Towards photorealistic rendering for mixed reality on mobile devices

• Main Author: Esguerra, Christian Terrence B.
• Format: text
• Language: English
• Published: Animo Repository 2018
• Subjects: Cell phones; Smartphones; Mixed reality
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/5633
• Institution: De La Salle University

This thesis aims to incorporate photorealistic rendering to mixed reality applications on mobile devices. Photorealistic rendering refers to the process of seamlessly merging virtual objects into real-world scenes. Previous research have been conducted to implement photorealistic rendering in mixed reality by capturing scene geometry information, environment lighting, and rendering. However, most of these systems require complex setups, heavy user input, and powerful hardware. Furthermore, very few of these systems have considered implementation for mobile devices especially through global illumination as mobile devices present limited computational power compared to modern desktop machines. In addition, there are a lot of physically based rendering approaches in computer graphics with differing qualities of photorealism. It is difficult to know which approach produce the most photorealistic results as photorealism is mostly subjective and is dependent on the viewer's perception. This thesis has implemented a system to address these issues. The system accepts two inputs, a panoramic photo of the real scene which is converted into an environment map and a virtual 3D scene. These are then rendered using several physically based rendering approaches and global illumination in the form of path tracing. These rendering approaches are then evaluated in terms of photorealistic quality in the form of a qualitative survey and performance through performance testing on mobile devices to gather which physically based rendering approaches produce the best results. Results show that it is indeed possible to implement photorealistic rendering on a mobile device that employs physically based rendering and global illumination and that there are indeed better rendering approaches that produce better photorealistic quality than others. Future AR applications can benefit from the findings of this research as it was able to present the most photorealistic physically based rendering BRDFs and the feasibility of implementing path tracing global illumination on limited mobile hardware.