지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 박순용
- 발행연도
- 2023
- 저작권
- 경북대학교 논문은 저작권에 의해 보호받습니다.
이용수1
이 논문의 연구 히스토리 (2)
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
In this paper, we propose a novel computer vision method for 3D reconstruction from urban multi-view satellite images. To do that, we approach monocular height estimation and fusion instead of using conventional RPC or Multi-view stereo methods.
By employing a monocular depth deep network, the proposed methods are substituting scale-invariant loss function with boundary-aware loss function and combining filtering with segmented rooftop mask for estimating higher accuracy height maps. Moreover, we perform affine transformation for reconstructing height maps from each camera coordinate system to reference coordinate system on fusing all height maps. After that, all of the transformed 3D points are projected to the image plane of each camera coordinate system, then by adapting median filtering, we finally can get final fusion-based height maps. The maps achieve higher accuracy than the monocular height estimation method and much denser 3D points than feature-based multi-view maps.
In training and evaluating, we used the DFC19 dataset covering Jacksonville, FL, and Omaha, NE. Additionally, for testing the out-domain on a new city, we also evaluate the ATL dataset covering Atlanta, GA. Experimental results show that the proposed method can be used as a new framework for 3D reconstruction from multi-view satellite images.
By employing a monocular depth deep network, the proposed methods are substituting scale-invariant loss function with boundary-aware loss function and combining filtering with segmented rooftop mask for estimating higher accuracy height maps. Moreover, we perform affine transformation for reconstructing height maps from each camera coordinate system to reference coordinate system on fusing all height maps. After that, all of the transformed 3D points are projected to the image plane of each camera coordinate system, then by adapting median filtering, we finally can get final fusion-based height maps. The maps achieve higher accuracy than the monocular height estimation method and much denser 3D points than feature-based multi-view maps.
In training and evaluating, we used the DFC19 dataset covering Jacksonville, FL, and Omaha, NE. Additionally, for testing the out-domain on a new city, we also evaluate the ATL dataset covering Atlanta, GA. Experimental results show that the proposed method can be used as a new framework for 3D reconstruction from multi-view satellite images.
목차
제 1 장 서론 1제 2 장 관련 연구 52.1. 위성 영상 처리 62.2. 단안 깊이 추정 7제 3 장 본론 123.1. 데이터 및 데이터 전처리 143.2. 지붕 분할 영역의 필터링 183.3. 경계 손실 함수 203.4. 단안 높이 추정의 융합 23제 4 장 실험 구성 및 결과 304.1. 단안 높이 추정 결과 314.2. 단안 높이의 융합 결과 37제 5 장 결론 43참고 문헌 44영문 초록 50
최근 본 자료
댓글(0)
0