Forest Modeling using Airborne Lidar

Building up forest inventories for performing forest management is currently done manually through field cruises and interpretation of the aerial imagery, which is a labor-intensive task. Airborne lidar technology is capable of building a point cloud off the top surface of the ground objects. Modeling urban areas from the lidar has been successfully accomplished. However, for forest modeling, trees are less conformed to a predefined geometric shape and size. That has made the previous work incapable of providing sufficiently robust approaches. Segmentation of individual trees within the point cloud is the important starting point. Previous work has mainly been based on identifying local maxima and trying to isolate trees starting off those local maxima; or identifying local minima in order to delineate tree crown borders; or a combination of the two approaches. However, both strategies are prone to pitfalls depending on the forest data. The current thesis work starts by trying out a novel tree segmentation approach. The new method is based on the fact that any two distinct objects on the ground, if distinguishable from the top view, are very likely to have horizontal gaps between them. The proposed approach tries to detect the gaps and extend them in order to segment trees apart within the lidar data. The initial result is very promising. For this thesis, we plan to complete the proposed method and test it over more varying forest data. Extracting other tree properties from the isolated trees will be another step toward modeling the forest. Due to the large scale of the forest data, applying high performance computing methods for efficient processing is inevitable and will be a part of this thesis. Theoretical analysis of the algorithms devised in this work, whether for single-processor or multiple ones, may also constitute another part of the thesis.