Selecting distinctive scene features for landmarks

Abstract

Autonomous finding of landmarks for guiding long distance navigation by a mobile is explored. In a trial navigation, the robot continuously views and memorizes scenes along the route. When the same route is subsequently pursued again, The robot locates and orients itself based on the memorized scene. Since the stream of images is highly redundant, it is transformed into an intermediate 2(1/2)D representation, called Panoramic Representation, with much less amount of data. Although the Panoramic Representation can be used for guidance of the autonomous navigation, it still contains a huge amount of data for a very long route. A human memorizes only very impressive objects along the route such as an old church or a very high tower and uses th,em as landmarks. Our robot also finds distinctive objects along the route and memorizes their features as well as spatial relationships f o r navigation guidance. 3 0 objects along the route are segmented in the Panoramic Representation by fusing range estimates and color attributes, and th,en a structure m a p representing their arrangement in space is yielded. I n order to find distinctive objects used for the landmarks, the spatial relationsh.ips, shapes and color attributes of the objects are examined.

Coresets for visual summarization with applications to loop closure

“…Studied intensely between the vision [26], [25], [15], [24] and the robotics [11], [4], [16] communities, attempts have been made to allow faster loop-closure processing of larger datasets [5] and to utilize 3D information in order to increase specificity [21]. Maddern et al [18] propose a way of minimizing the effects of life-long collection of images in the mapping phase, especially with respect to location model creation.…”

A strategy for acquiring an environmental model with panoramic sensing by a mobile robot

“…Rather than building a world model from a set of images taken at discrete points, our robot continuously views the environment and arranges its essential information into a route panoramic representation, a 2( 1/2)-D sketch [7]. The 2( 1/2)-D sketch yields arrangement of objects along the path and their visual patterns, such as object shapes, colors and textures.…”

“…Since the camera motion along a smooth path contains translation components, we can acquire range information to the objects from their image velocities across the slit. By integrating the panoramic view, depth, path geometry and camera velocity, we get a 2( 1/2) representation of scenes along the route, called Panoramic Representation [7].…”