Urban Road Extraction from Combined Data Sets of High-Resolution Satellite Imagery and Lidar Data Using GEOBIA

Authored by: Dale A. Quattrochi , Elizabeth A. Wentz , Nina Siu-Ngan Lam , Charles W. Emerson , Minjuan Cheng , Qihao Weng

Integrating Scale in Remote Sensing and GIS

Print publication date:  January  2017
Online publication date:  January  2017

Print ISBN: 9781482218268
eBook ISBN: 9781315373720
Adobe ISBN:


 Download Chapter



Remotely sensed data have been widely used in urban road extraction. Low resolution data cannot provide adequately detailed information for accurate estimation of road surface conditions. In low-resolution images, road extraction is often regarded as linear feature extraction because roads are considered to be “continuous and smooth lines” (Amini et al. 2002). Most traditional classification methods are based on statistical analysis of individual pixels. Those classifiers are well suited to images with relatively low spatial resolution (Wang et al. 2004). As high-resolution satellite images become available, a large amount of detailed information of ground features may be readily available to extract. On high-resolution images, a road is no longer a linear network of a few pixels, but a ribbon of features with a certain width. Therefore, many factors, such as lanes, cars, pedestrians, and shadows of trees and buildings, may affect the extraction of roads from the satellite images. Because of large data volume, high-resolution imagery is often difficult to process and analyze. Some researchers tried to apply traditional feature extraction and classification methods to high-resolution imagery for road network detection (Shi and Zhu 2002; Long and Zhao 2005; Zhu et al. 2005; Gautama et al. 2006; Hu et al. 2007; Péteri and Ranchin 2007). These studies faced difficulties due to two reasons. First, the traditional road extraction methods considered roads as linear features, whereas in high-resolution images roads are no longer linear features but “continuous and elongated homogeneous regions” along a certain direction (Long and Zhao 2005). Second, increased spectral variance makes it difficult to separate a section of road from spectrally mixed urban land-cover types by using traditional pixel-based methods (Shaban and Dikshit 2001). The high-resolution imagery can reveal very fine details of the Earth’s surface, and geometric information on ground features becomes clearly visible. Consequently, the advent of these images benefits the recognition and understanding of roads and expands the possibilities of the extraction methods. In contrast, the highly complex land-cover details brought by the high-resolution images cause an increase in geometric noise. For instance, a section of road may include pixels with various spectral values due to differences in materials, shading, or detailed conditions (Zhou and Troy 2008). Roads are often highly complex feature-packed information with geometric parameters that are critical in road recognition. Road extraction from high-resolution imagery also faces the problem that sharply increased noise such as trees, lanes, vehicles, and shadows would influence the data analysis. Numerous studies have been conducted to explore more effective and efficient methods for road extraction from high-resolution images as a surface. For example, Yang and Wang (2007) presented an improved model for road detection based on the principles of perceptual organization and classification fusion in the human vision system. The model consisted of four levels (pixel, primitives, structures, and objects) and two additional subprocesses (automatic classification of road scenes and global integration of multiform roads). Teng and Fairbairn (2002) used a GIS data set to assist the extraction of shape descriptors, reflectance, and height above ground characteristics of the road using a fuzzy expert system and an adaptive neuro-fuzzy system. Some approaches are based on edge- and line-detecting algorithms (Long and Zhao 2005), whereas Mourad et al. (2010) proposed a rule-based classification method using multispectral segmentation with inputs of digital maps and spectral data.

Search for more...
Back to top

Use of cookies on this website

We are using cookies to provide statistics that help us give you the best experience of our site. You can find out more in our Privacy Policy. By continuing to use the site you are agreeing to our use of cookies.