Simultaneous localization and mapping in unknown environment using dynamic matching of images and registration of point clouds

Abstract

This paper investigates the usage of the RGB-D cameras for generation of dense 3D maps of environments. The proposed approach to generation of dense three-dimensional maps of indoor environments with a complex infrastructure is based on multisensor dynamic information, spatial alignment of successive frames of data, view-based loop closure detection (the robot returns to the primary point), the alignment of the complete sequence of data (mapping and registration of image sequences and sparse clouds three-dimensional pixels). A new dynamic matching algorithm combining visual features and depth information to build global consistent maps is proposed. We evaluate the RGB-D Mapping in indoor environments of Chelyabinsk State University.