ABSTRACT
Accurately mapped roads are essential for a wide variety of applications including studies related to community and transportation planning, and water resource and wildlife management. Road extraction techniques have been applied to many different types of remote sensing data over recent decades and include both automated and semiautomated approaches. Development during this period has gone from low to high spatial resolution image data and more recently has extended into data fusion. Mena (2003) and Quackenbush (2004) reported many different types of remote sensed data used in linear feature extraction including multi-and hyperspectral imagery, synthetic aperture radar (SAR), and light detection and ranging (LiDAR) sources. Studies seeking to define roads from a single data source frequently face both spectral and spatial challenges. Road extraction is complicated by geometric noise-for example, because of cars or shadows-that increase spectral variance and decrease homogeneity in radiometry along a road (Weng 2012). Another problematic area is the similarity in radiometry between roads and other impervious features such as buildings and parking lots (Péteri and Ranchin 2007). Mena (2003) and Jung et al. (2006) reported semiautomatic algorithms that mitigate some of these challenges by guiding road extraction using existing geographic information system (GIS) databases or manual inputs; however, researchers have also explored factors related to successfully automating road extraction. Gecen and Sarp (2008) extracted roads automatically in an urban area from four satellite images with different ground sample distance-that is, IKONOS, QuickBird, ASTER, and Landsat-to study the impact of spatial resolution on
CONTENTS
9.1 Introduction ........................................................................................................................ 155 9.2 Overview of Recent LiDAR Applications in Road Extraction ..................................... 156 9.3 Road Extraction Using LiDAR Remote Sensing ............................................................ 158
9.3.1 Defining Road Clusters ......................................................................................... 158 9.3.1.1 Classification Framework....................................................................... 158 9.3.1.2 Algorithms Used for Road Identification ............................................ 160
9.3.2 Generating Road Networks .................................................................................. 162 9.3.2.1 Road Classification Refinement ............................................................ 162 9.3.2.2 Centerline Extraction .............................................................................. 163
9.4 Discussion and Conclusion .............................................................................................. 164 References ..................................................................................................................................... 165
automatic road extraction. Although accuracy tended to increase with spatial resolution, Gecen and Sarp (2008) found that objects with spectral property similar to roads, such as house roofs, were frequently misclassified into road clusters at all resolutions. Baumgartner et al. (1999) extracted roads from digital aerial imagery and found road edges contained significant noise caused by adjacent features such as trees and roofs. Amini et al. (2002) proposed an object-based approach for automatic extraction of major roads in a large-scale image, generating a binary road map through image segmentation. The project presented by Amini et al. (2002) suffered a problem that is common when working with impervious land cover types, in that their map characterized some elevated objects as roads.
