Deep fundamental matrix estimation without correspondences
Research output: Contribution to journal › Conference article › Research › peer-review
Standard
Deep fundamental matrix estimation without correspondences. / Poursaeed, Omid; Yang, Guandao; Prakash, Aditya; Fang, Qiuren; Jiang, Hanqing; Hariharan, Bharath; Belongie, Serge.
In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2019, p. 485-497.Research output: Contribution to journal › Conference article › Research › peer-review
Harvard
APA
Vancouver
Author
Bibtex
}
RIS
TY - GEN
T1 - Deep fundamental matrix estimation without correspondences
AU - Poursaeed, Omid
AU - Yang, Guandao
AU - Prakash, Aditya
AU - Fang, Qiuren
AU - Jiang, Hanqing
AU - Hariharan, Bharath
AU - Belongie, Serge
N1 - Publisher Copyright: © Springer Nature Switzerland AG 2019.
PY - 2019
Y1 - 2019
N2 - Estimating fundamental matrices is a classic problem in computer vision. Traditional methods rely heavily on the correctness of estimated key-point correspondences, which can be noisy and unreliable. As a result, it is difficult for these methods to handle image pairs with large occlusion or significantly different camera poses. In this paper, we propose novel neural network architectures to estimate fundamental matrices in an end-to-end manner without relying on point correspondences. New modules and layers are introduced in order to preserve mathematical properties of the fundamental matrix as a homogeneous rank-2 matrix with seven degrees of freedom. We analyze performance of the proposed models using various metrics on the KITTI dataset, and show that they achieve competitive performance with traditional methods without the need for extracting correspondences.
AB - Estimating fundamental matrices is a classic problem in computer vision. Traditional methods rely heavily on the correctness of estimated key-point correspondences, which can be noisy and unreliable. As a result, it is difficult for these methods to handle image pairs with large occlusion or significantly different camera poses. In this paper, we propose novel neural network architectures to estimate fundamental matrices in an end-to-end manner without relying on point correspondences. New modules and layers are introduced in order to preserve mathematical properties of the fundamental matrix as a homogeneous rank-2 matrix with seven degrees of freedom. We analyze performance of the proposed models using various metrics on the KITTI dataset, and show that they achieve competitive performance with traditional methods without the need for extracting correspondences.
KW - Deep learning
KW - Epipolar geometry
KW - Fundamental matrix
KW - Stereo
UR - http://www.scopus.com/inward/record.url?scp=85061693720&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-11015-4_35
DO - 10.1007/978-3-030-11015-4_35
M3 - Conference article
AN - SCOPUS:85061693720
SP - 485
EP - 497
JO - Lecture Notes in Computer Science
JF - Lecture Notes in Computer Science
SN - 0302-9743
T2 - 15th European Conference on Computer Vision, ECCV 2018
Y2 - 8 September 2018 through 14 September 2018
ER -
ID: 301824797