FACE RECOGNITION-BASED ATTENDANCE SYSTEM USING THE ARCFACE MODEL WITH VARIATIONS IN LIGHTING CONDITIONS, RESOLUTIONS, AND POSE IN THE DATASET

• Main Author: Gunawan, William
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85285
• Institution: Institut Teknologi Bandung

One of the key factors that can affect the performance of a face recognition attendance system is the variation in the conditions under which face dataset images are captured, such as different lighting, changes in pose, and image resolution. This final project aims to design a face recognition system for attendance purposes. The process involves collecting a face training dataset with varying conditions using a pre-designed setup and evaluating the performance of the ArcFace and FaceNet face recognition models. The results show that the ArcFace model outperforms FaceNet in terms of precision, recall, and F1-score. After evaluating the models, a deployment phase was conducted to simulate the face recognition attendance system in a real-world scenario. The training dataset used for deployment includes several types of variations in lighting conditions, image resolution, and pose. The deployment results indicate that the training dataset with 690 lux lighting, 1280 x 720 pixels resolution, and pose variation for ArcFace model training demonstrated the most reliable performance for attendance compared to other face training datasets. Additionally, it was concluded that dim lighting conditions in the training dataset can affect the face recognition results of the ArcFace model, pose variation can help improve the performance of the ArcFace model by reducing the number of false recognitions, and variations in image resolution conditions do not significantly affect the performance of the ArcFace model.