Abstract
Abstract. This paper addresses the problem of road region detection in urban areas using an image classification approach. In order to minimize the spectral superposition of the road (asphalt) class with other classes, the Artificial Neural Networks (ANN) image classification method was used to classify geometrically-integrated high-resolution RGB aerial and laser-derived images. The RGB image was combined with different laser data layers and the ANN classification results showed that the radiometric and geometric laser data allows a better detection of road pixel.
Highlights
Various methods have been proposed for classification of remote sensing images taken from complex urban regions (Bellens et al, 2008, Pálsson et al, 2012)
An unsupervised classification is applied to the IKONOS images, followed by post-processing steps applied to the resulting road class, including noise removal, skeletonization to extract road segments, and linking of road segments
The test area comprised an urban region of the city of Curitiba, Southern Brazil, for which an aerial high-resolution (GSD ~ 0.2 m) RGB image and an airborne laser scanning (ALS) point cloud at an average resolution of 0.5 m were available
Summary
Various methods have been proposed for classification of remote sensing images taken from complex urban regions (Bellens et al, 2008, Pálsson et al, 2012). A possible strategy to minimize this problem consists in integrating airborne laser scanning (ALS) data and multispectral images, thereby improving the classification results. Benkouider et al (2011) developed a method to separate the road class in RGB SPOT images using spectral characteristics of roads in a classification process using Artificial Neural Networks (ANN), followed by a morphological post-processing to regularize the previously classified roads. A method for extracting roads from multispectral IKONOS images was proposed by Gao and Wu (2004). An unsupervised classification is applied to the IKONOS images, followed by post-processing steps applied to the resulting road class, including noise removal, skeletonization to extract road segments, and linking of road segments. Mancine et al (2009) used the machine-learning algorithm AdaBoost (i.e., Adaptive Boosting) to classify multispectral aerial images that were integrated with ALS data.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
