Qualitative landmark recognition using visual cues

Understanding the surrounding scene and identifying man-made structures is an important task in autonomous vehicle navigation in an outdoor environment. The ability to generalize the task of landmark recognition to a variety of landmarks belonging to a particular class is a difficult problem compounded by its ill-defined nature. In this paper, we propose a technique for performing landmark recognition based on visual cues of texture, edges and functional form. We use a neural network based classifier to identify regions of interest and extract linear and curved edge features falling in these areas. We combine these cues using heuristics and inference based on the function of the landmark.