RGB-Infrared Cross-Modality Person Re-Identification Based On Convolutional Neural Network

RGB-Infrared person re-identification(RGB-IR re-ID)is an important subject to identify specific pedestrians photographed by different surveillance cameras in RGB and IR modes.Under the video surveillance network,RGB-IR re-ID faces greater challenges than re-ID,mainly because it needs to deal with the cross-modality and intra-modality differences at the same time.In order to solve this problem,a new model called AGF is proposed.And on the basis of convolutional neural networks(CNN),attention mechanism including non-local neural networks and squeeze-and-excitation networks are introduced to improve the network performance of RGB-IR re-ID.1.An AGF model is proposed,which consists of two modules:pixel alignment and feature alignment.Firstly,images are preprocessed by converting the images mode,and the way to conversion is Align GAN;Secondly,feature matching is carried out by pairing images,and the pairing method adopts disentangled representation learning;Finally,the ultimate goal of reducing the modal gap and instance alignment is achieved.2.In the aspect of network,the research and analysis are carried out in two aspects.Firstly,the non-local blocks are embedded in ResNet-50 network,and the model can effectively understand the picture globally.Secondly,SE-ResNet-50 network is used to replace the commonly used ResNet-50 network,and the model can automatically learn the importance of different channel features.3.All experiments were conducted on the large public data set SYSU-MM01,and mAP of the three models are 41.2%,41.8% and43.8%,respectively,which are improved in different degrees compared with many related methods.The models proposed in this paper have important significance and practical value for the research and application of RGB-IR re-ID.