IMPLEMENTASI SISTEM DETEKSI DAN COUNTING KENDARAAN BERBASIS VIDEO CCTV DENGAN METODE DEEP LEARNING YOLOv8

Muhammad Azfa Faadhilah, - (2025) IMPLEMENTASI SISTEM DETEKSI DAN COUNTING KENDARAAN BERBASIS VIDEO CCTV DENGAN METODE DEEP LEARNING YOLOv8. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Peningkatan volume dan kompleksitas lalu lintas di perkotaan menuntut adanya sistem pengawasan yang efisien, mengingat metode manual dan konvensional sering kali rawan kesalahan dan tidak dapat diandalkan. Penelitian ini menggunakan metode D&D yang bertujuan untuk mengimplementasi model YOLOv8 untuk mendeteksi, mengklasifikasikan, dan menghitung jumlah kendaraan serta mengevaluasi sistem tersebut. Sistem ini dibangun dengan metode deep learning YOLOv8 yang dilatih secara khusus pada dataset kustom untuk mendeteksi enam kelas kendaraan umum di Indonesia (motor, mobil, truk, bis, angkot, pickup). Sistem ini menerapkan algoritma pelacakan objek yang berfungsi untuk memberikan ID unik pada setiap kendaraan yang terdeteksi. ID ini memungkinkan sistem untuk mengikuti pergerakan kendaraan di sepanjang video, sehingga memastikan setiap kendaraan dihitung hanya satu kali saat melintasi garis virtual. Seluruh fungsionalitas ini diintegrasikan ke dalam sebuah aplikasi web yang memungkinkan pengguna mengunggah video, menentukan garis penghitung, dan memantau hasil secara langsung. Pengujian dilakukan pada 12 rekaman video dari empat lokasi di Bandung pada kondisi siang, sore, dan malam hari. Hasilnya menunjukkan performa sistem yang sangat baik secara keseluruhan, dengan micro average Akurasi 98.67%, Presisi 97.22%, Recall 95.72%, dan F1-Score 96.39%. Performa tertinggi tercatat pada kondisi siang hari, sementara kondisi sore hari justru menunjukkan penurunan performa terendah. Sistem secara umum berhasil terhindar dari perhitungan ganda, meskipun analisis mendalam menemukan beberapa kasus kesalahan klasifikasi dan overcounting pada kondisi tertentu. Sistem deteksi dan penghitungan kendaraan end-to-end berbasis YOLOv8 ini berhasil diimplementasikan dengan fungsionalitas penuh dan menunjukkan performa kuantitatif yang sangat baik, meskipun kinerjanya terbukti dipengaruhi oleh kondisi lingkungan. --------- The increasing volume and complexity of urban traffic demand an efficient surveillance system, as manual and conventional methods are often prone to errors and unreliable. This research employs the Design and Development (D&D) method to implement and evaluate a system using the YOLOv8 model for detecting, classifying, and counting vehicles. The system is built using the YOLOv8 deep learning model, specifically trained on a custom dataset to detect six common vehicle classes in Indonesia: motorcycles, cars, trucks, buses, angkot (public minivans), and pickup trucks. The system implements an object tracking algorithm that assigns a unique ID to each detected vehicle. This ID enables the system to track vehicle movement throughout the video, ensuring each vehicle is counted only once as it crosses a virtual line. All these functionalities are integrated into a web application that allows users to upload videos, define a counting line, and monitor the results in real-time. The system was tested on 12 video recordings from four locations in Bandung during day, afternoon, and night conditions. The results indicate excellent overall system performance, with a micro average Accuracy of 98.67%, Precision of 97.22%, Recall of 95.72%, and an F1-Score of 96.39%. The highest performance was recorded in daytime conditions, while afternoon conditions showed the lowest performance. The system generally avoided double counting, although in-depth analysis revealed some cases of misclassification and overcounting under specific conditions. This end-to-end YOLOv8-based vehicle detection and counting system was successfully implemented with full functionality and demonstrated excellent quantitative performance, although its performance was proven to be influenced by environmental conditions.

	Text S_TEKOM_2002878_Title.pdf Download (2MB)
	Text S_TEKOM_2002878_Chapter1.pdf Download (256kB)
	Text S_TEKOM_2002878_Chapter2.pdf Restricted to Staf Perpustakaan Download (406kB)
	Text S_TEKOM_2002878_Chapter3.pdf Download (866kB)
	Text S_TEKOM_2002878_Chapter4.pdf Restricted to Staf Perpustakaan Download (921kB)
	Text S_TEKOM_2002878_Chapter5.pdf Download (195kB)
	Text S_TEKOM_2002878_Appendix.pdf Restricted to Staf Perpustakaan Download (5MB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?hl=en&user=CCBsTo8AAAAJ ID SINTA Dosen Pembimbing Munawir: 6745899 Wirmanto Suteddy: 6745736
Uncontrolled Keywords:	Deteksi Kendaraan, Penghitungan Kendaraan, Deep Learning, Sistem Pengawasan Lalu Lintas, YOLOv8, Vehicle Detection, Vehicle Counting, Deep Learning, Traffic Monitoring System.
Subjects:	L Education > L Education (General) T Technology > T Technology (General) T Technology > TA Engineering (General). Civil engineering (General)
Divisions:	UPI Kampus cibiru > S1 Teknik Komputer
Depositing User:	Muhammad Azfa Faadhilah
Date Deposited:	15 Sep 2025 04:22
Last Modified:	15 Sep 2025 04:22
URI:	http://repository.upi.edu/id/eprint/137348

Actions (login required)

View Item