Date of Award

7-7-2020

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

Supervisor

Kemal Tepe

Rights

info:eu-repo/semantics/embargoedAccess

Abstract

Automakers are developing end-to-end three-dimensional (3D) mapping system for Advanced Driver Assistance Systems (ADAS) and autonomous vehicles (AVs). Using geomatics, artificial intelligence, and SLAM (Simultaneous Localization and Mapping) systems to handle all stages of map creation, sensor calibration and alignment. It is crucial to have a system highly accurate and efficient as it is an essential part of vehicle controls. Such mapping requires significant resources to acquire geographic information (GIS and GPS), optical laser and radar spectroscopy, Lidar, and 3D modeling applications in order to extract roadway features (e.g., lane markings, traffic signs, road-edges) detailed enough to construct a “base map”. To keep this map current, it is necessary to update changes due to occurring events such as construction changes, traffic patterns, or growth of vegetation. The information of the road play a very important factor in road traffic safety and it is essential for for guiding autonomous vehicles (AVs), and prediction of upcoming road situations within AVs. The data size of the map is extensive due to the level of information provided with different sensor modalities for that reason a data optimization and extraction from three-dimensional (3D) mobile laser scanning (MLS) point clouds is presented in this thesis. The research shows the proposed hybrid filter configuration together with the dynamic developed mechanism provides significant reduction of the point cloud data with reduced computational or size constraints. The results obtained in this work are proven by a real-world system.

Download

Share

COinS