Homography-based loss function for camera pose regression

Type Article
Date 2022-07
Language English
Author(s) Boittiaux Clementin1, Marxer Ricard2, Dune Claire3, Arnaubec Aurelien4, Hugel Vincent5
Affiliation(s) 1 : Equipe PRAO, Ifremer, La Seyne-sur-Mer, France, 83500
2 : Dept. of Computer Science, Universit de Toulon, Aix Marseille Univ, CNRS, LIS, Marseille, France, 83130
3 : COSMER LABORATORY, Université de Toulon, La Garde Cedex, France, 83041
4 : Ifremer, France
5 : COSMER, University of Toulon, LA GARDE, France, 83957
Source Ieee Robotics And Automation Letters (2377-3766) (Institute of Electrical and Electronics Engineers (IEEE)), 2022-07 , Vol. 7 , N. 3 , P. 6242-6249
DOI 10.1109/LRA.2022.3168329
Keyword(s) Localization, deep learning for visual perception
Abstract

Some recent visual-based relocalization algorithms rely on deep-learning methods to perform camera pose regression from image data. This paper focuses on the loss functions that embed the error between two poses to perform deep-learning based camera pose regression. Existing loss functions are either difficult-to-tune multi-objective functions or present unstable reprojection errors that rely on ground-truth 3D scene points and require a two-step training. To deal with these issues, we introduce a novel loss function which is based on a multiplane homography integration. This new function does not require prior initialization and only depends on physically interpretable hyperparameters. Furthermore, the experiments carried out on well established relocalization datasets show that it minimizes best the mean square reprojection error during training when compared with existing loss functions.

Full Text
File Pages Size Access
8 1 MB Access on demand
arXiv:2205.01937v1 - Preprint 8 974 KB Open access
Top of the page