A real-time algorithm for vision-based localization

We present a real-time algorithm for the localization of an agent in an area whose topography is given. The data is organized in a matrix whose cells correspond to square area units of the terrain. Our algorithm consists of two stages: preprocessing and localization. In the preprocess stage the a priori data is processed to yield information which will be used during the localization stage. Specifically, the skyline of the panoramic view which is seen from each point in the terrain is extracted. A feature vector is constructed based upon the skyline data. In the localization stage, the feature vector of the panoramic skyline, which is seen from the agent, is matched against the preprocessed feature data in order to find the location of the agent.