Search International and National Patent Collections

1. (WO2011136276) CREATION METHOD AND CREATION DEVICE OF THREE-DIMENSIONAL OBJECT RECOGNITION-USE IMAGE DATABASE

Pub. No.:    WO/2011/136276    International Application No.:    PCT/JP2011/060277
Publication Date: Fri Nov 04 00:59:59 CET 2011 International Filing Date: Thu Apr 28 01:59:59 CEST 2011
IPC: G06T 7/00
G06T 1/00
Applicants: OSAKA PREFECTURE UNIVERSITY PUBLIC CORPORATION
公立大学法人大阪府立大学
OLYMPUS CORPORATION
オリンパス株式会社
KASHIWAGI, Takahiro
柏木 隆宏
TOYAMA, Takumi
外山 託海
KISE, Koichi
黄瀬 浩一
FURUHASHI, Yukihito
古橋 幸人
MINE, Taiji
峯 泰治
Inventors: KASHIWAGI, Takahiro
柏木 隆宏
TOYAMA, Takumi
外山 託海
KISE, Koichi
黄瀬 浩一
FURUHASHI, Yukihito
古橋 幸人
MINE, Taiji
峯 泰治
Title: CREATION METHOD AND CREATION DEVICE OF THREE-DIMENSIONAL OBJECT RECOGNITION-USE IMAGE DATABASE
Abstract:
Disclosed is a creation method for performing selection of local feature values to be registered in a three-dimensional object recognition-use image database to reduce memory capacity while suppressing decrease in the rate of recognition. The disclosed creation method of a three-dimensional object recognition-use image database comprises a step in which a plurality of images of a three-dimensional object as viewed from different viewpoints are input, local features of each image are extracted, and each is represented as a feature vector; a step for generating a plurality of collections comprising feature vectors wherein each collection is a collection representing local features of the same spot of the object as viewed from a series of adjacent viewpoints, and generating a plurality of subspaces representing characteristics of each of the collections; a step for adding an ID of the object to each subspace and registering thereof in a database; and a step for extracting a plurality of feature vectors representing local features of a search question when an image of some object as viewed from some viewpoint is supplied as the search question, determining each of the subspaces which are most similar to each feature vector, performing aggregation processing for the object ID that has been added to each subspace, and acquiring the object most similar to the search question.