Recreating Past Environments in Virtual Reality

Abstract

Reconstructing environments from a collection of images or videos could allow people to revisit locations from their past. Existing reconstruction methods impose strict constraints on input images, and require more data than most people have at their disposal. This work provides a pipeline that takes a short video from a user’s past and creates a virtual environment that they can experience in VR. In particular, we at Photogrammetry and Neural Radiance Fields as ways of representing 3D environments that can be converted into meshes and exported to VR. A technical evaluation compares the two methods and one is selected for use in our pipeline. Through a human-subjects study, we found that experiencing past environments as immersive walkable spaces, when compared to simply watching a video, improves users’ sense of presence and ability to recall memories of the space.