The agility and versatility offered by UAV platforms still encounter obstacles for full exploitation in industrial applications due to their indoor usage limitations. A significant challenge in this sense is finding a reliable and cost-effective way to localize aerial vehicles in a GNSS-denied environment. In this paper, we focus on the visual-based positioning paradigm: high accuracy in UAVs position and orientation estimation is achieved by leveraging the potentials offered by a dense and size-heterogenous map of tags. In detail, we propose an efficient visual odometry procedure focusing on hierarchical tags selection, outliers removal, and multi-tag estimation fusion, to facilitate the visual-inertial reconciliation. Experimental results show the validity of the proposed localization architecture as compared to the state of the art.

翻译：无人机平台因其敏捷性和多功能性而在工业应用中仍面临障碍，主要受限于室内使用场景。在此方面，一个关键挑战是在无GNSS环境中找到可靠且经济高效的飞行器定位方法。本文聚焦于基于视觉的定位范式：通过利用密集且尺寸异构的标签地图的潜力，实现了无人机位置和姿态估计的高精度。具体而言，我们提出了一种高效的视觉里程计算法，重点包括层级标签选择、离群点剔除以及多标签估计融合，以促进视觉-惯性联合校准。实验结果表明，与现有技术相比，所提定位架构具有有效性。