Homography Estimation Network Based on Dense Correspondence

2021 ◽  
pp. 632-639
Author(s):  
Xiaohang Yang ◽  
Lingtong Kong ◽  
Ziyun Liang ◽  
Jie Yang
2015 ◽  
Vol 742 ◽  
pp. 290-293
Author(s):  
Xiu Zhi Li ◽  
Ai Lin Yang ◽  
Huan Qiu ◽  
Song Min Jia

This paper presents a technique for monocular Structure from Motion (SFM) that reconstructs 3D world shape. The technique proposed uses optical flow for 2D pixel pair matching and Angular Bundle Ajustment (ABA) for 3D structure refinement. The proposed strategy has two main advantages. Firstly, optical flow fields provide sufficient dense correspondence of image point pairs and secondly, ABA outperforms classic BA variants, especially for the points relatively far from camera. The reconstruction results obtained in realistic scenario demonstrate the effectiveness and accuracy of the proposed algorithm.


Sign in / Sign up

Export Citation Format

Share Document