Research On Person Re-identification Method Based On Deep Feature Fusion Network

In recent years,social security construction has become the hot topic that people pay most attention to,and security cameras are widely popularized in public areas.Facing massive monitoring data,it is already difficult to obtain,analyze,and organize the required information manually.Person re-identification is a technology for searching for specified pedestrians in cross-cameras,and it is receiving more and more attention in the fields of video content retrieval,video surveillance,and intelligent security.However,in actual application scenarios,background interference,arbitrary pedestrian poses,and uncontrollable camera angles,etc.,will bring great obstacles to person reidentification research.Therefore,the most important and challenging task in person re-identification tasks is how to accurately extract pedestrian features with high discriminative performance from pedestrian pictures.At the same time,the rapid development of deep learning provides an efficient research approach for computer vision.Compared with the previous manual design of features,the features extracted through deep convolutional neural networks are more robust.Therefore,it is feasible to apply deep learning to solve the problem of pedestrian re-identification.Based on the deep convolutional network,this paper proposes two person re-identification network models,which effectively improves the performance of pedestrian re-identification.The main research contents are as follows:(1)Aiming at the difficulty of extracting discriminatory pedestrian features,this paper proposes a multi-division attention network(MDA).By designing different branches,the network simultaneously learns global and local features,and then performs feature fusion to obtain the final pedestrian feature representation.Among them,the extraction of pedestrian local features adopts the method of horizontal segmentation,and the attention network is embedded in the local branch to enhance the local features.Attention network is composed of spatial attention network and channel attention network,which can effectively complement the characteristic information of pedestrians interested in learning and improve the performance of person reidentification network.(2)Aiming at the problem that pedestrians with different poses are difficult to extract local features of pedestrians,this paper proposes a pose attention network(PAN).This network is different from direct horizontal segmentation when extracting local features.By designing a local sub-network,the key points of the pedestrian skeleton are located to obtain local pedestrian areas.In order to more accurately divide the local area,the local division sub-network divides the local area into a rigid body part and a flexible body part according to the inherent properties of the body,so that it can extract the part of the pedestrian image well with only a small amount of a priori feature.At the same time,an attention mechanism is added to find prominent areas of different parts of the pedestrian's body and extract pedestrian features with more recognition capabilities.