PENERAPAN SINGLE SHOT MULTIBOX DETECTOR (SSD) UNTUK DETEKSI PELANGGARAN KENDARAAN PARKIR ILEGAL BERBASIS CITRA VIDEO (skripsi)

KHANSA FARRAS CALLISTA ARMANDSYAH, . (2025) PENERAPAN SINGLE SHOT MULTIBOX DETECTOR (SSD) UNTUK DETEKSI PELANGGARAN KENDARAAN PARKIR ILEGAL BERBASIS CITRA VIDEO (skripsi). Sarjana thesis, UNIVERSITAS NEGERI JAKARTA.

[img] Text
COVER.pdf

Download (1MB)
[img] Text
BAB I.pdf

Download (499kB)
[img] Text
BAB II_compressed.pdf
Restricted to Registered users only

Download (782kB) | Request a copy
[img] Text
BAB III_compressed.pdf
Restricted to Registered users only

Download (499kB) | Request a copy
[img] Text
BAB IV_compressed.pdf
Restricted to Registered users only

Download (931kB) | Request a copy
[img] Text
BAB V.pdf
Restricted to Registered users only

Download (475kB) | Request a copy
[img] Text
DAFTAR PUSTAKA.pdf

Download (501kB)
[img] Text
LAMPIRAN_compressed.pdf
Restricted to Registered users only

Download (390kB) | Request a copy

Abstract

Parkir liar di bahu jalan merupakan salah satu penyebab kemacetan lalu lintas dan gangguan ketertiban umum. Penelitian ini bertujuan untuk mengembangkan sistem deteksi dan pelacakan otomatis kendaraan yang berhenti secara ilegal menggunakan rekaman video pengawasan. Metode yang digunakan melibatkan deteksi objek berbasis Single Shot Multibox Detector (SSD) dengan tiga arsitektur model, yaitu MobileNetV2, MobileNetV2-FPN Lite, dan ResNet50 V1, serta algoritma pelacakan DeepSORT untuk identifikasi pergerakan kendaraan. Dataset berupa cuplikan video lalu lintas diperoleh dari PT Citra Persada Infrastruktur, kemudian dianotasi ke dalam lima kelas objek: mobil, truk, motor, bus, pickup. Hasil evaluasi menunjukkan bahwa model SSD-ResNet50V1 memiliki akurasi terbaik secara keseluruhan, dengan mAP tertinggi sebesar 0,9634. Untuk pengujian pada video MobileNetV2-FPN Lite menjadi model paling cocok karena memiliki kecepatan inferensi tertinggi (hingga 0,57 FPS) dan berjalan stabil pada perangkat dengan daya komputasi terbatas, meskipun akurasinya lebih rendah dari ResNet50V1. Penelitian ini membuktikan bahwa kombinasi SSD dan DeepSORT mampu diterapkan secara efektif untuk mendeteksi pelanggaran parkir liar pada rekaman CCTV, serta memberikan insight penting dalam memilih threshold yang sesuai untuk skenario nyata. ***** Ilegal parking on road shoulders is one of the causes of traffic congestion and public order disturbances. This research aims to develop an automatic detection and tracking system for vehicles that stop ilegally using surveillance video footage. The method employed involves object detection using the Single Shot Multibox Detector (SSD) with three model architectures MobileNetV2, MobileNetV2-FPN Lite, and ResNet50V1 and the DeepSORT tracking algorithm for vehicle movement identification. The dataset consists of traffic video footage obtained from PT Citra Persada Infrastruktur and was annotated into five object classes: car, truck, motorcycle, bus, and pickup. Evaluation results show that the ResNet50V1 model achieves the highest overall accuracy, with a mAP of 0,9634. For video testing, MobileNetV2-FPN Lite was the most suitable model due to its highest inference speed (up to 0,57 FPS) and ability to run stably on devices with limited computational resources, despite having lower accuracy compared to ResNet50V1. This study demonstrates that the combination of SSD and DeepSORT can be effectively applied for ilegal parking detection using CCTV footage, and also provides valuable insights into choosing an appropriate confidence threshold for real-world scenarios.

Item Type: Thesis (Sarjana)
Additional Information: 1). Dr. rer. nat Bambang Heru Iswanto, M.Si 2). Haris Suhendar, M.Sc
Subjects: Sains > Fisika
Divisions: FMIPA > S1 Fisika
Depositing User: Farras Callista Armandsyah .
Date Deposited: 09 Oct 2025 06:41
Last Modified: 09 Oct 2025 06:41
URI: http://repository.unj.ac.id/id/eprint/62574

Actions (login required)

View Item View Item