SISTEM TRACKING DOSEN BERBASIS FACE RECOGNITION MENGGUNAKAN YOLOv26 DAN ARCFACE DENGAN FITUR PELACAKAN LOKASI REAL-TIME SERTA NOTIFIKASI OTOMATIS MELALUI BOT TELEGRAM

Andreanto, Rizky Yuli (2026) SISTEM TRACKING DOSEN BERBASIS FACE RECOGNITION MENGGUNAKAN YOLOv26 DAN ARCFACE DENGAN FITUR PELACAKAN LOKASI REAL-TIME SERTA NOTIFIKASI OTOMATIS MELALUI BOT TELEGRAM. S1 thesis, Universitas PGRI MAdiun.

	Text Halaman Depan.pdf Download (918kB)
	Text Abstrak.pdf Download (218kB)
	Text BAB I.pdf Download (226kB)
	Text BAB II.pdf Download (1MB)
	Text BAB III.pdf Restricted to Repository staff only Download (296kB)
	Text BAB IV.pdf Restricted to Repository staff only Download (1MB)
	Text BAB V.pdf Download (213kB)
	Text Daftar Pustaka.pdf Download (208kB)
	Text Lampiran.pdf Restricted to Repository staff only Download (737kB)

Abstract

This study aims to design, implement, and evaluate a Dosen Tracker system based on edge computing and face recognition to provide information on a lecturer’s zone and last detected time. The system was developed using Extreme Programming (XP) and modeled with the Unified Modeling Language (UML). It integrates a Kiosk/Edge Node, an asynchronous FastAPI backend, SQLite, ReactJS, WebSocket, and a Telegram bot. Image processing is performed sequentially using YOLOv26 Nano for face detection, MiniFASNetV2 for liveness checking, and ArcFace Buffalo_SC for identity recognition. Lecturer status is managed using NIP as a persistent identity, a FILO algorithm, and virtual zone transitions to distinguish ENTER and EXIT events. Testing included black-box testing, white-box testing with basis path testing and Pytest, and evaluation of the three artificial intelligence models. Black-box testing results show that the main functions performed as expected in the development environment. A total of 35 white-box test cases covering five core backend logics all passed. YOLOv26 Nano achieved a precision of 89.31%, a recall of 78.94%, and an [email protected] of 87.94%. ArcFace reached an accuracy of 99.88% on 820 image pairs from six identities. MiniFASNetV2 achieved 90.00% accuracy on CelebA-Spoof, but dropped to 66.04% with an ACER of 42.58% on independent online-media-based data. All components were successfully integrated into a complete system; however, operational feasibility in a campus environment still requires direct enrollment, UAT, and field testing on the target devices.

Item Type:	Thesis/Skripsi/Tugas Akhir (S1)
Kata Kunci:	Dosen Tracker, Edge Computing, Pengenalan Wajah, Transisi Zona Virtual, Extreme Programming;Dosen Tracker, Edge Computing, Face Recognition, Virtual Zone Transition, Extreme Programming
Subjects:	T Technology > T Technology (General)
Divisions:	Fakultas Teknik > Teknik Informatika
Depositing User:	ANDREANTO YULI RIZKY
Date Deposited:	05 Aug 2026 06:24
Last Modified:	05 Aug 2026 06:24
URI:	http://eprint.unipma.ac.id/id/eprint/8411

Actions (login required)

View Item