SuperPoint-based visual feature front-end for visual simultaneous localization and mapping



Journal Title

Journal ISSN

Volume Title



Visual Simultaneous Localization and Mapping (V-SLAM) addresses the problem where robots concurrently keep track of ego-motions and construct a map understanding of the environments from visual observations. While SLAM employs various sensor data, cameras in Visual SLAM (V-SLAM) have gained popularity for their cost-effectiveness. However, the challenges of lighting variations and textureless environments make reliable feature detection crucial for V-SLAM success. This thesis leverages the progress of learning-based feature detectors, developing a visual front-end based on SuperPoint for V-SLAM. Through empirical evaluations on challenging outdoor trajectories, this work demonstrates the reliability of the SuperPoint-based feature tracking approach for V-SLAM, offering insights of the performance of various feature detectors on V-SLAM tasks.


LCSH Subject Headings