Automated Detection of Branch Dimensions in Woody Skeletons of Fruit Tree Canopies

Modeling the 3D canopy structure of trees provides the structural mapping capability on which to assign distributed values of light-driven physiological processes in tree canopies. We evaluate the potential of automatically extracted skeletons from terrestrial lidar data as a basis for modeling canopy structure. The automatic and species independent evaluation method for lidar data of trees is based on the SKELTRE algorithm. The SKELTRE skeleton is a graphical representation of the branch hierarchy. The extraction of the branch hierarchy utilizes a graph splitting procedure to extract the branches from the skeleton. Analyzing the distance between the point cloud points and the skeleton is the key to the branch diameter. Frequency distributions of branch length and diameter were chosen to test the algorithm performance in comparison to manually measured data and resulted in a correlation of up to 0.78 for the branch length and up to 0.99 for the branch diameter. Introduction Branch systems of trees are the result of ramification and branch elongation processes that occur, outside the tropics, in an annual rhythm. The pattern of branch elongation and radial diameter growth can reveal the dendritic growth history of trees with the same accuracy as growth-ring chronologies of the trunk (Roloff, 1986). The annual rhythm of growth conditions is reflected in the branching pattern of trees and will finally be represented in the skeleton. Furthermore, the dendrochronological patterns are closely correlated to other structural quantities of tree canopies like appending leaf or woody biomass (Niklas, 1994). Allometric equations were established on this basis for many tree species in order to derive the amount of woody biomass (Bartelink, 1997), leaf biomass (Burger, 1945) or distribution of leaf biomass in space (Fleck, 2002) from more easily measured features such as trunk or branch diameters. 3D-canopy light modeling depends on such spatial information as the distribution of biomass and is the key to a number of physiological processes in the canopy that express the vitality and performance of trees (Fleck et al., 2004). PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Mar ch 2011 229 Alexander Bucksch is at the Delft Institute of Earth Observation and Space Systems (DEOS), Delft University of Technology, Kluyverweg 1, 2629 HS Delft, The Netherlands, ([email protected]). Stefan Fleck is with the Department Environmental Control, North-West German Forest Research Institute, Graetzelstrasse 2, 37079 Göttingen, Germany, and formerly with Plant Ecology, Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, 37073 Göttingen, Germany. Photogrammetric Engineering & Remote Sensing Vol. 77, No. 3, March 2011, pp. 229–240. 0099-1112/11/7703–0229/$3.00/0 © 2011 American Society for Photogrammetry and Remote Sensing Automated Detection of Branch Dimensions in Woody Skeletons of Fruit Tree Canopies Alexander Bucksch and Stefan Fleck From a remote sensing viewpoint, the automated assessment of branch dimensions in the canopy is unprecedented. Terrestrial laser scanners measure thousands of distances per second between the instrument and its surroundings at regular horizontal and vertical angles (Shan and Toth, 2008) in order to represent a high-resolution 3D point cloud. Thus, terrestrial lidar enables the measurement of the complete three-dimensional structure of the branching system. This branching information can be made available to modelers in biology and forestry. An automated evaluation procedure would make it possible to overcome tedious measurement procedures or inaccurate estimations of the branching system. In one sense, laser scanning produces a discrete surface sampling of a real world object and represents it as a point cloud. Single scans must be made from different scanning positions to render the whole object. The scans have to then be co-located into one common coordinate system. The process of aligning scans into a common coordinate system is called registration. The drawback of the registration procedure is that regularity in the scan data vanishes, and the point cloud becomes unorganized. Furthermore, the height distribution of the reference points to perform the registration is critical (Henning and Radtke, 2006) and influences the registration result. The study of unorganized point clouds as an object representation and the possible information to be extracted from point clouds is an area of active research. Although the majority of research has focused on the extraction of surface parameters from the point cloud, e.g., Pfeifer et al. (2004) and Henning and Radtke (2008), this paper describes a new method to reveal the branching information using the example of leafless apple trees. The fully automatic approach presented here does not depend on species information, such as allometric relationships. Obtaining the branching system from unorganized point clouds (Figure 1) can help in various point cloud applications. The target application of this paper is the extraction of the branch length and diameter from laser-scanned orchard trees. The SKELTRE-skeleton used in this research represents the tree’s branching system as a graph. Such a graph consists of vertices which are connected by edges. Every vertex corresponds to a distinct part of the point cloud and is embedded into the centre of the corresponding point cloud part. The edges are assumed as straight connections between the embedded vertices. The skeleton extraction from a point cloud faces several algorithmic challenges, such as centeredness, topological