DEVELOPMENT OF FACE RECOGNITION WEB APPLICATION BASED ON FACE DETECTION USING YOLO DEEP LEARNING MODEL AND DLIB LIBRARY

• Main Author: A. F. Poeloengan, Azzahid
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73928
• Institution: Institut Teknologi Bandung

The electronic-based government system (SPBE) is an effort initiated by the Government of Indonesia to create clean and reliable governance of government services and public services. It includes the adoption of technology in each process and is held online. However, public services provided by the government involve sensitive data owned by Indonesian citizens. An identity verification system is needed to guarantee the security and integrity of the system and prevent misuse of the system. An identity verification system using a face (face recognition) is an option that is relatively easy to use. This is because access to desktop devices such as laptops or PCs equipped with cameras is increasingly affordable. In this final project, the author utilizes the pre-trained YOLOv5-Face and YOLOv7-Face detection object models for the face detection subsystem. Then the model is integrated with the Dlib library to classify faces based on their identities. Each model has a weight variation that has a different processing speed and accuracy. Testing is done by determining the confusion matrix for each weight used. The face recognition system with the best performance is implemented as a web application. Based on tests conducted on a dataset of 200 photos with 100 different identities using Google Colab with T4 GPU hardware accelerator, the best results were obtained for the YOLOv7 model with the yolov7-lite-s weight. The results obtained are 94.5% accuracy and 90.09% precision with an average processing time of 237.45 ms for each image.