Monocular Visual-Inertial Odometry with Point-Line Feature Fusion Assisted by GPS Signals

The current mainstream visual-inertial odometry primarily relies on extracting feature points from images for tracking and estimating the camera pose. However, in low-texture environments and under significant changes in illumination, the number of extractable point features is limited. To address this issue, the algorithm proposed in this paper introduces line features at the frontend for fusion. It tightly couples point features with line features, enhancing the precision of indoor localization. Additionally, to improve the algorithm’s adaptability to outdoor environments, this paper suggests leveraging the principle of multi-sensor fusion. The Global Positioning System (GPS) signals are fused with the monocular visual-inertial odometry featuring point-line fusion through a nonlinear optimization algorithm with a loose coupling, reducing cumulative positioning errors. This enhancement improves the algorithm’s outdoor positioning accuracy, making it more robust. The algorithm’s indoor positioning accuracy is validated using the classic EuRoC dataset, while its outdoor positioning accuracy is validated using the KITTI dataset.