Metode Pelabelan Dataset Segentasi CItra Jalan Otomatis menggunakan Lidar pada Mobil Otonom

Tantra, Pandu Surya (2024) Metode Pelabelan Dataset Segentasi CItra Jalan Otomatis menggunakan Lidar pada Mobil Otonom. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022212005-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (4MB) | Request a copy

Abstract

Sistem pendeteksi jalan adalah komponen kunci pada mobil otonom yang memungkinkan mobil untuk mendeteksi bidang jalan dan kemudian mengikuti jalur dengan tepat. Tanpa adanya sistem pendeteksi jalan yang andal maka mobil akan sulit melakukan navigasi dan aksi selanjutnya akan sulit ditentukan. Algoritma fusi sensor antara kamera dan lidar diyakini sebagai kunci keberhasilan sistem pendeteksi jalan. Namun implementasi kedua sensor tersebut pada skala besar tidak memungkinkan karena biayanya yang mahal. Oleh karena itu, sistem pendeteksi jalan banyak dikembangkan dengan kamera sebagai sensor utamanya menggunakan algoritma berbasis deep learning. Kualitas dan kuantitas dataset adalah masalah utama dan mendasar pada algoritma berbasis deep learning, menjadi tembok penghalang pada keberhasilan sistem. Penelitian ini mengusulkan metode otomatis untuk pelabelan dataset segmentasi citra jalan menggunakan lidar. Penelitian ini mencakup penentuan batas jalan, pembuatan peta lokal, dan algoritma road curve fitting. Penelitian ini juga mencakup pengembangan metode transformasi perspektif berbasis jaringan saraf tiruan. Hasil akhirnya adalah dataset segmentasi citra berbasis poligon untuk segmentasi citra jalan. Eksperimen yang dilakukan di kampus ITS pada segmen jalan sepanjang 1km menghasilkan lebih dari 2000 dataset dengan nilai rerata IoU yang tinggi sebesar 0.90.
=============================================================================================================================
The road detection system is a key component in autonomous vehicles, enabling them to detect the road surface and follow the path accurately. Without a reliable road detection system, navigation becomes challenging and subsequent actions become difficult to determine. The fusion algorithm between camera and lidar sensors is believed to be the key to the success of road detection systems. However, the large-scale implementation of both sensors is not feasible due to high costs. Therefore, many road detection systems are developed primarily using cameras as the main sensor, employing deep learning-based algorithms. The quality and quantity of datasets are major and fundamental issues in deep learning algorithms, posing significant barriers to the success of the system. This research proposes an automated method for labeling road image segmentation datasets using lidar. It encompasses road boundary determination, local map creation, and a road curve fitting algorithm. The research also includes the development of a perspective transformation method based on artificial neural networks. The final outcome is a polygon-based image segmentation dataset for road image segmentation. Experiments conducted on a 1km road segment within the ITS campus generated over 2000 datasets with a high average IoU score of 0.90.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	LiDAR, Segmentasi Citra, Dataset, Mobil Otonom, Image Segmentation, Dataset, Autonomous Vehicle
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL152.8 Vehicles, Remotely piloted. Autonomous vehicles. T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL589.2.N3 Navigation computer
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Pandu Surya Tantra
Date Deposited:	06 Feb 2024 01:29
Last Modified:	06 Feb 2024 01:29
URI:	http://repository.its.ac.id/id/eprint/106153

Actions (login required)

View Item