A Robust Visual Loop-Closure Detection Method of vSLAM for Ambiguous Environment

Abstract

Loop-closure detection is a significant component for autonomous robot SLAM to build a map of the environment. However, the robustness of visual SLAM loop-closure detection in outdoor complex environment is severely restricted by perceptual aliasing, which is caused by changes in illumination, viewpoint and scale. In this paper, we present a novel method with high robustness and low complexity for visual loop-closure detection of autonomous robot. Our method extracts invariable local features, then clusters environmental invariable image descriptors. To reduce the selection complexity of massive loop-closure candidates to sub-linear, hash the image descriptors to Locality-Sensitive Hashing table without losing the invariant ability of image descriptors. The proposed method is tested efficiency and complexity experimentally using a publicly available dataset, results demonstrate that our method can recognize complex surroundings online with good robustness and has good applied value in constructing map accurately for autonomous robot visual SLAM in outdoor environment.