Development of an object classification workflow applied on tree inventory

Kurzfassung

The German Government requires every ten years a forestry inventory to obtain data for monitoring of the wood stock and the condition of the woods. At present, the inventory is carried out collecting random samples manually to represent the forest condition. For this purpose, an inventory team requires two years to complete the inventory project, which is a big �nancial and resource e�ort. Thus, to decrease time e�ort and manpower and to improve the tree inventory an automatic optical method is introduced in this work. The method employs a stereo camera system, which measures the positions of the trees and its geometry in the 3D space. The multi-sensor Integral Positioning System (IPS) is developed by the German Aerospace Center (DLR), Institute of Optical Systems. IPS is a compact device that combines a stereo camera with an inertial measurement unit (IMU) and a GPS receiver to provide information about the attitude and position in 3D space. The contribution of this work is twofold: A robust algorithm for detection and segmentation of tree trunks in stereo images is developed. Afterwards, the trunk hypotheses are employed for spatial clustering of the trees in 3D space. The framework is evaluated on realistic data set, which is considered to be complex because of bushed and di�cult illumination conditions.