Design and Evaluation of a Face Verification Pipeline for Real-Time Online Assessment: A Case Study with InsightFace Models

Tóm tắt

The rapid development of online assessment systems has created a growing demand for reliable identity verification mechanisms to ensure fairness and trustworthiness of evaluation results. This study presents the design and evaluation of a real-time face verification pipeline for online assessment systems, implemented and validated through the real-time online assessment (PretestBooth) system. The proposed pipeline adopts an embedding-based approach, consisting of face preprocessing and feature matching in the embedding space using cosine similarity with an optimized decision threshold. System performance is evaluated using standard biometric metrics, including False Acceptance Rate (FAR) and False Rejection Rate (FRR), to simultaneously reflect security and reliability aspects of the verification process. The system leverages the InsightFace framework with two models, Buffalo_S and Buffalo_L, to analyze the trade-off between performance and computational cost under the same deployment conditions. Experimental results demonstrate that both models achieve high verification accuracy. Buffalo_S shows a clear advantage in terms of lower latency and faster inference speed, making it more suitable for real-time systems, while Buffalo_L exhibits better stability under varying data conditions. These findings highlight the trade-off between speed and stability and confirm the feasibility of deploying face verification pipelines in real-time online assessment systems.