Omnidirectional vision based mobile robot topological localization

A robust omni-directional vision based localization method that allows us to obtain accurate mobile robot pose of large indoor environments is proposed. To implement the localization based on vision. In a learning step, the robot is manually guided on a path and an omni-directional image frames sequence is recorded. From this sequence a topological map is built with robust affine and scale invariant features extraction and matching algorithm. Each topological node represented by a set of panoramic images described with affine and scale invariant features. In the on-line localization stage, the robot localizes itself to the most likely node through robust Monte Carlo localization algorithm, and ambiguous robot pose estimation is resolved by this high probability distribution method. This enables the system to deal with perceptual aliasing or absence of reliable observation data. Experiment results carried out with a real robot in an indoor environment show the performance of the proposed method.