A model-based method for indoor mobile robot localization using monocular vision and straight-line correspondences

A model-based method for indoor mobile robot localization is presented herein; this method relies on monocular vision and uses straight-line correspondences. A classical four-step approach has been adopted (i.e. image acquisition, image feature extraction, image and model feature matching, and camera pose computing). These four steps will be discussed with special focus placed on the critical matching problem. An efficient and simple method for searching image and model feature correspondences, which has been designed for indoor mobile robot self-location, will be highlighted: this is a three-stage method based on the interpretation tree search approach. During the first stage, the correspondence space is reduced by virtue of splitting the navigable space into view-invariant regions. In making use of the specificity of the mobile robotics frame of reference, the global interpretation tree is divided into two sub-trees; two low-order geometric constraints are then defined and applied directly on 2D-3D correspondences in order to improve pruning and search efficiency. During the last stage, the pose is calculated for each matching hypothesis and the best one is selected according to a defined error function. Test results illustrate the performance of this approach.