基于多粒度特征的行人重识别方法研究

Research on Pedestrian Re-identification Method Based on Multi-granularity Features

由于采集的图像中存在遮挡、图像分辨率低、人姿态发生改变等干扰因素,行人重识别的研究极具挑战性。为此,文章提出基于注意力机制与多粒度特征的行人重识别网络。首先,针对行人姿态的改变,设计了一种多粒度特征提取模块,使用多分支网络联合注意力机制提取多层次全局特征与局部特征。其次,针对行人局部未对齐问题,文章提出了一种邻域自适应特征融合模块。此外,为保留更多的有用信息,文章还设计了一个自适应特征池化模块。在两个公开数据集进行了实验,与其他方法的比较结果验证了所提出方法的有效性。

Due to interference factors such as occlusion,low image resolution,and changes in person poses in the collected images,the research on person re-identification is extremely challenging.To this end,this paper proposes a pedestrian re-identification network based on Attention Mechanism and multi-granularity features.Firstly,in response to the change of pedestrian posture,this paper designs a multi-granularity feature extraction module,which uses a multi-branch network joint Attention Mechanism to extract multi-level global features and local features.Secondly,for the pedestrian local misalignment problem,this paper proposes a neighborhood adaptive feature fusion module.In addition,in order to retain more useful information,this paper also designs an adaptive feature pooling module.It conducts experiments on two public data sets,and the comparison results with other methods verify the effectiveness of the proposed method.