SISLY DESTRI AGUSTIN
Program Studi S1 Teknik Telekomunikasi, Universitas Telkom

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

Perbandingan Algoritma YOLOv4 dan Scaled YOLOv4 untuk Deteksi Objek pada Citra Termal AZIZAH AULIA RAHMAN; SISLY DESTRI AGUSTIN; NUR IBRAHIM; NOR CAECAR KUMALASARI
MIND (Multimedia Artificial Intelligent Networking Database) Journal Vol 7, No 1 (2022): MIND Journal
Publisher : Institut Teknologi Nasional, Bandung

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.26760/mindjournal.v7i1.61-71

Abstract

ABSTRAKMinimnya visibilitas pejalan kaki dan pengendara pada malam hari karena kurangnya pencahayaan pada lampu jalan menyebabkan kecelakaan rentan terjadi pada rentang waktu tersebut. Sistem penglihatan komputer berbeda dengan manusia, semua objek dengan suhu di atas nol dapat memancarkan radiasi inframerah jika direkam menggunakan kamera termal. Dalam penelitian ini penulis mengidentifikasi citra termal dalam bentuk citra RGB dengan algoritma YOLOv4 dan Scaled YOLOv4 sebagai deteksi objek. Performa sistem diukur berdasarkan nilai presisi, recall, f1-score, dan mAP. Eksperimen dilakukan pada dataset citra termal dengan objek manusia. Skenario yang digunakan adalah mendeteksi objek dengan jarak 5m, 10m, 15m, dan 20m. Hasil deteksi didapatkan algoritma Scaled YOLOv4 CSP lebih unggul dengan nilai pengujian precision 94,3%, recall 83,8%, f1-Score 88,7%, dan mAP 86,9%. Hasil tersebut dipengaruhi oleh ukuran citra dan jumlah dataset dari citra training, citra validation, dan citra uji.Kata kunci: Citra Termal, YOLO, YOLOv4, Scaled-YOLOv4, Deteksi ObjekABSTRACTThe lack of visibility of pedestrians and drivers at night due to lack of lighting in street lights makes accidents prone to occur during this time. The computer vision system is different from the humans, any object with a temperature above zero can emit infrared radiation when using a thermal camera. In this study, the authors identify thermal images in RGB using YOLOv4 and Scaled YOLOv4 as object detection algorithms. System performance is measured based on the value of precision, recall, f1-score, and mAP. Experiments were carried out on a thermal image dataset with human objects. The scenario used was to detect objects at a distance of 5m, 10m, 15m, and 20m. The detection results show that Scaled YOLOv4 CSP algorithms is the best, based on the test value of 94.3% precision, 83.8% recall, 88.7% f1-Score, and 86.9% mAP. These results are influenced by the size of the image and the number of datasets from training images, validation images, and test images.Keywords: Thermal Image, YOLO, YOLOv4, Scaled-YOLOv4, Object Detection