Edge-Based Loop Closure Detection in Visual SLAM

摘要

In robotics research which is focused on vision-based localization estimation, Loop Closure Detection (LCD) algorithms used to recognize visited places, have been established as the solution to the problem of error drift in Visual SLAM systems. In this paper, an LCD system based on the popular DoW2 module is analyzed using a classical state-of-art solution for keypoints extraction like FAST, and our proposal for keypoint selection based on image edges detection. The behavior of the system is analyzed and tested in three different datasets (one indoors and two outdoors). This preliminary work aims to be an initial design for future development of an LCD module to add to Visual SLAM Rebvo system.