[AG-TECH] Access Grid Based Seminar

[Lee Margetts]

*** Apologies for Cross Posting ***

Robust Real-Time Camera Localisation and Mapping

Andrew Calway, Dept. of Computer Science, University of Bristol

Friday, 23rd February, 2-3p.m.

Room 1.10, Kilburn Building

AG Node Operator, ben.green at manchester.ac.uk

Significant advances have recently been made in algorithms for real-time
estimation of the pose of a moving camera using only visual measurements.
There are typically two kinds of scenario. In the general case, no a
priori information is available about the structure of the scene, and thus
localisation must proceed in tandem with mapping depth values. This is the
simultaneous localisation and mapping problem (visual SLAM). In other
applications, prior knowledge of scene structure may be available, in the
form of CAD or wireframe models, for instance, and these can be utilised
to guide camera localisation. In both of these cases, an effective
mechanism for robust operation is stochastic filtering, which provides a
sound statistical framework for obtaining 'optimal' estimates of the 6-D
camera pose and 3-D map. Several systems now exist which can operate in
real-time (around 30 fps). In this talk I will describe recent work
carried out in these areas at Bristol. We are particularly interested in
designing localisation and SLAM algorithms which are robust to effects
caused by 'normal' camera use, such as camera shake and visual occlusion.
This is a challenging task and many existing algorithms fail in such
cases. We have utilised generalised stochastic filtering in the form of
particle filters and robust view-invariant feature matching to give
algorithms which are able to withstand both severe shake and visual
occlusion. This makes them particularly suitable for applications in
wearable computing and augmented reality, in which camera movement is
often agile and unpredictable. The talk will consist of an overview of
stochastic approaches to localisation and SLAM, along with details and
examples from our own work.

For location and Access Grid information, please see
http://www.mc.manchester.ac.uk/research/seminars/