Designing a Roll Call System with Facial Recognition on Kubeflow

This study is based on the Kubeflow machine learning development platform in order to deploy a real-time roll call system. Kubeflow is based on Kubernetes, which is convenient for container management and portability. Face recognition is done in three steps. First, MTCNN detects a face in the image. Then, FaceNet extracts the features from the face. Finally, SVM finds out the identity of the face closest to the detected face. The average accuracy of the 30 classes in this study is approximately 94.2%, and the execution speed is about 35fps, with Intel Core i7-10700 CPU and NVIDIA GeForce RTX 3060 GPU.