Skripsi

DETEKSI DAN PENGENALAN WAJAH UNTUK SISTEM KEAMANAN MENGGUNAKAN CCTV BERBABIS DEEP LEARNING

---

Penilaian

---

Currently, security systems can be implemented for surveillance in rooms that are only accessible to authorized or relevant personnel. However, the methods applied in Closed-Circuit Television (CCTV)-based security systems still cannot detect and recognize faces from a sufficient distance and have not been directly implemented. Therefore, this research aims to develop a room security system that can identify and implement the use of faces as input from a remote distance using CCTV cameras. The research involves several main stages, namely video capture from a distance of 1-3 meters, image processing, and training facial image data. This study employs YOLOv5 and YOLOv8 algorithms using pre-trained models with variations of sizes M and X for each algorithm, used for face detection and recognition. The training is conducted with 200 epochs and 32 batch sizes for each model, resulting in training mAP (mean average precision) scores for YOLOv5m, YOLOv5x, YOLOv8m, and YOLOv8x as 82.7%, 83%, 85%, and 85.2% respectively. The offline testing results show that all classes can be correctly recognized with average accuracy values of 94%, 95%, 90%, and 91% respectively. These results indicate that the YOLO version x architecture achieves better accuracy, leading to online testing being performed using YOLOv5x and YOLOv8x models. The online testing is conducted in different parts of the room, using 1, 2, and 3 faces in both normal and dark conditions. The results show that YOLOv5x achieves an accuracy of 88.4%, while YOLOv8x achieves 82.2%. During the testing with a single face, the model can detect and recognize faces more quickly and accurately, but as the number of faces in the camera's view increases, the model needs to work harder to recognize faces. This research demonstrates that faces can still be recognized when the distance between CCTV and faces is 1-3 meters long. Kata kunci: Security system, Biometric, CCTV, Face Detection and Recognition, You Only Look Once, YOLO, Deep Learning.

---

Availability

Inventory Code	Barcode	Call Number	Location	Status
2307004205	T123520	T1235202023	Central Library (Referens)	Available but not for loan - Not for Loan

Detail Information

Series Title

-

Call Number

T1235202023

Publisher

Indralaya : Jurusan Teknik Elektro, Fakultas Teknik Universitas Sriwijaya., 2023

Collation

xiii, 76 hlm.; Ilus.; 29 cm

Language

Indonesia

ISBN/ISSN

-

Classification

620.009 207

Content Type

-

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Jurusan Teknik Elektro
Ahli Teknologi

Specific Detail Info

-

Statement of Responsibility

PITRIA

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment