Autonomous Surface Vehicle Pencari Korban Kecelakaan Laut Berbasis Computer Vision.

Akbar, Achmad Zidan (2022) Autonomous Surface Vehicle Pencari Korban Kecelakaan Laut Berbasis Computer Vision. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 05111840000005-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (43MB)

Abstract

Berdasarkan laporan hasil investigasi Komite Nasional Keselamatan Transportasi (KNKT) di wilayah perairan Indonesia pada kurun waktu tahun 2010-2016, terjadi 54 kecelakaan di laut dengan berbagai jenis kejadian seperti tenggelam, terguling, kandas, dan tabrakan. Dalam proses pencarian dan evakuasi korban oleh Tim Search and Resue (SAR) kerap dijumpai hambatan seperti cuaca buruk, lokasi kecelakaan, keterbatasan peralatan, dan keterbatasan tenaga tim SAR. Hambatan tersebut tentu akan mengurangi probabilitas korban kecelakaan untuk selamat. Diusulkan sebuah Autonomous Surface Vehicle (ASV) yang dikembangkan untuk membantu proses penyelamatan korban kecelakaan laut. Jenis lambung kapal catamaran diimplementasikan dalam ASV ini karena memiliki tingkat stabilitas yang baik. ASV ini menggunakan sistem propulsi elektrik thruster T200 yang disematkan pada bagian belakang robot menggunakan sistem fix mounting. Robot Operating System (ROS) diimplementasikan sebagai sistem arsitektur perangkat lunak utama. ASV dilengkapi sensor navigasi seperti Global Positioning System (GPS), kompas, Inertial Measuring Unit (IMU), dan gyroscope. Sensor navigasi tersebut digunakan sebagai referensi program kontrol untuk melakukan navigasi autonomous. ASV juga menggunakan sensor ultrasonic untuk mendukung kemampuan obstacle avoidance. Untuk mendukung fungsionalitas ke aktuator, digunakan STM32F4 sebagai microcontroller. Computer Vision ditambahkan untuk meningkatkan kemampuan dari ASV untuk mendeteksi korban kecelakaan laut menggunakan kamera. Arsitektur YOLOv4 CNN digunakan sebagai pendeteksi orang tenggelam. TensorRT juga digunakansebagai optimizer model dengan perangkat graphics processing unit (GPU) Nvidia. Dengan menggunakan sistem navigasi yang dikembangkan, ASV mampu melakukan navigasi melewati perairan dalam operasi penyelamatan dengan baik. Sistem navigasi mampu memberikan output kontrol yang baik meskipun terdapat noise atau gangguan pada sensor kompas. ASV mampu menghindari halangan dengan cukup baik pada kecepatan rendah. Anotasi dataset dilakukan secara manual, dengan data gambar yang diambil di Danau 8 ITS. YOLOv4 sebagai arsitektur yang dipilih menghasilkan evaluasi mAP@IoU=0.5 pada data testing sebesar 0.840203. Terdapat peningkatan performa inference, model darknet YOLO yang diubah menjadi model TensorRT, dari 27 FPS menjadi 85 FPS. Dengan naiknya performa inference time, mAP@IoU=0.5 dari model tensorRT turun menjadi 0.776.
==================================================================================================================================
Based on the investigation report of the National Transportation Safety Committee (NTSC) in Indonesian waters in the period 2010-2016, there were 54 accidents at sea with various types of events such as drowning, overturning, running aground, and collisions. During the search and evacuation process for victims, the Search and Rescue (SAR) Team often encounters obstacles such as bad weather, accident locations, limited equipment, and limited staff for the SAR team. These obstacles will certainly reduce the possibility of accident victims to survive. An Autonomous Surface Vehicle (ASV) is proposed to be developed to assist the evacuation process of victims of marine accidents. Catamaran hull type was implemented in this ASV because it had a good stability. This ASV used the T200 thruster electronic propulsion system which was attached to the rear of the robot using a fix mounting system. Robot Operating System (ROS) was implemented as the main software architecture system. ASV was equipped with navigation sensors such as the Global Positioning System (GPS), Compass, Inertial Measuring Unit (IMU), and gyroscope. The navigation sensor was used as a program control reference for autonomous navigation. The ASV also used an ultrasonic sensor to support obstacle avoidance capabilities. To support the functionality to the actuator, STM32F4 was used as a microcontroller. Computer Vision was added to enhance the ability of the ASV to detect marine accident victims using cameras. YOLOv4 CNN Architecture was used as a drowning person detector. TensorRT was also used as an optimizer model with Nvidia graphics processing unit (GPU) devices. By using the developed navigation system, the ASV could navigate quite well through the waters for rescue operation. The navigation system could provide good output control although there was noise or interference in compass sensor. ASV could avoid obstacles quite well at low speeds. Dataset annotation was done manually, with image data taken at Lake 8 ITS. YOLOv4 as the chosen architecture produced the evaluation of mAP@IoU=0.5 on the testing data of 0.840203. There was an increase in inference performance, the YOLO darknet model was changed to the TensorRT model, from 27 FPS to 85 FPS. As the performance inference time increased, the mAP@IoU=0.5 of the tensorRT model decreased to 0.776.

Item Type:	Thesis (Other)
Additional Information:	RSIf 006.37 Akb a-1 2022
Uncontrolled Keywords:	Autonomous Surface Vehicle, Computer Vision, YOLO, Search and Rescue, Global Navigation Control, TensorRT. Autonomous Surface Vehicle, Computer Vision, Search and Rescue, Global Navigation Control, Catamaran.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Mr. Marsudiyana -
Date Deposited:	25 May 2026 02:44
Last Modified:	25 May 2026 02:44
URI:	http://repository.its.ac.id/id/eprint/133378

Actions (login required)

View Item