Robust Photogeometric Localization over Time for Map-Centric Loop\n Closure

Map-centric SLAM is emerging as an alternative of conventional graph-based\nSLAM for its accuracy and efficiency in long-term mapping problems. However, in\nmap-centric SLAM, the process of loop closure differs from that of conventional\nSLAM and the result of incorrect loop closure is more destructive and is not\nreversible. In this paper, we present a tightly coupled photogeometric metric\nlocalization for the loop closure problem in map-centric SLAM. In particular,\nour method combines complementary constraints from LiDAR and camera sensors,\nand validates loop closure candidates with sequential observations. The\nproposed method provides a visual evidence-based outlier rejection where\nfailures caused by either place recognition or localization outliers can be\neffectively removed. We demonstrate the proposed method is not only more\naccurate than the conventional global ICP methods but is also robust to\nincorrect initial pose guesses.\n