Research And Application Of Person Re-Identification Technology Based On Deep Learning

Person re-identification is a technique that uses computer vision techniques to determine whether a particular pedestrian exists in an image or video sequence.Since the application scene is more common in surveillance video,video person re-identification has gradually become a research hotspot.Due to the different poses of pedestrians,illumination and changes in perspective,the same pedestrian image taken by different cameras often has large differences in visual appearance characteristics,which leads to mismatching problems.The contribution of this paper can be summarized as follows:In this paper,a new pedestrian re-identification algorithm based on deep learning is proposed for the pedestrian re-identification database with low resolution and too similar visual characteristics of pedestrians.1.This paper proposed to use the Generative Adversarial Networks to enhance the image at the pixel level.2.The paper proposes a classification network which introduces semantic information so that the network can identify more representative pedestrians feature.At the same time,this paper uses the attention model to dynamically assign different weights to different parts of pedestrians,and automatically highlights the most resolving features of pedestrians.3.This paper proposes to use the combination of traditional cross entropy loss function and the triplet loss to maximize the inter-class distance of the pedestrian image,minimize the intra-class distance of the pedestrian image,and maintain the constraints of the overall classification.4.Experiments from two challenging person re-identification datasets iLIDS-VID and PRID-2011 show that the proposed algorithm significantly improves the recognition rate of pedestrian re-identification.5.In this paper,the proposed model and the benchmark model(removing all the models proposed in this paper)are visualized in the dimension reduction of the features extracted from the same pedestrian image.The results of the visualization strongly prove that the proposed algorithm can reduce the difference between cameras.