Dense 3D map building for autonomous mobile robots

Abstract

Numerous approaches for vision-based navigation of mobile robots using depth maps are classified into two classes: The first class is to extract image features and to produce sparse depth maps, and the second class is based on correlation algorithms to produce depth maps. One of the problems of the first class is that it may not offer enough information of the scene for autonomous navigation. In this paper, we propose a method to build a 3-dimensional map, not a depth map, which can give 3D visual cues for autonomous navigation. The main purpose of the 3D map is not only to decrease computational time needed for building a depth map, but also to provide enough 3D information for autonomous navigation than a sparse map does. We demonstrate the procedure to build a 3D map and present that our method is appropriate for mobile robot navigation.