An Occupancy Grid Mapping enhanced visual SLAM for real-time locating applications in indoor GPS-denied environments

Current Real-Time Locating Systems (RTLS)typically deployed in indoor built environments are generally based on wireless technologies, fixed cameras, or Lidar-based Simultaneous Localization and Mapping (SLAM), which generally suffer from the drawbacks of low accuracy, reliance on existing infrastructures that may be not available in the deployed environments, labor-intensive environment instrumentation, or economic infeasibility for wide deployment. By improving an ORB RGB-D SLAM with Occupancy Grid Mapping, this paper proposes a new indoor RTLS that can be readily adapted and deployed for a broad range of indoor locating applications while overcoming the limitations faced by current solutions. In addition to the sparse feature map that is maintained by ORB SLAM itself, a new 2D mapping module is developed to build and maintain an additional 2D Occupancy Grid Map (OGM). The 2D OGM is built with the 3D camera poses estimated by Visual SLAM (vSLAM)and laser scans extracted from the point cloud observed by the camera from those poses. In addition, the Robot Operating System (ROS)visualization tools are used to overlay real-time current camera poses and observations (virtual laser scans)on the OGM. This approach not only provides more intuitive pose information to users and allows them to interact with the system, but also enables path planning and continuous navigation, which cannot be implemented directly on vSLAM's original feature map. The localization accuracy of the proposed system is experimentally evaluated with a set of visual landmarks that are installed in a large-scale building environment. The achieved marker position measurement accuracy ranges from 0.039 m to 0.186 m and the marker distance measurement accuracy ranges from 0.018 m to 0.235 m, proving the method's feasibility and applicability in providing real-time and accurate localization for a wide range of applications within constructed facilities and the built environment. Three examples are provided to highlight such potential applications, including path planning and real-time navigation, geo-tagged date collection and location-aware point cloud updating.