Title Bundle Adjustment using Conjugate Gradients with Multiscale Preconditioning
Authors Martin Byröd, Karl Åström
Full-text Available as PDF
Alternative Location http://www.bmva.org/bmvc/20...
Publication British Machine Vision Conference
Year 2009
Pages 10
Document type Conference paper
Conference name British Machine Vision Conference
Conference Date 2009-09-07
Conference Location London, UK
Status Published
Quality controlled Yes
Language eng
Abstract English Bundle adjustment is a key component of almost any feature based 3D reconstruction<br> system, used to compute accurate estimates of calibration parameters and structure and<br> motion configurations. These problems tend to be very large, often involving thousands<br> of variables. Thus, efficient optimization methods are crucial. The traditional Levenberg<br> Marquardt algorithm with a direct sparse solver can be efficiently adapted to the special<br> structure of the problem and works well for small to medium size setups. However, for<br> larger scale configurations the cubic computational complexity makes this approach pro-<br> hibitively expensive. The natural step here is to turn to iterative methods for solving the<br> normal equations such as conjugate gradients. So far, there has been little progress in this<br> direction. This is probably due to the lack of suitable pre-conditioners, which are con-<br> sidered essential for the success of any iterative linear solver. In this paper, we show how<br> multi scale representations, derived from the underlying geometric layout of the problem,<br> can be used to dramatically increase the power of straight forward preconditioners such<br> as Gauss-Seidel.
Keywords Computer vision, non-linear least squares problems, simultaneous localization and mapping, bundle adjustment,

Questions: webmaster
Last update: 2013-04-11

Centre for Mathematical Sciences, Box 118, SE-22100, Lund. Telefon: +46 46-222 00 00 (vx)