Object-aware semantic mapping of indoor scenes using octomap

In order to enable robots to provide high-level services in complex indoor environments, it is necessary to improve the robots' ability to cognize the environments. Most of the existing research is focused on indoor 3D reconstruction and semantic segmentation without the organization and maintenance of object recognition results. In this paper, we present an approach to build a 3D semantic map that includes both voxel-based geometrical demonstrations and object-aware entities with the combination of Simultaneous Localization and Mapping (SLAM) and Mask Region-based Convolutional Network (Mask R-CNN). An extended seeded region growing algorithm is designed for 3D segmentation refinement, and an octree-based framework octomap is used to present 3D map in replacement of point cloud map. We present experiments in a simulated home environment and the experimental results verify the accuracy and efficiency of our method.