A RGB-D Vision based Indoor SLAM using 2.5D Map by Multiple UAVs

This paper presents an approach to build an indoor 2.5D map with multiple UAVs(Unmanned Aerial Vehicles). In an indoor environment, GPS system could not be used so each UAV adopted a tracking camera(Intel RealSense T265) to localize itself and 3D data for map building is acquired by the stereo depth camera(Intel RealSense D435). However, the raw data of the above sensors can have significant noise and large size of data, so the octree filter is applied as a solution. Usually indoor aerial robots have not enough flying time to build entire map of the floor so this paper suggested an improved way to build entire map in 2.5D with multiple UAVs. Each UAV can build a map for part of the global floor and each 3D local maps are converted to 2.5D, then merged together. To make a global map from several local maps, feature points are extracted that can be detected in 2.5D of the indoor environment. Through a series of experiments, the proposed method creates a map that takes into account obstacles and complex structures in the indoor environment, and shows that the ground robot can autonomously drive using this map.