In English

Road Extraction from Aerial Images

Rickard Sirefelt
Göteborg : Chalmers tekniska högskola, 2015. Diploma work - Department of Applied Mechanics, Chalmers University of Technology, Göteborg, Sweden, ISSN 1652-8557; 2015:82, 2015.
[Examensarbete på avancerad nivå]

Road extraction from aerial images is essential to everyday life having several useful application in for example urban planning and automotive navigation. Roads are currently extracted using methods that requires a lot of manual work conducted by humans which is both time consuming and error prone. The aim of this thesis is to develop a robust algorithm that can automatically extract roads from aerial images. It was concluded, based on a literature review, that the most suitable method for automatic road extraction is a machine learning approach based on stacked convolutional neural networks. The method was implemented and evaluated against four different road images in the vicinity of the motorway E6 in southern Sweden. The best network achieved a recall of 0.845, precision of 0.878 and quality of 0.760 over a test set of previously unseen images. Considering that the method used was relatively simple, the result is to be considered competitive compared to other published works.

Nyckelord: automatic road extraction, pattern recognition, machine learning, image processing, image analysis, convolutional neural networks, classified road detection, aerial image analysis

Publikationen registrerades 2015-10-05. Den ändrades senast 2015-10-05

CPL ID: 223677

Detta är en tjänst från Chalmers bibliotek