“Retexturing under Self-Occlusion using Hierarchical Markers” by Miyagawa, Fukuhara, Narita, Ogata and Morishima

  • ©

Conference:


Type(s):


Entry Number: 27

Title:

    Retexturing under Self-Occlusion using Hierarchical Markers

Presenter(s)/Author(s):



Abstract:


    Marker-based retexturing is to superimpose the texture on a target object by detecting and identifying markers from within the captured image. We propose a new marker that can be identified under a large deformation that involves self-occlusion, which was not taken into consideration in the following markers. Bradley et al. [Bradley et al. 2009] designed the independent markers, but it is difficult to recognize them under complicated occlusion. Scholz et al.[Scholz and Magnor 2006] created a circular marker with a single color selected from multiple colors. They created ID corresponding to the alignment of colors by one marker and the markers around it and identified by placing the marker so that the ID would be unique. However, when some markers are covered by self-occlusion, the positional relationship of the markers appears to be different from the original, so markers near the self-occlusion are failed to identify. Narita et al. [Narita et al. 2017] considered self-occlusion by improving the identification algorithm. They succeeded in improving the accuracy of identification by creating triangle meshes whose vertices are the center of gravity of markers and assuming that they are close to a right isosceles triangle. However, since outliers are removed using angles, identification of markers may fail in the case of the object that is likely to be deformed in the shear direction like a cloth. Therefore, we considered self-occlusion by designing hierarchical markers so that they can be referred to in a global scope. We designed a color based marker for easy recognition even at low resolution.

Keyword(s):



Acknowledgements:


    This work was supported by JSTACCEL Grant Number JPMJAC1602, Japan.


PDF:



ACM Digital Library Publication:



Overview Page: