Improved Loop Closure Detection Algorithm for VSLAM with Spatial Coordinate Index

Abstract

This paper presents an improved loop closure detection algorithm for VSLAM with spatial coordinate index in perceptual aliasing scenes. Firstly, we model the visual odometry accumulative error with Gaussian mixed distribution. Secondly, we eliminate the false loop closures by limiting the detection regions with accumulative error. Thirdly, the accumulative error is corrected by the appearance of loop closures for reducing the detection regions. Besides, we construct a dynamic octree for spatial coordinate index to accelerate the search of detection regions. Results demonstrate that, compared with IAB-MAP, FAB-MAP and RTAB-MAP, our algorithm has a better precision rate and a fine real-time property.