Secondary abstract: |
This thesis proposes a new approach for large-scale forest visualization with geometrically diverse trees. First, we propose a new approach for tree geometry synthesis that is based on particle flow simulation. The simulation constructs the trees in the direction from the leaves towards the trunk. Therefore, the final tree structure is defined by a tree crown envelope. This approach is later adapted for tree synthesis on graphics processing units to enable interactive tree geometry synthesis of diverse trees during visualization without noticeable delays. To visualize a larger number of trees, the nearest trees are generated at a higher level of detail, while the distant ones are generated directly at a lower level of detail, without the need for post-processing of existing geometric data. However, all the trees cannot be synthesized simultaneously, therefore, we propose a new approach for volumetric tree crown visualization of distant trees within the graphics pipeline. The proposed approach generates the leaves implicitly during a ray casting algorithm on a graphics processing unit, without requirements for intermediate geometric representation. This approach also considers the tree crown envelope representation, and, thus, enables achieving the same tree crown appearance in respect to the tree crown based on geometric representation. In the continuation, both approaches are combined within a new comprehensive framework for interactive on-the-fly generation and visualization of large-scale forests, while navigating through the scene. In the results, tree synthesis and visualization speed are evaluated, and the suitability of the proposed approach is demonstrated for interactive large scale forest visualization. Currently available hardware allows generating trees consisting of 400,000 internodes in less than 25 ms, which enables generating larger numbers of trees between individual frames. Additionally, generating trees directly at a specific level of detail allows faster generation and visualization of the trees at the lower level of detail. Finally, we demonstrate that the procedural volumetric tree crown visualization of distant trees achieves higher rendering rates in relation to the visualization based on existing geometric representation. |