Accurate Visual Odometry by Leveraging 2DoF Bearing-Only Measurements

Based on the workshop measurement positioning system (wMPS) and visual sensors, the joint localization and mapping system (wMPS-Point) has demonstrated significant enhancements in positioning accuracy compared to the conventional visual simultaneous localization and mapping (SLAM), which employs 3DoF point measurements of tags (wMPS receivers) obtained from multiple anchors (wMPS transmitters <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\ge ~2$ </tex-math></inline-formula>). However, when the anchors encounter occlusion or failure, the 2DoF bearing-only observations from a single anchor become insufficient to recover the 3-D position of the tag attached to the camera sensor using the triangulation principle, resulting in a significant decrease in the localization accuracy. To address the challenge, this article proposes a general robust positioning approach that integrates 2DoF bearing measurements into visual-based camera tracking and optimization modules. First, the 2-D bearing measurement is represented as the angle parameterization to establish a residual edge connecting camera poses and fixed wMPS anchor. Furthermore, a tightly coupled optimization framework is designed to take advantage of both visual and bearing measurements to achieve more accurate camera pose estimation performance. To ensure algorithmic robustness, an outlier filtering method is also presented to cope with anomalous data in wMPS bearing-only measurements. The effectiveness of the proposed method was evaluated using open-source datasets and real-life experiments. The results indicate that the proposed method outperforms wMPS-Point in the consistency of multiple sequences localization results. Even when our method leverages bearing-only measurements from a single anchor, its localization accuracy is comparable to that achieved using multiple wMPS anchors.