Implementasi Convolutional Neural Network Untuk Real-Time Face Recognizing Menggunakan Python

Rahman, Muhammad Aditya (2026) Implementasi Convolutional Neural Network Untuk Real-Time Face Recognizing Menggunakan Python. S1 Teknik Informatika thesis, STMIK Widya Cipta Dharma.

	Text 2443902-S1-Jurnal.pdf Download (474kB)
	Text 2443902-S1-Teknik Informatika.pdf Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

ABSTRAK Pengawasan keamanan dan pemantauan keberadaan individu di lingkungan kerja masih banyak bergantung pada pengamatan petugas serta koordinasi secara langsung. Keterbatasan jumlah personel keamanan dan luasnya area kantor menyebabkan proses pemantauan belum dapat dilakukan secara optimal. Selain itu, pencarian informasi mengenai keberadaan pimpinan maupun pegawai masih sering dilakukan dengan bertanya kepada pegawai lain, menghubungi individu yang bersangkutan, atau melakukan pengecekan langsung ke ruangan. Proses tersebut menjadi kurang efisien ketika informasi dibutuhkan dalam waktu cepat. Penelitian ini bertujuan mengembangkan sistem pengenalan wajah secara real-time untuk membantu proses identifikasi individu dan pemantauan keberadaan pegawai di lingkungan kerja. Pengembangan sistem dilakukan menggunakan tahapan Cross-Industry Standard Process for Data Mining (CRISP-DM). Sistem mengintegrasikan YOLOv8n-Face sebagai metode deteksi wajah dan FaceNet dengan arsitektur InceptionResnetV1 berbobot pretrained VGGFace2 sebagai metode ekstraksi karakteristik wajah. Tahapan pengolahan meliputi pengambilan citra wajah, deteksi wajah, square cropping, penambahan margin, perubahan ukuran menjadi 160 × 160 piksel, peningkatan kontras pencahayaan menggunakan Contrast Limited Adaptive Histogram Equalization (CLAHE), penajaman citra, ekstraksi face embedding berdimensi 512, normalisasi L2, serta pencocokan identitas menggunakan cosine similarity terhadap centroid setiap identitas. Sistem menerapkan similarity threshold dan similarity margin untuk menentukan apakah wajah diterima sebagai identitas terdaftar atau diberikan label Unknown. Selain itu, temporal smoothing dan pelacakan berbasis Intersection over Union diterapkan untuk menjaga kestabilan hasil pengenalan ketika beberapa wajah terdeteksi secara bersamaan. Berdasarkan hasil pengujian, sistem memperoleh Known Recognition Rate sebesar 0,8779 atau 87,79%, Unknown Detection Rate sebesar 1,0 atau 100%, False Acceptance Rate sebesar 0,0 atau 0%, dan False Rejection Rate sebesar 0,1220 atau 12,2%, dengan kecepatan pemrosesan rata-rata sebesar 12,45 frame per second dan accuracy sebesar 89,12%. Hasil penelitian menunjukkan bahwa integrasi YOLOv8n-Face, FaceNet, dan cosine similarity mampu mendeteksi serta mengenali beberapa wajah secara real-time melalui kamera internal maupun eksternal. Sistem juga mampu membedakan wajah yang terdaftar dan wajah yang belum terdaftar dengan memberikan label Unknown, sehingga dapat digunakan sebagai prototipe alat bantu identifikasi dan pemantauan keberadaan individu di lingkungan kerja. Kata kunci: pengenalan wajah, YOLOv8n-Face, FaceNet, face embedding, cosine similarity, real-time. ============================================================ ABSTRACT Workplace security supervision and the monitoring of individuals’ presence still largely depend on direct observation by security personnel and manual coordination. The limited number of security personnel and the extensive office area may prevent monitoring activities from being carried out optimally. In addition, information regarding the presence of managers or employees is often obtained by asking other employees, contacting the individuals concerned, or checking their offices directly. This process becomes inefficient when information is required immediately. This study aims to develop a real-time face recognition system to assist in identifying individuals and monitoring employees’ presence in the workplace. The system was developed using the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology. It integrates YOLOv8n-Face as the face detection method and FaceNet with the InceptionResnetV1 architecture pretrained on VGGFace2 as the facial feature extraction model. The processing stages include facial image acquisition, face detection, square cropping, margin addition, resizing to 160 × 160 pixels, illumination contrast enhancement using Contrast Limited Adaptive Histogram Equalization (CLAHE), image sharpening, extraction of 512-dimensional face embeddings, L2 normalization, and identity matching using cosine similarity against the centroid of each registered identity. The system applies a similarity threshold and a similarity margin to determine whether a detected face is accepted as a registered identity or labeled as Unknown. Temporal smoothing and Intersection over Union-based tracking are also applied to maintain recognition stability when multiple faces are detected simultaneously. Based on the test results, the system achieved a Known Recognition Rate of 0.8779 or 87,79%, an Unknown Detection Rate of 1.0 or 100%, a False Acceptance Rate of 0.0 or 0%, and a False Rejection Rate of 0.1220 or 12,2%, with an average processing speed of 12.45 frames per second and with accuracy of 87,12%. The results demonstrate that the integration of YOLOv8n-Face, FaceNet, and cosine similarity can detect and recognize multiple faces in real time using internal or external cameras. The system can also distinguish between registered and unregistered faces by assigning the Unknown label, making it applicable as a prototype to support individual identification and presence monitoring in the workplace. Keywords: face recognition, YOLOv8n-Face, FaceNet, face embedding, cosine similarity, real-time.

Item Type:	Thesis (S1 Teknik Informatika)
Additional Information:	Wahyuni, S.Kom., M.Kom. Pitrasacha Adytia, S.T., M.T.
Uncontrolled Keywords:	pengenalan wajah, YOLOv8n-Face, FaceNet, face embedding, cosine similarity, real-time
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Divisions:	Teknik Informatika
Depositing User:	Mr Muhammad Aditya Rahman
Date Deposited:	06 Aug 2026 07:22
Last Modified:	06 Aug 2026 07:22
URI:	http://repository.wicida.ac.id/id/eprint/6391

Actions (login required)

View Item