Research On Person Re-identification Algorithms For Complex Scenes

With the continuous development of surveillance technology and image processing technology,the subject of pedestrian re-identification has received more and more attention.Pedestrian re-identification uses computer-related technology to determine whether the pedestrians captured by different cameras are the same person,and then hope to achieve fast tracking and positioning of important pedestrians.This technology is of great significance and value to social public safety.However,due to the influence of various factors such as illumination changes,complex environments,changing human postures and occlusions,there are still many difficulties to be overcome in pedestrian re-identification.From the perspective of practical application,this paper discusses the two most common complex scenarios in pedestrian re-identification: cross-domain and crossmodal.This paper mainly studies these two scenarios,and the main work is as follows:(1)Cross-domain pedestrian re-identification.In this paper,a pedestrian re-identification framework is designed to improve the generalization ability of the model.The features extracted by the middle layer of the network are used to estimate the image distribution through the clustering algorithm,and the clustering labels are used to guide the distribution alignment of the batch regularization layer.In order to distinguish the features of different distributions,this paper proposes to stack the statistics of layer outputs in multiple networks as features of the estimated distribution.In order to better bridge the data distribution,the batch regularization layer of the framework will normalize the data of different clusters separately.Experimental results on multiple pedestrian datasets verify the effectiveness of our method.(2)Cross-modal pedestrian re-identification.In this paper,a novel crossmodal pedestrian re-identification method is proposed to extract robust features by simultaneously eliminating inter-modal differences for both global and local features.The method employs an attention mechanism to extract local features of the image and uses a novel quadruple loss to align the intermodal differences.The experimental results on RegDB and SYSU-MM01 datasets show that this method improves the performance of cross-modal person re-identification task and outperforms other cross-modal person reidentification methods.