A Robot Visual Homing Algorithm Based on Reduced Landmarks

Among all the local visual homing algorithms inspired by the navigation of living creatures, ALV (average landmark vector) algorithm gains extensive attention for its simple model, better homing performance and less storage space, etc. In unstructured environments, ALV algorithm often uses local features of the image as natural landmarks. Under this circumstance, it is hard to ensure the correspondence of landmarks, at the same time, on the premise of ensuring its homing performance, there is no effective way to reduce the number of landmarks reasonably. In order to solve these problems, the concept of horizon ring-shaped region is put forward by the study on catadioptric panoramic image based on hyperboloidal mirror. An improved ALV algorithm based on natural landmarks is proposed by combining natural landmarks which are SIFT (scale invariant feature transform) features extracted from the ring-shaped region with ALV model. The improved algorithm effectively narrows the area for extracting landmarks and ensures the correspondence of landmarks, at the same time, it reduces the number of landmarks. Many experiments in different actual scenes indicate that the proposed algorithm improves homing precision effectively.