Active perception in visual simultaneous localization and mapping by feature / map point selection

One traditional and typical method to improve the computational efficiency of Visual Simultaneous Localization and Mapping (Visual SLAM) is selecting features which hold most contributions in estimating the pose in Tracking thread of SLAM. In this thesis, we first analyze the disadvantages of some most recent feature selection algorithms by adapting them onto ORB-SLAM2, then present a brand-new map point selection algorithm which significantly enhances the efficiency, while avoids some disadvantages in feature selection as well as keeps the accuracy at the same level. Finally, we test our map point selection algorithm on standard datasets with ORB-SLAM2 and analyze the performance improvement.U of I OnlyAuthor requested U of Illinois access only (OA after 2yrs) in Vireo ETD syste