Dense Pixel-wise Micro-motion Estimation of Object Surface by using Low Dimensional Embedding of Laser Speckle Pattern

Ryusuke Sagawa ("AIST, Japan")*, Yusuke Higuchi (Kyushu University), Hiroshi Kawasaki (Kyushu univ.), Ryo Furukawa (Hiroshima city univ.), Takahiro Ito (AIST)

Keywords: Computational Photography, Sensing, and Display

Abstract: This paper proposes a method of estimating micro-motion of an object at eachpixel that is too small to detect under a common setup of camera andillumination. The method introduces an active-lighting approach to make themotion visually detectable. The approach is based on speckle pattern, which isproduced by the mutual interference of laser light on object's surface andcontinuously changes its appearance according to the out-of-plane motion of thesurface. In addition, speckle pattern becomes uncorrelated with large motion. Tocompensate such micro- and large motion, the method estimates the motionparameters up to scale at each pixel by nonlinear embedding of the specklepattern into low-dimensional space. The out-of-plane motion is calculated bymaking the motion parameters spatially consistent across the image. In theexperiments, the proposed method is compared with other measuring devices toprove the effectiveness of the method.

SlidesLive

Similar Papers

Learning Multi-Instance Sub-pixel Point Localization
Julien Schroeter (Cardiff University)*, Tinne Tuytelaars (KU Leuven), Kirill Sidorov (Cardiff University), David Marshall (Cardiff University)
Chromatic Aberration Correction Using Cross-Channel Prior in Shearlet Domain
Kunyi Li (Tsinghua University), Xin Jin (Tsinghua University)*
Deep Snapshot HDR Imaging Using Multi-Exposure Color Filter Array
Takeru Suda (Tokyo Institute of Technology), Masayuki Tanaka (Tokyo Institute of Technology), Yusuke Monno (Tokyo Institute of Technology)*, Masatoshi Okutomi (Tokyo Institute of Technology)