Work

Real-Time Road Estimation for Autonomous Driving

Year: 2016
Type: dissertation
Abstract: During the last decade, there has been an increasing interest and dedication towards developing self-driving cars.The enhancements of sensor technology and processing power have made more advanced autonomous systems possible.For self-driving cars, robust and accurate navigation is essential.In that context, cameras, in collaboration with specialized graphics processing units (GPU), has been shown to serve as a valuable tool to map the local scenery surrounding the vehicle.This thesis proposes a computer vision based method for detecting road in a local scenery, purposed for road following.The method utilizes on-line machine learning techniques to model the road, in terms of a set of features.The choice of features is widely discussed in this thesis, as is the methods for modeling the road.The proposed methods for both modeling and detecting the road are executable in real-time and are adaptable to variation in road appearance.The algorithm has been thoroughly tested in four different sceneries, which includes three recorded sequences containing tarmac road, and one containing dirt road.The algorithm has been evaluated in relation to similar state-of-the-art algorithms which has proved reliable for road detection.The proposed algorithm shows great promise for detecting tarmac road in homogeneous scenery, where it almost matches the state-of-theart algorithms.However, it struggles when the road appearance is less consistent throughout the images.Dirt road, inconsistently shadowed road and extreme illumination conditions currently pose as challenging situations.iii A special thanks to Anders Ueland, for all your help, and for all the helpful discussions we have had.
Author: Eirik Sundet
Institution: University of Oslo
Language: English
Open Access status: green
OpenAlex ID: W2551914308
Cites: 27
Cited by:
Related to: 10