Utvärdering av metoder för procedurell generation av vegetationsfördelning : Att balansera beräkningseffektivitet och visuell trovärdighet

Within the field of video game creation and design, procedural generation (PCG) is often applied to limit the time and financial effort required to create feasible virtual environments. One particular area of PCG that plays a large role in the visual fidelity of virtual worlds is procedural vegetation generation (PVG), in which virtual worlds are enriched with vegetation. As these methods become more complex, they place a greater demand on computational hardware. This in turn limits the accessibility of video game design to those with the economy to keep up with increased hardware demands. This thesis aimed to evaluate methods of PVG in relation to their computational demand and visual fidelity of the outcome, to determine which method is best suited to achieve a balance between the two fields, and as such is best suited for video game creation for those without advanced computational hardware. Three methods of procedural vegetation generation were evaluated and compared: Perlin Noise Distribution (PND), Poisson Disk Sampling Distribution (PDD) and Field-of-Neighborhood (FON). Through the Unity game engine, a 3D terrain was created and enriched with vegetation using each of the three methods. Each method placed four different 3D models of vegetation: a large conifer tree, a medium and a small deciduous tree, and a patch of grass. The aim was to utilize the methods to create a virtual version of a mixed forest with both types of trees, common in many countries around Europe. By implementing these methods and evaluating their computational efficiency in terms of processing time, a conclusion could be drawn on the most efficient method. Through a user study consisting of 25 participants, where each participant navigated through the different worlds from a first person perspective, the visual fidelity of each method could then be evaluated and ranked according to the best visual outcome. The results suggest that PND is the most computationally efficient method, but also the worst method in relation to visual fidelity. Both PDD and FON ranked high in terms of visual fidelity, with PDD receiving slightly higher scores. PDD also had a shorter processing time and was thus the most efficient of the two methods. The conclusions drawn from the results suggest that the best method of PVG to utilize when aiming for a balance between computational demand and visual fidelity is the Poisson Disk Sampling Distribution (PDD).