Author

Fatima Kuehn

Date of Award

January 2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Ronald Marsh

Abstract

Virtual reality (VR) is a very promising technology with many compelling industrial applications. As many advancements have been made recently to deploy and use VR technology in virtual environments, they are still less mature to be used to render real environments. The current VR systems settings, which are developed for virtual environments rendering, fail to adequately address the challenges of capturing and displaying real-world virtual reality that these systems entail. Before these systems can be used in real life settings, their performance needs to be investigated, more specifically, depth perception and how distances to objects in the rendered scenes are estimated. The perceived depth is influenced by Head Mounted Displays (HMD) that inevitability decrease the virtual content’s depth perception. Distances are consistently underestimated in virtual environments (VEs) compared to the real world. The reason behind this underestimation is still not understood. This thesis investigates another version of this kind of system, that to the best of authors knowledge has not been explored by any previous research. Previous research used a computer-generated scene. This work is examining distance estimation in real environments rendered to Head-Mounted Displays, where distance estimations is among the most challenging issues that are still investigated and not fully understood.This thesis introduces a dual-camera video feed system through a virtual reality head mounted display with two models: a video-based and a static photo-based model, in which, the purpose is to explore whether the misjudgment of distances in HMDs could be due to a lack of realism, or not, with the use of a real-world scene rendering system. Distance judgments performance in the real world and these two evaluated VE models were compared using protocols already proven to accurately measure real-world distance estimations. An improved model based on enhancing the field of view (FOV) of the displayed scenes to improve distance judgements when displaying real-world VR content to HMDs was developed; allowing to mitigate the limited FOV, which is among the first potential causes of distance underestimation, specially, the mismatch of FOV between the camera and the HMD field of views. The proposed model is using a set of two cameras to generate the video instead of hundreds of input cameras or tens of cameras mounted on a circular rig as previous works from the literature. First Results from the first implementation of this system found that when the model was rendered as static photo-based, the underestimation was less as compared with the live video feed rendering. The video-based (real + HMD) model and the static photo-based (real + photo + HMD) model averaged 80.2% of the actual distance, and 81.4% respectively compared to the Real-World estimations that averaged 92.4%. The improved developed approach (Real + HMD + FOV) was compared to these two models and showed an improvement of 11%, increasing the estimation accuracy from 80% to 91% and reducing the estimation error from 1.29% to 0.56%. This thesis results present strong evidence of the need for novel distance estimation improvements methods for real world VR content systems and provides effective initial work towards this goal.

Share

COinS