Use this URL to cite or link to this record in EThOS:
Title: Towards real-time lighting-aware scene capture from a moving camera
Author: Jachnik, Jan
ISNI:       0000 0004 6346 972X
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
The level of complexity of maps created by monocular SLAM is on the rise. Increases in computational power have taken us from sparse feature maps to fully dense 3D reconstructions. Still, none of these are making full use of the wealth of information available from a live, monocular video feed. Aside from geometry there are the effects of lighting, reflection and shadow which are often ignored but give us vital clues into the types of surfaces being observed. We take some steps to extend the maps generated by monocular SLAM by considering real-time acquisition of surface reflectance and lighting information. In robotics, such information could be used to help determine materials, aiding object detection and semantic under- standing, thus enabling better interaction with the environment. In augmented reality lighting and reflectance information is essential to make virtual objects blend seamlessly into the real world. In this thesis we will demonstrate real-time capture of planar surface light-fields, a convenient representation to infer lighting and reflectance information. On a tangent we then investigate how sculptors manipulate geometry to alter the effects of illumination and reflectance, changing our perception and enhancing details; a technique required to bring a piece of work to life when sculpting in a medium of constant albedo. We attempt to apply some of these sculptors techniques in a mesh editing tool we call sculptural stylisation. Finally, we investigate methods for recovering the geometry of surfaces from a monocular camera in an attempt to extend our work on planar surface light-fields to work in 3D: we present various ways to generate depth-maps from a monocular video stream and detail a system to fuse them together into a consistent 3D model.
Supervisor: Davison, Andrew Sponsor: Engineering and Physical Sciences Research Council
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral