TY - GEN

T1 - Robustifying relative orientations with respect to repetitive structures and very short baselines for global SfM

AU - Wang, Xin

AU - Xiao, Teng

AU - Gruber, Michael

AU - Heipke, Christian

N1 - Funding Information: Parts of works were done at Vexcel Imaging GmbH, Graz, financially supported by the EU project “innoVation in geOspatiaL and 3D daTA — VOLTA” funded under the Marie-Curie RISE scheme as no. 734687. The author Xin Wang would like to thank the China Scholarship Council (CSC) for financially supporting his PhD studying at Leibniz Universität Hannover, Germany. The author Xiao Teng would like to thank the Graduate Student Exchange Program of Wuhan University for his visiting scientist scholarship for studying in Germany.

PY - 2019

Y1 - 2019

N2 - Recently, global SfM has been attracting many researchers, mainly because of its time efficiency. Most of these methods are based on averaging relative orientations (ROs). Therefore, eliminating incorrect ROs is of great significance for improving the robustness of global SfM. In this paper, we propose a method to eliminate wrong ROs which have resulted from repetitive structure (RS) and very short baselines (VSB). We suggest two corresponding criteria that indicate the quality of ROs. These criteria are functions of potentially conjugate points resulting from local image matching of image pairs, followed by a geometry check using the 5-point algorithm combined with RANSAC. RS is detected based on counts of corresponding conjugate points of the various pairs, while VSB is found by inspecting the intersection angles of corresponding image rays. Based on these two criteria, incorrect ROs are eliminated. We demonstrate the proposed method on various datasets by inserting our refined ROs into a global SfM pipeline. The experiments show that compared to other methods we can generate the better results in this way.

AB - Recently, global SfM has been attracting many researchers, mainly because of its time efficiency. Most of these methods are based on averaging relative orientations (ROs). Therefore, eliminating incorrect ROs is of great significance for improving the robustness of global SfM. In this paper, we propose a method to eliminate wrong ROs which have resulted from repetitive structure (RS) and very short baselines (VSB). We suggest two corresponding criteria that indicate the quality of ROs. These criteria are functions of potentially conjugate points resulting from local image matching of image pairs, followed by a geometry check using the 5-point algorithm combined with RANSAC. RS is detected based on counts of corresponding conjugate points of the various pairs, while VSB is found by inspecting the intersection angles of corresponding image rays. Based on these two criteria, incorrect ROs are eliminated. We demonstrate the proposed method on various datasets by inserting our refined ROs into a global SfM pipeline. The experiments show that compared to other methods we can generate the better results in this way.

UR - http://www.scopus.com/inward/record.url?scp=85083328892&partnerID=8YFLogxK

U2 - 10.1109/CVPRW.2019.00349

DO - 10.1109/CVPRW.2019.00349

M3 - Conference contribution

AN - SCOPUS:85083328892

SN - 9781728125077

T3 - IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops

SP - 2896

EP - 2904

BT - 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2019

Y2 - 16 June 2019 through 20 June 2019

ER -