Visual Landmark Design, Detection and Identification for Vision-guided CAD Model-based Mobile Robot Localisation

Visual Landmark Design, Detection and Identification for Vision-guided CAD Model-based Mobile Robot Localisation

  A new pattern landmark has been designed and its detection and identification algorithms are proposed for an efficient CAD model-based localisation system for a vision-aided mobile robot. The known horizontal projection model of the vertical wall closed to the landmark, together with the sensed positions based on a CAD-model relative to the robot, help to determine the robot"s current position without any reference to its previous positions. The landmark is used to guide the searching process while the local CAD model is used for localisation. The accurate co-ordinates of the landmark are not important. A self-similar landmark with bar-colour-coding pattern after defining a "barlet" function is designed such that an upward-aiming camera can reliably detect it even if it is partially blocked by unforeseeable obstructions. The landmark can be effectively detected and recognised by using wavelet-based MRA for landmark searching and matching the colour pattern one by one with the pre-tested colour bands of HSV space for landmark recognition. Experimental results show the feasibility of the proposed landmark and its identification algorithm for localisation.