Please use this identifier to cite or link to this item: https://hdl.handle.net/1889/3823
Full metadata record
DC FieldValueLanguage
dc.contributor.advisorBertozzi, Massimo-
dc.contributor.advisorBroggi, Alberto-
dc.contributor.authorGhidini, Francesca-
dc.date.accessioned2019-04-17T10:28:12Z-
dc.date.available2019-04-17T10:28:12Z-
dc.date.issued2019-
dc.identifier.urihttp://hdl.handle.net/1889/3823-
dc.description.abstractIn the last three decades we have assisted to an accelerated growth of intelligent systems for automated driving. These advancements are pushed by the idea of reducing traffic accidents and by important economic impacts. Advanced Driving Assistance Systems (ADAS) are nowadays readily available even in small and cheap car models. Autonomous lane keeping and adaptive cruise control are just some examples. Moreover, the automotive industry and many technology developers are working on highly-automated vehicles. All these systems require a complete knowledge of the vehicle surroundings. In this scenario computer vision based on the use of cameras plays a fundamental role since cameras are cheap, easily integrable on a vehicle and they provide a huge quantity of information. The main drawback of these sensors is the amount of processing required for extracting useful information from pixels. At the same time, space is one of the most an exiguous resource on a car and this put a challenge in implementing computer vision techniques on resource- constrained embedded platforms. Highways and non-urban roads are well limited by lane markings. However, urban areas are more complex scenarios where the drivable area could be delimited by lane markings, parked cars, walls, and curbs. For these reasons the use of ADAS is usually limited to highway scenarios where markings are visible and the geometry of the road is regular.The purpose of this research is to develop robust perception systems for autonomous vehicle using computer vision methods able to run with real-time performance on a embedded platform. The works in this field are focused on the specific problem of road boundaries detection, in particular curbs and barriers. Both these structures represent an important input for many high-level applications (path-planning, ego-lane estimation, obstacle avoidance, localization, etc) especially when lane markings are not present or when the complexity of the scene is high. The main contributions of this research are: two innovative approaches to curbs detection, one based on stereovision and the other based on monocamera and deep learning, and a new approach to road barriers detection with stereo camera.During this research emerged that the process of development computer vision solutions for the automotive industry is accompanied by several constraints: reliable and robust results are required as well as methods for handling adverse environmental conditions, moreover software has to be designed in order to require as little as hardware as possible.it
dc.language.isoIngleseit
dc.publisherUniversità degli studi di Parma. Dipartimento di Ingegneria e architetturait
dc.relation.ispartofseriesDottorato di ricerca in Tecnologie dell'informazioneit
dc.rights© Francesca Ghidini, 2019it
dc.subjectstereovisionit
dc.subjectDEMit
dc.subjectroad boundariesit
dc.subjectcurbsit
dc.subjectbarriersit
dc.subjectdeep learningit
dc.subjectsemantic segmentationit
dc.subjectembeddedit
dc.titleStudy and development of road boundaries detection applications for embedded platform implementationit
dc.title.alternativeStudio e sviluppo di applicazioni per l'individuazione dei bordi stradali su piattaforma embeddedit
dc.typeDoctoral thesisit
dc.subject.miurING-INF/05it
Appears in Collections:Tecnologie dell'informazione, tesi di dottorato

Files in This Item:
File Description SizeFormat 
relazione-finale-EN.pdf
  Until 2100-01-01
Relazione finale attività di dottorato55.63 kBAdobe PDFView/Open Request a copy
TesiDottorato.pdfTesi68.32 MBAdobe PDFView/Open


This item is licensed under a Creative Commons License Creative Commons