PRIOR-SLAM: Enabling Visual SLAM for Loop Closure under Large Viewpoint Variations
PRIOR-SLAM is the first system which leverages scene structure extracted from monocular input to achieve accurate loop closure under significant viewpoint variations and to be integrated into prevalent SLAM frameworks.
PRIOR-SLAM is developed based on the framework of ORB-SLAM3 and refines the 3D mesh reconstruction module from Kimera.
We provide demonstrations comparing PRIOR-SLAM with well-known baselines on the selected sequences with challenging loop closures, including KITTI dataset, OpenLORIS-Scene dataset, and UrbanLoco dataset.
We provide demonstrations comparing PRIOR with well-known features on the interpolated Oxford Graffiti sequence, where the first image is taken as a reference and subsequent images are used to form pairs with increasing viewpoint variations.
