2020 Fiscal Year Final Research Report
Visualization of time-lapse imaging by 3D reconstruction on cultural heritage buildings
| Project/Area Number |
19K13416
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 03060:Cultural assets study-related
|
|---|
| Research Institution | University of Tsukuba |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2019-04-01 – 2021-03-31
|
| Keywords | 画像マッチング / 文化遺産建造物 / 自己符号化器 / クラウドソーシング / 深層学習 |
|---|
| Outline of Final Research Achievements |
In this research, a method to improve the matching accuracy between the image of a cultural heritage building taken 50 to 100 years ago and the current image was proposed. A solution to that problem was proposed by using an autoencoder and a Guided Matching method. By using the self-encoder, the problem that image features different from the prototype are detected at the same position such as deterioration or damage due to aging of the building, or remodeling / modification has been eliminated. By re-searching for the positive correspondence obtained by the Guided Matching method, the false matching problem due to the symmetry of the building was solved.
|
| Free Research Field |
コンピュータビジョン
|
| Academic Significance and Societal Importance of the Research Achievements |
文化遺産建造物の５０年～１００年前の写真は，様々な角度からの多数の画像は存在しない．任意の年代の疎らな角度からの数枚の画像が存在する文化遺産建造物が多い．本研究は，１枚の昔の画像を情報源として，現在の画像情報とつなぎ合わせることで，時系列変化映像の生成を目的とした．従って，本研究成果によって，多くの文化遺産建造物へ適用を可能とし，過去と現在の比較を視覚的に把握することを可能とした．この知見は，考古学の分野において，文化遺産建造物の保存を目的とした活動の貴重な資料となり，多大な貢献が期待できる．
|
