Robust Wearable Camera Localization as a Target Tracking Problem on SE(3)

Guillaume Bourmaud and Audrey Giremus

Abstract

This paper deals with the trajectory estimation of a wearable camera evolving in an indoor environment where a database of images has been previously annotated with map coordinates. The difficulty of this problem resides in the fact that: i) hand-held objects are frequently interposed between the camera and the environment during daily living activities; ii) strong motion blur and differences in illumination occur; iii) the environment changes between the images of the database and the video frames to localize, and the viewpoints can be significantly different. The contribution of this paper is threefold: 1) We formulate the localization problem as a target tracking problem on the Lie group of camera motions SE (3), where the measurements are map coordinates obtained by applying a Content Based Image Retrieval algorithm to the video frames. 2) In order to solve this problem, we derive a novel Rao-Blackwellized particle smoother on Lie groups, which builds upon the recently proposed Extended Kalman Filter on Lie groups and the Rauch-Tung-Striebel Smoother on Lie groups that we also derive in this paper. 3) To take into account the topology of the environment, we propose to introduce virtual measurements that guide the particles and prevent them from crossing walls. We demonstrate, on several challenging video sequences, where the person wearing the camera performs daily living activities, that the proposed method is able to efficiently deal with outlier measurements and achieves a sub-meter level accuracy while the state of the art algorithms lack in robustness.

Session

Poster 1

Files

Extended Abstract (PDF, 2M)
Paper (PDF, 4M)
Supplemental Materials (ZIP, 7M)

DOI

10.5244/C.29.39
https://dx.doi.org/10.5244/C.29.39

Citation

Bibtex

@inproceedings{BMVC2015_39,
	title={Robust Wearable Camera Localization as a Target Tracking Problem on SE(3)},
	author={Guillaume Bourmaud and Audrey Giremus},
	year={2015},
	month={September},
	pages={39.1-39.12},
	articleno={39},
	numpages={12},
	booktitle={Proceedings of the British Machine Vision Conference (BMVC)},
	publisher={BMVA Press},
	editor={Xianghua Xie, Mark W. Jones, and Gary K. L. Tam},
	doi={10.5244/C.29.39},
	isbn={1-901725-53-7},
	url={https://dx.doi.org/10.5244/C.29.39}
}