Object Detection-based Semantic Map Building for A Semantic Visual SLAM System

In this paper, we propose a novel semantic visual simultaneous localization and mapping (SLAM) system which can provide rich semantic information of landmarks in the map. The system is created based on a visual SLAM system using a RGBD camera. We integrate an object detection stage on the top of the SLAM system to obtain semantic landmarks of the scene. The semantic object-based landmarks are continuously saved into a semantic map to not only describe context information, but also update the semantic map in dynamical environments where objects can be moved passively. The experimental results demonstrated the validity of the proposed system.