【摘要】 行人重识别技术（person Re-Identification,Re-ID）旨在从多个不同的摄像头中检索与查询图像相同身份的行人图像。在行人重识别中，不同的行人可能具有相似的外观，而且受到摄像机视角、行人姿势等因素的影响，行人重识别可能存在着较为严重的遮挡问题。为了解决这些问题，我们提出了文本辅助图像信息的行人重识别算法，构建图像通道和文本通道，分别使用OSNet (Omni-Scale Network)网络和SRU (Simple Recurrent Units)网络来提取图像特征和文本特征，其中图像特征分为全局特征和局部特征。文本信息可以准确地描述一个人，对遮挡问题有较高的鲁棒性，全局特征关注行人的整体信息，局部特征提供细粒度的细节信息，因此用文本信息辅以图像信息，可以提高网络模型的鲁棒性。在Market1501和CUHK03两个数据集上进行的实验可以证明本方法的有效性。更多还原

【Abstract】 The person re-identification(Re-ID) technology aims to retrieve pedestrian images with the same identity as the query image from multiple different cameras. In Re-ID, different pedestrians may have similar appearance. And affected by camera viewpoints, pedestrian pose and other factors, person re-identification may face serious occlusion problems. In order to solve these problems, we propose a Re-ID algorithm for text assisted image information, construct image channels and text channels,and use OSNet(Omni-Scale Network) network and SRU(Simple Recurrent Units) network to extract image features and text features, respectively. Image features are divided into global features and local features. Text information can accurately describe a person, and has high robustness to occlusion problems. Global features focus on the overall information of pedestrians, and local features provide fine-grained details. Therefore, image information supplemented with text information can improve the robustness of the network model. Experiments on Market1501 and CUHK03 datasets can prove the effectiveness of this method.更多还原