基于姿态估计和Transformer模型的遮挡行人重识别

行人重识别是利用人工智能解决边防检查、人员追踪等公共安全应用问题的技术，具有从跨设备采集的图像中识别某一特定行人的能力。但是在人员追踪等问题中，往往会出现行人刻意遮挡、复杂场景环境遮挡等因素，大大提高了行人重识别的难度。针对行人重识别遮挡问题，基于ResNet50网络，结合姿态估计（Pose estimation）和转换器（Transformer）模型，提出了一种改进的行人重识别网络PT-Net，以提高遮挡条件下的行人重识别能力。该方法首先利用现有的姿态估计方法对输入图像进行关键点检测，并将关键点信息与行人特征图像结合起来生成一个基于姿态的行人特征表示；然后利用Transformer模型对基于姿态的行人特征表示编码，用来实现特征对齐和特征融合。论文基于国际公开的数据集Occluded-Duke开展实验验证，结果表明，PT-Net方法相对于基线模型，其均值精度mAP和相似度排序Rank-1指标分别提高了1.3和1.5个百分点，验证了该方法的有效性和优越性。

Person re-identification based on pose estimation and Transformer model

Person re-identification (ReID) is a technology that utilizes artificial intelligence to solve public safety application problems such as border inspection and personnel tracking. It has the ability to identify a specific person from images collected across devices. However, in person tracking and other issues, deliberate person occlusion and complex scene environment occlusion greatly increases the difficulty of person re-identification. An improved person re-identification network PT-Net based on ResNet50 network is proposed, which combined with Pose estimation and Transformer models to improve the person re-identification ability under occlusion conditions. The existing pose estimation method is utilized to detect key-points in the input image, and combines the key-point information with the person feature maps to generate a pose based person feature representation; Then, the Transformer model is used to encode the pose-based person feature representation for feature alignment and fusion. Based on the internationally available dataset Occluded-Duke, the experimental validation is conducted. And the results shows that the PT-Net method improves its mean accuracy mAP and similarity ranking Rank-1 metrics by 1.3 and 1.5 percentage points compared to the baseline model, respectively, verifying the effectiveness and superiority of the method.