Methodology of Assigning Terrain Object Images to the Class of Landmarks for Autonomous Mobile Robots

The problem of autonomous navigation of mobile robots on unfamiliar terrain in the absence of GPS remains relevant. Navigation is impossible without topological binding of the robot to the coordinates of several landmarks, for example, pillars, tree trunks, or objects of other (non-cylindrical) shape. Therefore, for autonomous navigation of a mobile robot, it is necessary to develop a technology for classification of the detected landmarks. This can be achieved by evaluating the correspondence of the images on the frame: the image of the reference landmark with a known shape and color and the image of a real object. This chapter analyzes the similarity between the reference image of a landmark and a real object image based on the estimation of the Hausdorff distance between the specified images: the smaller it is, the greater similarity. The technology of determining the Hausdorff distance and its threshold value using statistical characteristics for various type of landmarks and surrounding area has been developed. The probabilistic characteristics of correct classification of an object image in the frame are evaluated, and directions for further research are proposed.