Real-Time Lane Detection on Embedded Systems for Control of Semi-Autonomous Vehicles

Thumbnail Image
Files
master_Young_Jacob_2022.pdf (7.45 MB)
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Author
Date
2022-10-17
Department
Major/Subject
Visual Computing and Communication
Mcode
SCI3102
Degree programme
Master's Programme in ICT Innovation
Language
en
Pages
63 + 9
Series
Abstract
Car accidents are the leading cause of death and injuries in most countries. advanced driving assistance systems and intelligent autonomous vehicles aim to improve road safety, traffic issues, and the comfort of passengers. Lane detection is a pivotal element in advanced driving assistance systems as lane understanding is essential in maneuvering the car safely on roads. Detecting lanes in real-world scenarios is challenging due to adverse weather, lighting conditions, and occlusions. However, as the computational budget available for lane detection in the systems above is limited, a lightweight, fast and accurate lane detection system is crucial. This thesis proposes a simple, lightweight, end-to-end deep learning-based lane detection framework following the row-wise classification approach. The inference speed is significantly increased by reducing the computational complexity and using a light backbone. In contrast to other systems, the proposed method can handle lane-changing scenarios by offering three lane candidates within the model. Additionally, we introduced a second-order polynomial fitting method and Kalman filter for tracking lane points as post-processing steps to improve the overall accuracy and stability of the system. The proposed lane detection method can provide over 500 frames per second on an Nvidia GTX 3080 notebook with our lightweight model and a median 48 frames per second on an Nvidia Jetson AGX Xavier while producing comparable accuracy to most of the state-of-the-art approaches.
Description
Supervisor
Kannala, Juho
Thesis advisor
Abrishami, Vahid
Tossavainen, Timo
Keywords
lane detection, deep learning, embedded systems, ADAS, autonomous driving, machine learning
Other note
Citation
Permanent link to this item
https://urn.fi/URN:NBN:fi:aalto-202210236108
Collections
[dipl] Perustieteiden korkeakoulu / SCI