With the advancement of the Vehicle-based Internet of Things (IoT), the demand for positioning in challenging environment is growing. As a commonly employed positioning method, satellite navigation faces challenges in occlusion-prone environments, which poses difficulties in meeting the navigational requirements of vehicles. Inspired by insects orienting with the polarized skylight, we developed a bionic polarized compass and integrate it into our vision-based vehicular navigation system, aiming to enhance the performance of satellite systems in the IoT. Correspondingly, we introduce a multisensor fusion framework with polarized orientation and metric constraints visual inertial navigation system, which has capabilities for real-time location in large-scale areas, even in satellite-denied environments. First, we describe an optimization-based framework, which integrates visual, inertial, polarized orientation, and global position to get metric constraints. Subsequently, the visual-inertial system is coupled with metric constraints in practical applications, achieving large-scale drift-free positioning in satellite-denied environments. Finally, we conducted self-driving experiments in the city and walking experiments in the jungle, as well as tested the algorithm on Ford multi-AV seasonal data sets. The results of experiments, which lasted over 16 h and covered a distance of more than 150 km, show that the root mean square error of location is no more 6.18 m (no more than 0.04% $\circ $ of the total mileage). The experiments demonstrate that our method has better accuracy and robustness than existing vision-based methods and can be a useful complement to satellite-based navigation methods in the IoT.

中文翻译：

---

OMC-VINS：仿生天窗方向和基于视觉的度量限制了在具有挑战性的环境中进行车辆定位的建议


随着基于车辆的物联网 （IoT） 的进步，在具有挑战性的环境中定位的需求正在增长。卫星导航作为一种常用的定位方法，在遮挡易受阻的环境中面临挑战，难以满足车辆的导航需求。受到昆虫利用偏振天窗定向的启发，我们开发了一种仿生偏振指南针，并将其集成到我们基于视觉的车载导航系统中，旨在提高物联网中卫星系统的性能。相应地，我们引入了一种具有极化方向和度量约束视觉惯性导航系统的多传感器融合框架，该系统具有在大尺度区域实时定位的能力，即使在没有卫星的环境中也是如此。首先，我们描述了一个基于优化的框架，该框架集成了视觉、惯性、极化方向和全局位置以获得度量约束。随后，视觉惯性系统在实际应用中与度量约束相结合，在无卫星环境中实现大规模无漂移定位。最后，我们在城市中进行了自动驾驶实验，在丛林中进行了步行实验，并在福特多 AV 季节性数据集上测试了算法。持续 16 小时以上，覆盖距离超过 150 公里的实验结果表明，位置的均方根误差不超过 6.18 m（不超过总里程的 0.04% $\circ $）。实验表明，我们的方法比现有的基于视觉的方法具有更好的准确性和鲁棒性，可以成为物联网中基于卫星的导航方法的有用补充。