Local visual homing by warping of two-dimensional images

Local visual homing methods can be used in the context of topological maps to travel between neighboring locations. These methods take two images as input and produce a home vector that points from the vantage point of one image to that of the other. “Warping” [M.O. Franz, B. Schölkopf, H.A. Mallot, H.H. Bülthoff, Where did I take that snapshot? Scene-based homing by image matching, Biological Cybernetics 79 (3) (1998) 191–202] is an attractive homing method since it does not require an external compass. Here we describe how the performance of warping can be substantially improved by extending the method from one- to two-dimensional images, with only a moderate increase in the computational effort. Experiments on several image databases confirm the improved performance.