Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.691261
Title: Vision based lane detection for intelligent vehicles
Author: Ozgunalp, Umar
ISNI:       0000 0004 5917 3491
Awarding Body: University of Bristol
Current Institution: University of Bristol
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Abstract:
Most vehicle accidents are due to driver error or slow reaction time. To prevent or minimize the consequences of these accidents, Advanced Driver Assistance Systems (ADAS) are introduced and lane detection is one of the most important building blocks of ADAS. Thus, the main focus of this thesis is lane detection. In this thesis, initially, lane detection algorithms based on a single camera as a sensor are investigated and proposed. First, an Inverse Perspective Mapping (IPM) based lane detection algorithm is proposed, where the global lane orientation and lane connectivity in this direction is exploited to increase the Signal to Noise Ratio (SNR). Using the initially estimated lane orientation, feature map is iteratively shifted and matched with itself to eliminate noise. Furthermore, based on the global lane orientation, an accurate, and linear Region of Interest (ROI) is efficiently formed using a I-D likelihood accumulator, where lane pairs are fitted to the feature points in estimated ROI. Later, an extension to the Symmetrical Local Threshold (SLT) is proposed for more accurate feature map extraction. Despite low computational complexity of the SLT, the algorithm outperformed all of the tested lane feature extractors in the Road Marking (ROMA) data sets. However, the main drawback of this algorithm is it cannot supply orientation information for the feature points. The proposed extension to the SLT, both reduced the noise (tested using ROMA data sets) , and outputs orientation information for the extracted feature points. Then, the extracted feature map and feature point orientations, are exploited for an efficient lane detection, where lane categorization is achieved by using a mask in the Hough domain. Although, single camera can be used for lane detection, single camera cannot supply depth information. Thus, many lane detection algorithms using single camera input are based on assumptions such as flat road assumption. However, 3D input can be utilized for lane detection application on non-flat roads.
Supervisor: Dahnoum, Naim Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.691261  DOI: Not available
Share: