Modeling and Information Processing

SAM - Structure and motion

Objective

• To use data from onboard imaging sensors to obtain a 3-D reconstruction of the  environment and the ego-motion of the carrier.

Contribution

While moving around an imperfectly known environment, an autonomous system should construct a more accurate model of its surrounding (estimation of the 3-D structure) and to locate itself within that environment (estimation of the ego-motion). This is a non-linear problem coupled in these two sets of variables. Moreover, it involves a data association problem in the case of a passive sensor like an optic imager. Therefore, this topic has close relationships with the EMI and ADO topics.

Depending on the applications, the focus is either on the reconstruction of the environment, in which case we are talking about SFM (Structure from Motion), or on the ego-motion (which amounts to SLAM: Simultaneous Localization And Mapping). Our recent work has mainly been concerned with dense SFM, for which we have created a complete processing system using an aerial oblique image sequence. This system takes occultation problems into account and relies on super-resolution results (SR topic) to provide a high-accuracy elevation map of the terrain. This work is currently being applied to support the autonomy of a UAV in the framework of the ReSSAC federating research project.

Illustrations

• SFM from oblique aerial image sequence
• Optical flow estimation and accuracy assessment

Last Update: 11 October 2006 - © ONERA 2009 - Terms of use