利用Dlib深度殘差學習實作人臉辨識──以google meet線上會議點名為例

Face Recognition Using Dlib Deep Residual Learning - Take Google Online Meeting As an Example

本文利用現今人工智慧的技術，實作自動執行人臉辨識與比對程式，並將此程式實際應用於疫情時代大量使用的線上教學系統中，以快速得到線上教學參與人數之數量及參與對象的身分識別。經過一連串程式設計的過程，實作成果如下：一、利用OpenCV 的CascadeClassifier 分類器來搜尋影像中的人臉時，使用的是Haar 特徵產生的人臉分類器，針對影像中過於歪斜的人臉，這種方法還是偵測不到的。二、Dlib 的人臉搜尋主要是利用HOG 特徵實作，這種方法準確度較高，當找出了人臉位置後，Dlib 進而使用Ensemble of Regression Trees 理論來預測每一個人臉上的指定特徵位置。當有了所有人臉的特徵位置資料，經過人臉對正程序，最後使用深度殘差學習模型計算每張人臉的128D 向量資訊而能確實使用歐式距離的計算進行人臉比對。在我們的實作中，人臉辨識的準確度可以百分之百的達成。

This article uses today's artificial intelligencetechnology to implement an automatic face recognitionand comparison program, and practically applies thisprogram to the online teaching system that is widelyused in the epidemic era, so as to quickly obtain thenumber of online teaching participants and the number ofparticipants. After a series of programming process, theimplementation results are as follows:1. When using OpenCV's CascadeClassifier tosearch for faces in the image, the face classifier generatedby Haar features is used. For faces that are too skewed inthe image, this method is still unable to detect.2. Dlib's face search is mainly implemented usingHOG features. This method has high accuracy. When theface position is found, Dlib then uses the Ensemble ofRegression Trees theory to predict the specified featureposition on each face. When the feature position data ofall faces is available, after the face alignment procedure,the deep residual learning model is used to calculate the128D vector information of each face, and the Euclideandistance calculation can be used for face comparison. Inour implementation, the accuracy of face recognition canbe achieved 100%.