Vision-based simultaneous localization and mapping (vSLAM) is awell-established problem in mobile robotics and monocular vSLAM is one of themost challenging variations of that problem nowadays. In this work we study oneof the core post-processing optimization mechanisms in vSLAM, e.g. loop-closuredetection. We analyze the existing methods and propose original algorithm forloop-closure detection, which is suitable for dense, semi-dense andfeature-based vSLAM methods. We evaluate the algorithm experimentally and showthat it contribute to more accurate mapping while speeding up the monocularvSLAM pipeline to the extent the latter can be used in real-time forcontrolling small multi-rotor vehicle (drone).
展开▼
