TY - GEN
T1 - Reconstructing 3D Contour Models of General Scenes from RGB-D Sequences
AU - Wang, Weiran
AU - Di, Huijun
AU - Song, Lingxiao
N1 - Publisher Copyright:
© 2022, Springer Nature Switzerland AG.
PY - 2022
Y1 - 2022
N2 - General 3D reconstruction methods use voxels, surfels, or meshes to represent the 3D model of a given scene. These surface-based methods are vulnerable to the loss of boundary details, which affects the completeness of the reconstructed model. In this paper, we focus on the boundary information of the scene and propose a novel method to reconstruct 3D models by using 3D contours extracted from input image sequences. We design a robust frame-to-model contour matching algorithm to solve the problem of finding many-to-many contour correspondences between different frames, and use contour-enhanced optimization to obtain more accurate camera poses. In order to make the reconstructed model more expressive of structural information, we propose a contour fusion algorithm that considers the connections between 3D contours. Compared with other methods which use straight lines or curve segments to reconstruct the scene model, our method can generate a more complete and regular 3D contour model with topological relationship. Experiments on several public datasets demonstrate the effectiveness of our method for both modeling and pose estimation.
AB - General 3D reconstruction methods use voxels, surfels, or meshes to represent the 3D model of a given scene. These surface-based methods are vulnerable to the loss of boundary details, which affects the completeness of the reconstructed model. In this paper, we focus on the boundary information of the scene and propose a novel method to reconstruct 3D models by using 3D contours extracted from input image sequences. We design a robust frame-to-model contour matching algorithm to solve the problem of finding many-to-many contour correspondences between different frames, and use contour-enhanced optimization to obtain more accurate camera poses. In order to make the reconstructed model more expressive of structural information, we propose a contour fusion algorithm that considers the connections between 3D contours. Compared with other methods which use straight lines or curve segments to reconstruct the scene model, our method can generate a more complete and regular 3D contour model with topological relationship. Experiments on several public datasets demonstrate the effectiveness of our method for both modeling and pose estimation.
KW - 3D Reconstruction
KW - 3D contour model
KW - Topological relationship
UR - http://www.scopus.com/inward/record.url?scp=85127093506&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-98355-0_14
DO - 10.1007/978-3-030-98355-0_14
M3 - Conference contribution
AN - SCOPUS:85127093506
SN - 9783030983543
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 158
EP - 170
BT - MultiMedia Modeling - 28th International Conference, MMM 2022, Proceedings
A2 - Þór Jónsson, Björn
A2 - Gurrin, Cathal
A2 - Tran, Minh-Triet
A2 - Dang-Nguyen, Duc-Tien
A2 - Hu, Anita Min-Chun
A2 - Huynh Thi Thanh, Binh
A2 - Huet, Benoit
PB - Springer Science and Business Media Deutschland GmbH
T2 - 28th International Conference on MultiMedia Modeling, MMM 2022
Y2 - 6 June 2022 through 10 June 2022
ER -