Publication Details
The effect of different parameterisations in incremental structure from motion
Vincent Lui W. (MONASH)
Ila Viorela S., Ing., Ph.D. (ANU)
Drummond Tom, prof. (MONASH)
Mahony Robert, prof. (ANU)
autonomous robotics, condition numbers, incremental solving, on-line estimation, point coordinates, reprojection error, structure from motion
Accurate online estimation of the structure of the environment together with the pose of the robot is an important component to enable autonomous robotic applications. This paper analyses the different parameterisations used in structure from motion (SFM) problem in the context of accuracy and efficiency of the online solutions. Three point parameterisations are compared: Euclidean, inverse depth and inverse distance. At the same time two representations, global and local point coordinates are tested. Different metrics are used to compare the results, camera localisation errors, reprojection errors, execution time as well as a complete analysis on how different parameterisations affect the convergence, system's condition number and the incremental solving are provided. The paper shows that, with the correct parameterisation, efficient globally consistent SFM is possible, which under the assumption of small, bounded number of correspondences performs in constant time in open loop.
@INPROCEEDINGS{FITPUB12266,
author = "Luk\'{a}\v{s} Polok and W. Lui Vincent and S. Viorela Ila and Tom Drummond and Robert Mahony",
title = "The effect of different parameterisations in incremental structure from motion",
pages = "1--9",
booktitle = "Proceedings of 2015 Australasian Conference on Robotics and Automation, ACRA 2015; Australian National UniversityCanberra; Australia",
journal = "Australasian Conference on Robotics and Automation (ACRA)",
volume = 2015,
number = 1,
year = 2015,
location = "Canberra, AU",
publisher = "Australian National University",
ISSN = "1448-2053",
language = "english",
url = "https://www.fit.vut.cz/research/publication/12266"
}