Implementation of YOLOv8n-Based Liveness Detection to Enhance the Security of Facial Attendance Systems Against Real-Time Spoofing Attacks

Abstract : This research aims to develop an anti-spoofing security system based on liveness detection to address the vulnerabilities of facial attendance systems against identity manipulation using photos or videos. The methodology employed is quantitative experimental, utilizing the YOLOv8n (You Only Look Once) deep learning architecture. The research stages involved the independent collection of a dataset comprising 8,000 images, categorized into “real” and “fake” face classes. This dataset was processed using a distribution ratio of 70% for training, 20% for validation, and 10% for testing. The model training process was conducted over 50 epochs using the Google Colab platform supported by a Tesla T4 Graphics Processing Unit (GPU). The results indicate that the developed model achieves high performance, with a mean Average Precision (mAP50) of 99.4%, precision of 98.8%, and recall of 99.0%. Real-time testing demonstrated an average inference speed of 20 frames per second (FPS), ensuring system responsiveness in practical applications. Implementation on low-power hardware, specifically the NVIDIA Jetson Nano, confirmed the computational efficiency of the model for edge device deployment. The primary conclusion of this study is that the YOLOv8n-based liveness detection system is highly effective and viable for integration into automated organizational attendance systems due to its high accuracy and efficient power consumption.