；在DukeMTMC-Reid数据集上Rank1提高了5.5%，mAP提高了约7%。
|
| 关 键 词： | 行人重识别 特征加权 深度神经网络 特征多样性 |
| 收稿时间： | 2020-03-04 |
Weighted PCB for Person Re-Identification |
| |
| Affiliation: | College of Communication and Information Engineering, Nanjing University of Posts and Telecommunication |
| |
| Abstract: | In order to fully exploit the features of any input image for person re-identification (Re-ID), part-based convolutional baseline (PCB) was proposed to employ a uniform partition for any input image and a refined part pooling (RPP) method is followed for enhanced within-part consistency. In order to further improve the performance of Person Re-ID, this paper proposes a weighted PCB algorithm, which combines the global feature and local part-based features in a weighted form. Experiments show that the proposed algorithm is better than other weighted methods. Experiments over Market1501 and DukeMTMC-Reid show that the proposed algorithm can achieve better performance in both the Rank1 accuracy and the mean average accuracy (mAP). Compared with the PCB+RPP algorithm, the proposed algorithm provides a margin of 0.8% and 4.5% over Market1501 for Rank1 and mAP, respectively. For the dataset of DukeMTMC-Reid, it improves PCB by 5.5% in Rank1 accuracy and by about 7% in mAP. |
| |
| Keywords: | |
|
| 点击此处可从《信号处理》浏览原始摘要信息 |
|
点击此处可从《信号处理》下载免费的PDF全文 |
|
