Natural landmark extraction in 2D laser data based on local curvature scale for mobile robot navigation

This paper proposes a novel algorithm for natural landmark extraction from 2D laser scan data. The algorithm permits to extract several types of landmarks such as dominant points, edges, center of curve segments and virtual corners. This is achieved by using a new geometrical feature identification approach based on a novel concept of local scale called local curvature scale. Intuitively, this approach is computationally efficiencies due to needless of the construction of scale space map which is required in the other methods based on scale space theory. Experimental results show that the algorithm is efficient to detect landmarks for semi-structured environments.