Design And Implementation Face Recognition System Based On Deep Learning

Due to its nonintrusive and natural characteristics,face recognition has been the prominent biometric technique for identity authentication and has been widely used in many areas,such as military,finance,public security and daily life,these areas usually require face recognition application to be deployed on servers.With the increasing demand for deploying deep learning algorithms on resource-constrained devices,researching to improve the speed and reduce the loss of accuracy of inference models has became hotspot.Technologies such as lightweight network design and model pruning have simplified the model complexity,and techniques such as model quantization and compression further accelerate model inference speed.These technologies have improved the feasibility of deploying face recognition technology on edge devices.Deep learning has brought artificial intelligence into new era.Deep-learning-based face recognition system is a typical representative.Building a lightweight face recognition system,which mainly used in classroom attendance and event check-in and deployed on personal computer.The system involves face detection algorithm named retinaface and feature extraction algorithm named mobilefacnets with lightweight network backbone.These two algorithms are trainning on large-scale datasets with pytorch deep learning framework and optimize and infer models with Open VINO.When building the system,this system used Qt Designer and Py Qt5 to design and implement user interfaces.After analyzing the system requirements,a face recognition system with modules of face detection,face landmarks detection,face alignment,feature extraction,face registration,face verification,and face search was designed and implemented.The deep learning-based face recognition system read images from the video stream from the camera with a resolution of 640x640.After completing a series of operations,the recognition results are displayed in real time in the user interface.The inference speed on the CPU is improved by nearly 10 times after using Open VINO to optimize and accelerate,compared to inference on deep learning framework directly.