Analisis Performa Deteksi Jenis Kendaraan pada Rekaman CCTV Lalu Lintas Berdasarkan Kondisi Pencahayaan

Sianturi, Sarma Elvita Malona (2026) Analisis Performa Deteksi Jenis Kendaraan pada Rekaman CCTV Lalu Lintas Berdasarkan Kondisi Pencahayaan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221031-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (20MB) | Request a copy

Abstract

Untuk membangun sistem pemantauan lalu lintas kendaraan bermotor secara otomatis dan real-time, diperlukan sistem deteksi kendaraan bermotor berdasarkan rekaman video CCTV. Rekaman CCTV merupakan sumber data yang potensial untuk mendukung pemantauan lalu lintas. Penelitian ini bertujuan untuk merancang bangun sistem deteksi kendaraan bermotor menggunakan model YOLOv11 serta mengevaluasi performanya berdasarkan tingkat pencahayaan. Data yang digunakan berupa rekaman CCTV lalu lintas dari aplikasi SITS (Surabaya Intelligent Transportation System) dengan empat kelas kendaraan, yaitu motorcycle, car, bus, dan truck. Tahapan penelitian meliputi ekstraksi citra dari video, anotasi data, pembentukan dataset daytime, nighttime, dan gabungan, praproses data, augmentasi data, pelatihan model, serta evaluasi menggunakan metrik Mean Average Precision pada ambang IoU 0,5 (mAP50). Evaluasi dilakukan melalui tujuh skenario pengujian, yaitu: (1) model dilatih dan diuji menggunakan dataset daytime; (2) model dilatih dan diuji menggunakan dataset nighttime; (3) model dilatih menggunakan dataset daytime dan diuji menggunakan dataset nighttime; (4) model dilatih menggunakan dataset nighttime dan diuji menggunakan dataset daytime; (5) model dilatih menggunakan dataset gabungan (daytime dan nighttime) kemudian diuji menggunakan dataset daytime; (6) model dilatih menggunakan dataset gabungan (daytime dan nighttime) kemudian diuji menggunakan dataset nighttime; dan (7) model dilatih dan diuji menggunakan dataset gabungan (daytime dan nighttime). Hasil penelitian menunjukkan bahwa performa terbaik diperoleh pada model yang dilatih menggunakan dataset gabungan dan diuji menggunakan dataset daytime dengan nilai mAP50 sebesar 0, 886. Sementara itu, performa terendah diperoleh pada model yang dilatih menggunakan dataset daytime dan diuji menggunakan dataset nighttime dengan nilai mAP50 sebesar 0, 568. Hasil tersebut menunjukkan bahwa penggunaan dataset gabungan mampu meningkatkan kemampuan generalisasi model sehingga menghasilkan performa yang lebih stabil pada berbagai kondisi pencahayaan ======================================================================================================================================
To build a real-time and automatic motor vehicle traffic monitoring system, a motor vehicle detection system based on CCTV video recordings is required. CCTV recordings are a potential data source to support traffic monitoring. This study aims to design and build a motor vehicle detection system using the YOLOv11 model and to optimize its performance based on lighting levels. The data used are traffic CCTV recordings from the SITS (Surabaya Intelligent Transportation System) application with four vehicle classes, namely motorcycle, car, bus, and truck. The research stages include image extraction from videos, data annotation, formation of daytime, nighttime, and combined datasets, data preprocessing, data augmentation, model training, and evaluation using the Mean Average Precision metric at an IoU threshold of 0.5 (mAP50). The evaluation is carried out through seven test scenarios, namely: (1) the model is drilled and tested using the daytime dataset; (2) the model was drilled and tested using the nighttime dataset; (3) the model was drilled using the daytime dataset and tested using the nighttime dataset; (4) the model was drilled using the nighttime dataset and tested using the daytime dataset; (5) the model was drilled using the combined dataset (daytime and nighttime) and then tested using the daytime dataset; (6) the model was drilled using the combined dataset (daytime and nighttime) and then tested using the nighttime dataset; and (7) the model was drilled and tested using the combined dataset (daytime and nighttime). The results showed that the best performance was obtained by the model drilled using the combined dataset and tested using the daytime dataset with an mAP50 value of 0.886. Meanwhile, the lowest performance was obtained by the model drilled using the daytime dataset and tested using the nighttime dataset with an mAP50 value of 0.568. These results indicate that the use of the combined dataset can improve the model’s generalization capability, resulting in more stable performance under various lighting conditions

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CCTV SITS, Deteksi Objek, Klasifikasi Kendaraan, Tingkat Pencahayaan, YOLOv11, SITS CCTV, Object Detection, Vehicle Classification, Lighting Conditions, YOLOv11
Subjects:	Q Science > QA Mathematics
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Sarma Elvita Malona Sianturi
Date Deposited:	04 Aug 2026 01:01
Last Modified:	04 Aug 2026 01:01
URI:	http://repository.its.ac.id/id/eprint/142697

Actions (login required)

View Item