Dhiyya'ul Haq, Ikhwanul Abiyu (2025) Sistem Deteksi Kendaraan Overdimension secara Real-time di Gerbang Tol Menggunakan SSD-MobileNetV2 pada Edge Device. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5024211048-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5024211048-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy |
Abstract
Penggunaan truk sebagai moda transportasi barang di Indonesia terus meningkat, namun pelanggaran ODOL (Overdimension Overloading) pada truk menjadi salah satu penyebab utama kecelakaan lalu lintas. Penelitian ini bertujuan mengembangkan sistem deteksi kendaraan overdimension menggunakan teknologi deep learning, khususnya Convolutional Neural Network (CNN) dengan arsitektur SSD-MobileNetV2, yang dapat berjalan secara real-time di edge device. Model dilatih dengan dataset kendaraan yang dianotasi, mencapai nilai mAP 0,805 dan akurasi deteksi 80,72%. Pengujian pada dua jenis perangkat edge menunjukkan NVIDIA Jetson Nano memberikan performa terbaik dengan kecepatan inferensi 46,86 FPS, jauh lebih tinggi dibanding Beelink Gemini T34 (3,63 FPS). Sistem diimplementasikan di Gerbang Tol Dupak 2, Surabaya dengan tingkat keberhasilan transfer data mencapai 100% meskipun terdapat keterbatasan bandwidth. Dibandingkan penelitian sejenis, sistem ini memiliki keunggulan dalam keseimbangan akurasi dan kecepatan, serta efisiensi penggunaan bandwidth (4 KB/s). Sistem juga diintegrasikan dengan backend cloud dan dilengkapi fitur notifikasi otomatis kepada pihak berwenang saat pelanggaran terdeteksi. Dengan peningkatan akurasi dan efisiensi, sistem ini memberikan solusi efektif untuk mendeteksi kendaraan ODOL di berbagai lokasi dengan keterbatasan konektivitas.
===================================================================================================================================
The use of trucks as a mode of transportation for goods in Indonesia continues to in crease, but violations of ODOL (Overdimension Overloading) on trucks have become a major cause of traffic accidents. This research aims to develop an overdimension vehicle detection system using deep learning technology, specifically Convolutional Neural Network (CNN) with SSD-MobileNetV2 architecture, that can operate in real-time on edge devices. The model was trained with annotated vehicle datasets, achieving a mAP value of 0.805 and detection accu racy of 80.72%. Testing on two types of edge devices showed that NVIDIA Jetson Nano provided the best performance with an inference speed of 46.86 FPS, significantly higher than Beelink Gemini T34 (3.63 FPS). The system was implemented at Dupak 2 Toll Gate, Surabaya with a 100% data transfer success rate despite bandwidth limitations. Compared to similar research, this system has advantages in balancing accuracy and speed, as well as bandwidth efficiency (4 KB/s). The system is also integrated with a cloud backend and equipped with automatic notification features to authorities when violations are detected. With improved accuracy and efficiency, this system provides an effective solution for detecting ODOL vehicles in various locations with connectivity limitations
Actions (login required)
 |
View Item |