Pengembangan Autonomous Underwater Vehicle untuk Menjejak Pipa Bawah Laut Menggunakan Visi Komputer

Dzulfadli, Muhammad Hasan (2024) Pengembangan Autonomous Underwater Vehicle untuk Menjejak Pipa Bawah Laut Menggunakan Visi Komputer. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5022201210_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (5MB)

Abstract

Menjejak pipa bawah laut adalah proses mendeteksi dan mengikuti jalur pipa di dasar laut yang penting dilakukan untuk pemeliharaan dan inspeksi guna mencegah kegagalan sistem yang dapat menyebabkan dampak lingkungan dan ekonomi yang signifikan. Variasi karakteristik lingkungan bawah laut membuat deteksi pipa menjadi sulit bagi visi komputer konvensional. Dalam tugas akhir ini, sistem Autonomous Underwater Vehicle (AUV) dikembangkan menggunakan teknologi visi komputer berbasis Convolutional Neural Network (CNN) melalui pendekatan YOLOv8. Algoritma YOLOv8 Instance Segmentation digunakan untuk mendeteksi pipa dalam gambar yang diambil oleh kamera AUV, dan hasil deteksi diolah menggunakan OpenCV untuk menentukan posisi objek dalam frame. Data ini digunakan oleh algoritma pipe-following yang dirancang untuk menavigasi AUV secara otomatis di sepanjang jalur pipa. Sistem ini menggunakan framework ROS2 Humble pada OS Ubuntu 22.04 yang tertanam di Mini-PC Intel NUC 13 dalam AUV, melibatkan komponen elektronik lainnya seperti mikrokontroler STM32F407 Discovery, kamera USB Bluerobotics, sensor IMU BNO055, dan pressure sensor MS5837. Tiga variasi model YOLOv8 (nano, small, dan medium) diuji, dengan hasil evaluasi menunjukkan bahwa model YOLOv8-nano memberikan performa terbaik untuk deteksi pipa secara real-time dengan nilai mAP@[0.5:0.95] sebesar 0.78143 dan dapat dijalankan dengan kecepatan rata-rata 6,54 FPS. Pada pengujian algoritma pipe-following dilakukan di kolam renang KONI Jawa Timur diperoleh nilai MAE sebesar 52,22% dengan standar deviasi 196 piksel untuk titik atas dan MAE sebesar 24,83% untuk titik bawah dengan standar deviasi 116 piksel. Pada pengujian di perairan Gili Ketapang Probolinggo diperoleh nilai MAE sebesar 30,41% untuk titik atas dengan standar deviasi 125 piksel dan nilai MAE sebesar 12,46% untuk titik bawah dengan standar deviasi 79 piksel.
=================================================================================================================================
Subsea pipeline tracing is the process of detecting and following the path of a pipeline on the seabed, essential for maintenance and inspection to prevent system failures that can cause significant environmental and economic impacts. Variations in the characteristics of the underwater environment make pipeline detection difficult for conventional computer vision. In this final project, an Autonomous Underwater Vehicle (AUV) system is developed using Convolutional Neural Network (CNN)-based computer vision technology through the YOLOv8 approach. The YOLOv8 Instance Segmentation algorithm is used to detect pipes in images taken by the AUV camera, and the detection results are processed using OpenCV to determine the position of the object in the frame. This data is used by the pipe-following algorithm designed to automatically navigate the AUV along the pipeline path. The system uses Humble's ROS2 framework on Ubuntu 22.04 OS embedded in an Intel NUC 13 Mini-PC inside the AUV, involving other electronic components such as STM32F407 Discovery microcontroller, Bluerobotics USB camera, BNO055 IMU sensor, and MS5837 pressure sensor. Three variations of the YOLOv8 model (nano, small, and medium) were tested, with evaluation results showing that the YOLOv8-nano model provides the best performance for real-time pipe detection with mAP@[0.5:0.95] values close to other models and lower inference time. In testing the pipe-following algorithm carried out at the KONI East Java swimming pool, an MAE value of 52.22% was obtained with a standard deviation of 196 pixels for the upper point and an MAE of 24.83% for the lower point with a standard deviation of 116 pixels. In testing in the waters of Gili Ketapang Probolinggo, an MAE value of 30.41% was obtained for the top point with a standard deviation of 125 pixels and an MAE value of 12.46% for the bottom point with a standard deviation of 79 pixels.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Autonomous Underwater Vehicle, Image Processing, Convolutional Neural Network, You Look Only Once.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK6592.A9 Automatic tracking.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Hasan Dzulfadli
Date Deposited:	31 Jul 2024 02:56
Last Modified:	31 Jul 2024 02:57
URI:	http://repository.its.ac.id/id/eprint/110744

Actions (login required)

View Item