Visual navigation for a mobile robot using landmarks

Abstract

The visual navigation system in this paper models the human visual system and utilizes the well-known physiological concepts of peripheral vision and central vision fields to conduct image processing separately and to integrate the results from these processes. With this system, we attempt to realize an efficient image processing method. The peripheral vision field, which has a low resolution, rapidly detects the presence of the object of interest from the scene, and is able to deal with differences in the size and shape of the object brought about by a change in the viewpoint. Because of its speed and its capability in dealing with ambiguities, such as analogy and association, a three-layer neural network is utilized here to realize this function. On the other hand, the central vision field focuses on the object detected by the peripheral vision field and zooms in on the candidate to check it further. If the object detected is adjudged to be the object of interest, the robot's visual navigation system computes the direction and the distance towards the object using the algorithm to be introduced here that extracts three-dimensional (3D) vectors from 2D coordinates. Great demand is seen for such machines in the near future.