[Back]

A. Grafl, M. Hollaus, L. Eysn, W. Karel, N. Pfeifer:
"A novel approach for forest road detection based on ALS data and orthophotos";
Poster: 13th International Conference on LiDAR Applications for Assessing Forest Ecosystems (SilviLaser 2013), Beijing, China; 2013-10-09 - 2013-10-11.

Up-to-date knowledge about the position and the geometric properties of forest roads is essential for forest management and monitoring activities. Furthermore, a complete and topologically correct forest road network is a crucial requirement for any track routing systems which is an important task for optimizing the timber harvesting chain. In the last decade small footprint airborne laserscanning (ALS) data proofed to be an excellent data source for deriving detailed terrain models (DTMs) even in tree covered areas, which provide the main input data source for forest road detection.

This contribution describes a novel approach for a semi-automatic extraction of forest roads based on ALS digital terrain models (DTMs) and digital aerial orthophotos, in which the use of orthophotos is optional. As the developed algorithm is intended for country-wide operational forest road detection a rasterized ALS DTM with a spatial resolution of 1 m is used which is a standard topographic model derived from commercial country-wide ALS acquisitions in Austria. Additional to the ALS data the developed algorithms allows the optional integration of digital aerial orthophotos, which are acquired in a regular time interval of three years in Austria.

The main assumption of the road detection using ALS data is that a road is characterized with lower slopes in relation to their surroundings. In addition to this geometric property the radiometric information from orthophotos is used for rather flat terrains with little slope differences between the road and the surrounding areas. In a first processing step different slope intervals (e.g. 0-30%; 10-40%) are defined and saved as binary images. Similar to the slope the spectral bands of the orthophotos are binarized in potential road areas. Each of these binarized images contains potential road areas. In the second processing step, these areas are thinned to a line (pixel width 1). A watershed detection algorithm is applied for this on each binary layer. After morphologic operations of the extracted watersheds the individual watersheds are weighted and summed up resulting in a weighted graph. Applying Dijkstra´s algorithm shortest paths are detected from starting vertices to destination vertices based on the weighted graph. At the current stage of the developed algorithms the starting and ending vertices are manually defined by an operator, whereas the lengths between start and end point can vary between several meters up to several kilometres depending on the complexity of the road network. For finalizing the road centre lines curve adjustments and smoothing operations are applied to the detected shortest paths in an automatic way. Based on the derived road centre lines the curve radii and the mean with and slope are automatically determined based on the ALS DTM.
The developed approach was applied to different study areas (i.e. with different slope and crown cover conditions) in Austria. The extracted roads were validated with manually digitized road networks and showed a very high degree of completeness and correctness. Furthermore, the approach has shown high robustness against varying DTM quality originating from different ALS terrain point densities. Additionally, the optional integration of the digital aerial images has shown its expected benefit especially for roads located in flat terrain. An advantage of the manual interaction is the topologically correctness and completeness of the derived road network.