Graph Convolutional Network Based on Transformer(Tran-GCN) Model for Improving Monitoring Vidwo Person Re-Identification

With the rapid development of society and economy, video monitoring networks are widely used in many fields. In real-world scenarios, camera installation is limited by height and density, and most of the images captured in videos are incomplete persons. In complex scenes with high person flow density, it is difficult to accurately identify, track, and detect a certain person. The person Re-Identification method is mainly divided into two stages: the first stage is to detect all persons in the monitoring video, and the second stage is to compare persons in numerous videos based on a certain target object, in order to find the same character. This study proposes a Transformer graph based Convolutional neural network (Tran-GCN) model to improve person recognition in monitoring video. The Tran-GCN model includes two parts: First, design a person graph convolutional network learning model. Firstly, the basic CNN is used to extract the global appearance features of persons. Then, through the OpenPose human pose estimation network, the rich pose information of persons is fully combined with the inherent skeletal structure information of persons to obtain key point information of persons. Through the graph convolutional network, the relationship information of person joint structure is integrated to obtain the relationship features between nodes. Secondly, design a person Re-Identification model for the Transformer learning branch, learning more fine-grained semantic local feature relationships of persons, so that the model focuses on more important local regions. Finally, through experimental verification, more accurate discriminative person features were captured to re identify personnel in monitoring videos, improving recognition accuracy and effectiveness.