RGB-D 영상을 이용한 실내 레이아웃의 빠른 복원

Abstract

This thesis presents a fast recovering method that estimates an indoor layout from a singleRGB-D image of Kinect in a second. We recover the layout, consisting of room structuralobjects and cuboid-like objects, by optimizing the combination of axes-aligned cuboidsfrom candidates. For accelerating, we develop fast polygon extraction method and generatecandidate cuboids based on extracted polygons. With these approaches, a computationof optimization process is dramatically reduced as acquring a small number of candidatecuboids. In the optimization step, we formulate the mixed integer programming modelfor selecting optimal cuboids that suitably represent the indoor scene. Experimental resultsdemonstrate that our method recovers the layout fast, also quality of result is reliable.