Abstract: 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 ...