Hierarchical VSLAM and NBV Framework for UAV Mapping of Urban Environments
Urban mapping using Unmanned Aerial Vehicles (UAVs) presents significant challenges due to complex structures, occlusions, and the need for varying levels of detail. Current approaches face a challenge with efficiently managing the trade-off between coverage and detail, particularly in dense urban cities. The current approaches enable mapping capabilities, but achieving fully autonomous mapping remains challenging, particularly in dense urban settings. There is a need for systems that can autonomously manage both global coverage and local detail acquisition without human intervention. While VSLAM enables real-time localization and mapping, and NBV strategies help optimize viewpoint selection, there remains a gap in effectively combining these approaches for autonomous multi-scale urban mapping that can adapt to different environmental complexities and application requirements.
The aimed framework in the thesis will focus on optimizing viewpoint selection at multiple levels (global and local) to ensure comprehensive, high-resolution 3D reconstruction of urban structures such as buildings, streets, and dense urban features.
The expected methodology is to develop a multi-level approach where the urban environment is divided into global and local segments. The global mapping is to use VSLAM to generate a coarse 3D map of the entire environment with key urban landmarks. Then followed by a local refinement by planning local NBV UAV trajectories that focus on specific regions for detailed scanning and reconstruction. You are expected to implement the framework in simulation environments such as Gazebo, AirSim, or a similar tool.
A Review on Viewpoints and Path-planning for UAV-based 3D Reconstruction, Mehdi Maboudi, MohammadReza Homaei, Soohwan Song, Shirin Malihi, Mohammad Saadatseresht, and Markus Gerke. https://arxiv.org/abs/2205.03716