Novel indoor mobile robot navigation using monocular vision

Novel autonomous navigation system for an indoor mobile robot based on monocular vision is presented. The navigation system is composed of on-line and off-line stages. During the off-line learning stage, the robot records an image frame sequence. From this sequence a hybrid environment map is built with Rao-Blackwellized particle filters (RBPF), where the number of resampling steps is determined adaptively for significantly reducing the particle depletion problem, and the evolution strategies (ES) are introduced for avoiding particle impoverishment. The map is partitioned into topological locations characterized by a set of geometrical scale invariant key-points. These key-points, represented with multi-dimension descriptors, can be robustly matched despite changes in contrast, scale and viewpoint, demonstrated with nearest neighbor search based on KD-tree. In the on-line navigation stage, the robot recognizes the most likely location through robust location recognition algorithm, estimates the relative pose between the locations, and then navigates the environment autonomously. Experiment results carried out with a real robot in an indoor environment show the superior performance of the proposed method.