This paper contains two parts: a short description of our tree finder, and a concept for the automatic evaluation of the results. The basic idea of our delineation approach is, that the coarse structure of the crown can be approximated with the help of a sphere or an ellipsoid. This assumption is true, if the fine structure of the crown is ignored and the coarse structure is revealed in an appropriate level of the multi-scale representation of the surface model. However, this scale level is unknown, because it is correlated with the unknown diameter of the crown. The proposed solution for this chicken-and-egg problem is to investigate a wide range of scale levels, and to subsequently select the best hypothesis for a crown from all these scale levels. This selection must be based on an internal evaluation of the obtained results. The evaluation concept is based on the classification of the topological relations between the crown’s projection from different data sets onto the ground level. In general, eight different topological relations exist in 2D space: disjoint, touch, overlap, covers, contains, contained by, covers, and covered by. These topological relations can be subdivided into two clusters C1 and C2, where the C1 cluster includes the relations disjoint, touch and C2 the other ones apart from the overlap relation. The overlap relation is between these two clusters, it can be divided into weak-overlap (C1) and strong-overlap (C2). The motivation behind this partitioning is that the relations in C1 are similar to disjoint, and in C2 to equal. In our approach the type of the relation cluster is used in the internal evaluation to decide whether hypotheses from different scale levels are identical. In the external evaluation, we use the relation cluster to decide which trees from a reference data set are identical to automatically extracted trees.
[1]
S. Reutebuch,et al.
BAYESIAN OBJECT RECOGNITION FOR THE ANALYSIS OF COMPLEX FOREST SCENES IN AIRBORNE LASER SCANNER DATA
,
2002
.
[2]
R. Reulke,et al.
Remote Sensing and Spatial Information Sciences
,
2005
.
[3]
Bernd-Michael Wolf,et al.
Automatische Extraktion von Bäumen aus Fernerkundungsdaten
,
2003
.
[4]
Bernd-M. Straub,et al.
A TOP-DOWN OPERATOR FOR THE AUTOMATIC EXTRACTION OF TREES-CONCEPT AND PERFORMANCE EVALUATION
,
2003
.
[5]
Åsa Persson,et al.
Detecting and measuring individual trees using an airborne laser scanner
,
2002
.
[6]
M. Egenhofer.
Categorizing Binary Topological Relations Between Regions, Lines, and Points in Geographic Databases
,
1998
.
[7]
Tomas Brandtberg,et al.
Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis
,
1998,
Machine Vision and Applications.
[8]
Richard J. Pollock,et al.
Model-based approach to automatically locating tree crowns in high spatial resolution images
,
1994,
Remote Sensing.
[9]
Mathias Schardt,et al.
ASSESSMENT OF FOREST PARAMETERS BY MEANS OF LASER SCANNING
,
2002
.
[10]
Robert J. Woodham,et al.
The automatic recognition of individual trees in aerial images of forests based on a synthetic tree crown image model
,
1996
.
[11]
Tony Lindeberg,et al.
Scale-Space Theory in Computer Vision
,
1993,
Lecture Notes in Computer Science.