ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume II-3/W4
https://doi.org/10.5194/isprsannals-II-3-W4-87-2015
https://doi.org/10.5194/isprsannals-II-3-W4-87-2015
11 Mar 2015
 | 11 Mar 2015

INFORMATION FILTERING WITH SUBMAPS FOR INERTIAL AIDED VISUAL ODOMETRY

M. Kleinert and U. Stilla

Keywords: Indoor Positioning, Inertial Aided Visual Odometry, Bundle Adjustment, Submapping

Abstract. This work is concerned with the fusion of inertial measurements (accelerations and angular velocities) with imagery data (feature points extracted in a video stream) in a recursive bundle adjustment framework for indoor position and attitude estimation. Recursive processing is achieved by a combination of local submaps and the Schur complement. The Schur complement is used to reduce the problem size at regular intervals while retaining the information provided by past measurements. Local submaps provide a way to propagate the gauge constraints and thereby to alleviate the detrimental effects of linearization errors in the prior. Though the presented technique is not real-time capable in its current implementation, it can be employed to process arbitrarily long trajectories. The presented system is evaluated by comparing the estimated trajectory of the system with a reference trajectory of a prism attached to the system, which was recorded by a total station.