Um framework para distribuição, gerenciamento e renderização de vegetação em cenários virtuais massivos em tempo real

• Ano de defesa: 2020
• Autor(a) principal: Franzin, Flavio Paulus
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/26339/0013000007t80
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Ciência da Computação; UFSM; Programa de Pós-Graduação em Ciência da Computação; Centro de Tecnologia
• Palavras-chave em Português: Vegetação procedimental; Paisagens virtuais; Deformação de vegetação rasteira; GPU-Based; Cenários de larga escala; Procedimental vegetation; Virtual landscapes; Undergrowth collision simulation; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.ufsm.br/handle/1/22425

Vegetation is one of the leading graphic elements in virtual environments, as it integrates the vast majority of natural landscapes. As a result, to maximize the immersion of its users, several applications need to represent the vegetation with high fidelity. However, vegetation is composed of large amounts of plants, usually represented by complex 3D models, generating a high demand for storage and graphic processing to achieve satisfactory visual results. We present a GPU-based framework for handling vegetation in large-scale scenarios in real-time. Our proposal includes an architecture to distribute, render, and deal with plants’ interaction and moving objects. The plants are grouped and distributed procedurally by similarity, based on pre-established areas in the scenario and the terrain’s topographic characteristics. Consistent parameters are used to direct procedural distribution, providing considerable levels of artistic control. Support for manual placement of the plants is also offered, providing full control for the artist. A new approach based on vector fields and data compression is proposed for undergrowth deformation. The vector fields encode the undergrowth deformation, which can remain deformed for arbitrary periods and even permanently. Large plants are structured in a dynamic hash, which ensures efficient access so that moving objects can identify and deal with these plants. An efficient LOD system and GPU-Instancing optimize rendering performance. The proposed architecture also exploits a high GPU parallelism level, and processes are managed and invoked on demand. Also, to guarantee the solution’s applicability, efficient structures and optimizations are proposed to minimize processing and storage demands. The results show that the proposed approach can deal with vegetation in arbitrary scale scenarios, ensure a pleasant appearance, and have computational costs compatible with real-time applications.