Research On Large-scale Conference Sign-in System Based On Face Clustering Retrieval

In recent years,face recognition technology has been promoted and popularized in various industries such as security.As face recognition technology is applied to more and more scenes,the challenges it faces are also increasing.Firstly,the robustness of the face recognition system is required,that is,the system can still recognize the face under complex conditions such as poor lighting conditions and changes in facial expressions.Secondly,the real-time performance of face recognition system is required,that is,the system is required to recognize the face quickly in the real-time changing video.Thirdly,the ability of the face recognition system to process large-scale face data is required,that is,the system is required to maintain low time-consuming and high accuracy when processing large-scale face data.To improve the above situation,we need to optimize the algorithm or take some measures.This paper makes in-depth research on large-scale face database processing measures in face recognition technology,and adopts the clustering measure to process large-scale face data.At the same time,this paper compares and analyzes different kinds of clustering algorithms,and finally uses density-based clustering algorithms for cluster analysis.The main work of this paper can be summarized as follows:1.This paper studies the application status of face recognition technology,and conducts in-depth research on the large-scale face database processing technology.In view of the current challenges,two strategies including coarse-thin-fine-finish matching and clustering are introduced.At the same time,this paper makes further research on clustering measures,analyzes and compares the advantages and disadvantages of different kinds of clustering algorithms.2.This paper adopts the clustering measures in large-scale face database processing technology to design and implement the face cluster retrieval process.That is to say,the large-scale face database is firstly clustered,and the face is retrieved in the class according to the distance between the class and the image.This can reduce the amount of retrieved data and improve the retrieval speed.In specific implementation,this paper uses a multi-task convolutional neural network with strong robustness,high real-time performance,and support for multi-face recognition for face detection and face alignment.At the same time,this paper uses the Facenet model based on the triple loss function tolearn the encoding of face images in Euclidean space.This model can well reduce the distance between faces within a class and increase the distance between faces of different classes.Finally,based on the mapping results of Facenet model,this paper uses SDBSCAN for clustering analysis,which is a density clustering method based on data sampling.3.Based on the face clustering retrieval process,this article uses Python to build a server based on the Flask framework.At the same time,it uses HTML,CSS,and jQuery for front-end development.In combination with the MongoDB database,a large-scale conference sign-in system is designed and implemented.The system is divided into management backend and sign-in platform.The main functions are designed according to the needs of the three stages: before the meeting,during the meeting,and after the meeting.System functions include pre-meeting image acquisition,sign-in statistics during the meeting,post-meeting file organization,etc.The result shows that the system supports real-time sign-in for multiple people,has good robustness for face detection in complex environments,and can quickly and accurately recognize faces.