Research On Cross-modal Person Re-identification Based On Deep Learning

Person re-identification,as an important research direction in the area of intelligent security,has gained the focus of researchers and scholars.In practical scenarios,visible light cameras are highly dependent on lighting conditions and have limited detection capability in poor light.Therefore,many scholars have gradually shifted their research goals to the field of cross-modality person re-identification,but there are few relevant studies and challenges remained in resolving the differences in the images of different modalities.To address these problems,this paper uses two approaches of deep learning networks to reduce the discrepancy between the two modalities separately and to guide network in a more reasonable direction,thus improving the identification performance of the network,with the following main work.(1)To address the problems that simple pooling methods cannot obtain detailed information about pedestrians and that traditional loss functions fail to reduce the variability between heterogeneous modalities,resulting in poor model accuracy,a crossmodality person re-identification method based on multi-granularity pooling is presented.Using a combination of global average pooling and global max pooling,a new module of global multi-granularity pooling is designed to obtain more pedestrian information.Meanwhile,a new cross-modality triplet loss function is designed to achieve a better reduction of the difference between the two modalities.(2)A cross-modality person re-identification method based on the attention mechanism is proposed to address the problem that the use of the attention mechanism method can improve training efficiency but at the same time is prone to unstable model training.A new attention mechanism module is designed to allow the network to use less time to focus on more key features of person.In addition,a cross-modality hard center triplet loss is designed to supervise the model training better.The paper has conducted extensive experiments on the above two methods on two publicly available datasets,which obtained better performance than current similar methods and verified the effectiveness and feasibility of the proposed methods in this paper.Furthermore,to facilitate the application exploitation of cross-modality person reidentification,a cross-modality person re-identification system is designed,which integrates two methods proposed in this paper not only improves the availability of algorithms,but also realizes the functional visual analysis of the above two algorithms.