Vision-based algorithms provide the accurate localisation necessary for mobile Augmented Reality applications, but achieving robust behaviour in unprepared environments is a challenging problem. We have developed a real-time camera localisation and mapping system based on particle filtering. The system maps visual interest point features as landmarks for localisation in unprepared environments. The tracking is resilient to erratic motion and camera occlusion.  Further work developed a simultaneous localisation and mapping system that gives very robust performance, even in the presence of severe occlusion of features or camera shake. The key component is a novel utilisation of multi-resolution descriptors in a coherent top-down framework.  Currently we are investigating the use of line segments as visual features for camera localisation. The use of line segments provides viewpoint invariance and robustness to partial occlusion of features. The landmark map generated from the line segments can be used to infer higher level structure in the scene.  |