Multiple light source optical flow for translations and rotations in the image plane

The classic optical flow constraint equation is accurate under conditions of translation and distant light sources, but becomes inaccurate under conditions where the object may rotate or deform. The inaccuracies are generally a result of the changing intensities at points under rotation or deform...

Full description

Bibliographic Details
Main Author: Takagi, Shingo Jason
Language:English
Published: 2009
Online Access:http://hdl.handle.net/2429/11457
id ndltd-LACETR-oai-collectionscanada.gc.ca-BVAU.2429-11457
record_format oai_dc
spelling ndltd-LACETR-oai-collectionscanada.gc.ca-BVAU.2429-114572014-03-14T15:44:55Z Multiple light source optical flow for translations and rotations in the image plane Takagi, Shingo Jason The classic optical flow constraint equation is accurate under conditions of translation and distant light sources, but becomes inaccurate under conditions where the object may rotate or deform. The inaccuracies are generally a result of the changing intensities at points under rotation or deformation. The changing intensities of these points are associated with the changing surface gradients. We investigate a novel approach to multiple light source optical flow under known reflectance properties. This novel approach is specialized for translation and for rotation around axes which are parallel to the optical axis. The assumption that the object is moving under translation, rotation, or a combination of translation androtation in the image plane allows us to introduce a physical surface area constraint. Even with this additional constraint, however, the problem still remains locally underdetermined. At each time frame, our multiple light source optical flow approach assumes that three images are acquired from the same viewpoint, but under three different illumination conditions. Photometric stereo determines many of the coefficients in our underdetermined system, which has six equations in seven unknowns at each pixel in the image. Two of the unknowns are the optical flow components. Another three of the unknowns are the total derivatives of the three intensities with respect to time. The last two are the total derivatives of the surface gradients with respect to time. A variety of local regularization methods were investigated to select optical flow estimates which best matched the known motion fields. All the experimental results for this approach were obtained from synthetic data, in which the motion fields were known. 2009-07-29T18:05:18Z 2009-07-29T18:05:18Z 2001 2009-07-29T18:05:18Z 2001-05 Electronic Thesis or Dissertation http://hdl.handle.net/2429/11457 eng UBC Retrospective Theses Digitization Project [http://www.library.ubc.ca/archives/retro_theses/]
collection NDLTD
language English
sources NDLTD
description The classic optical flow constraint equation is accurate under conditions of translation and distant light sources, but becomes inaccurate under conditions where the object may rotate or deform. The inaccuracies are generally a result of the changing intensities at points under rotation or deformation. The changing intensities of these points are associated with the changing surface gradients. We investigate a novel approach to multiple light source optical flow under known reflectance properties. This novel approach is specialized for translation and for rotation around axes which are parallel to the optical axis. The assumption that the object is moving under translation, rotation, or a combination of translation androtation in the image plane allows us to introduce a physical surface area constraint. Even with this additional constraint, however, the problem still remains locally underdetermined. At each time frame, our multiple light source optical flow approach assumes that three images are acquired from the same viewpoint, but under three different illumination conditions. Photometric stereo determines many of the coefficients in our underdetermined system, which has six equations in seven unknowns at each pixel in the image. Two of the unknowns are the optical flow components. Another three of the unknowns are the total derivatives of the three intensities with respect to time. The last two are the total derivatives of the surface gradients with respect to time. A variety of local regularization methods were investigated to select optical flow estimates which best matched the known motion fields. All the experimental results for this approach were obtained from synthetic data, in which the motion fields were known.
author Takagi, Shingo Jason
spellingShingle Takagi, Shingo Jason
Multiple light source optical flow for translations and rotations in the image plane
author_facet Takagi, Shingo Jason
author_sort Takagi, Shingo Jason
title Multiple light source optical flow for translations and rotations in the image plane
title_short Multiple light source optical flow for translations and rotations in the image plane
title_full Multiple light source optical flow for translations and rotations in the image plane
title_fullStr Multiple light source optical flow for translations and rotations in the image plane
title_full_unstemmed Multiple light source optical flow for translations and rotations in the image plane
title_sort multiple light source optical flow for translations and rotations in the image plane
publishDate 2009
url http://hdl.handle.net/2429/11457
work_keys_str_mv AT takagishingojason multiplelightsourceopticalflowfortranslationsandrotationsintheimageplane
_version_ 1716652271371026432