Visual Inertial Odometer Based on GNSS Assisted Fusion of Point Line Features in Large Scale Environments

In order to achieve local accuracy and global drift free positioning capabilities of autonomous robots in complex large-scale environments, this paper proposes a visual inertial odometer (VIO) based on point line features and a global satellite navigation system (GNSS) multi-source information fusion algorithm for simultaneous positioning and map construction. Firstly, this system incorporates a line feature algorithm that can perform edge segment detection to more intuitively extract structural information in the environment, effectively improving the accuracy of pose estimation in weak texture situations; Secondly, using a multi-source fusion algorithm, a loose combination positioning model of VIO and GNSS is introduced to eliminate the cumulative error of VIO pose estimation results using GNSS positioning information; Finally, this system was tested on three different datasets. Experiments have shown that this system can improve the motion estimation accuracy of VIO between adjacent frame images in weak texture environments and suppress the cumulative error of VIO in large-scale environments, with strong real-time and robustness.