next up previous
Next: Selected Approaches for Automatic Up: Road Extraction from Aerial Previous: Road Extraction from Aerial

Introduction

  Aerial imagery is one of the standard data sources to extract topographic objects such as roads or buildings for geographic information systems (GIS). Road data in GIS are of major importance for applications like car navigation or guidance systems for police, fire service, or forwarding agencies. Because their manual extraction is time consuming, their is a need for automation.

For practical applications, there will be, at least for some time, human interaction needed, which leads to semi-automatic approaches. Algorithms relying strongly on this interaction use, for instance, road tracking [19,37] starting from a given point and a given direction. The road tracking can be based on parallel edges or on extrapolating and matching profiles in high resolution images. Other semi-automatic approaches are based on finding an optimal path between a few given points, using, for instance, dynamic programming [11,20] or the F*-algorithm [8] after detecting lines from low-resolution. When more than one image is available, it is possible to track the line in 3D. This constrains the path of the road and makes it possible to handle occlusions using robust optimization. So-called ``ziplock snakes'' [21] are another method to connect given points in the presence of obstacles.

Another way to tackle the problem is to start with fully automatic extraction and manually edit the result afterwards. Because for automatic extraction a much more specific modeling of the roads is needed, the survey focuses on the modeling with two intentions: On one hand, an overview about the approaches developed in the last decades shall be given. This is done based on a few selected approaches that have been chosen because they either were (or still are) most prominent or because they include some (at the time of their development) new important ideas that have led to an advancement of the field. To show the development, the ordering is mostly chronological for the first part of Section 2. To give further insight into the approaches, a characterization of the models and the strategies of these approaches is given (Sect. 2.1 and 2.2) and the approaches are classified (Sect 2.3).

On other hand, in Section 3 parts of a model and a strategy are presented (Sect. 3.1 and 3.2) which represent state of the art ideas of how to extract roads from aerial imagery. Additionally, approaches making use of the model and the strategy especially well are listed. Since there are parts that are useful for other object types as well, the presentation is split into more general parts and specific parts for roads. Finally in Section 3.3 issues, of which their outstanding importance has become clear only recently, are highlighted.


next up previous
Next: Selected Approaches for Automatic Up: Road Extraction from Aerial Previous: Road Extraction from Aerial
Helmut Mayer
11/22/1998