Evaluation of LIDAR and photogrammetric elevation data for feature-based forest analyses

The main aim of this work is to study the surface models, generated from LIDAR data and airborne photogrammetric imagery for feature extraction approaches. In this research neither the spectral information of the images nor the intensity values of the LIDAR data are used. A novel concept, called eye-finger, is developed to simulate and analyze the surface models. To translate the feeling of the human sense of touch into the language of machine vision, the geometric and morphological features are defined and evaluated on laser-based and SGM-based canopy height models. A smart filter called non-smoothing is developed to remove the problematic pixels from the LIDAR data, while the roughness and the form of the trees remain unchanged. As complex surface model the European mixed-forests, consisting of coniferous and deciduous trees is studied. The roughness parameters such as average-based parameters showed the dependency of these surface evaluators on the age and tree type of the stand. The roughness parameters related to the standard deviation of the surface, measure the finer variations on the surface. An advanced roughness evaluator called Rfstd is developed in this research to capture minor height variations on the canopy independent from the mean height of the stand. For single tree exctraction, a novel object-oriented method, called marble-rolling is developed. The seed-objects, as the result of this process are employed for the supervised region-growing algorithm. Simultaneously, the shape characteristics of the growing segments are evaluated and optimized with morphological functions. For the characterisation of the single trees, both geometric features and morphological feature are implemented. The results of the segmentation are compared with a precise reference dataset, provided with terrestrial laser scanning and close-range photogrammetry. The evaluation of the topological relationships of the reference and target dataset provides high completeness and correctness results for the single tree extraction.