3D mapping of indoor environments using RGB-D data

Abstract

In this work a method for 3D mapping of indoor environments using RGB-D data is presented. Our method exploited the RGB and depth images provided by Kinect sensor for pairwise registration. The proposed method contains five main steps. Firstly, keypoints are detected and automatically established their correspondences in the RGB images. In the second step we propose a rectification between RGB and IR images to find the best association of the correspondent keypoints and depth values. Following, the XYZ coordinates can be computed. Then, a pairwise registration is realized. Finally, we present a linear model for the global consistency. The performance of the method is demonstrated using four datasets of an office environment. An assessment of relative accuracy of the sensor trajectory shows pairwise registration errors around 3,0 cm

Keywords:
Photogrammetry; Kinect; RGB-D data; Indoor 3D mapping; Pairwise registration; Global consistency