An improved bag of words method for appearance based visual loop closure detection

Abstract

Loop closure detection is an essential component of many robotics applications such as SLAM (Simultaneous Localization and Mapping) and place recognition. This paper presents an appearance based loop closure detection algorithm based on bag of words method and inverse depth of feature words. Our proposed approach represent the feature points and descriptors in the images as the visual words according to the off-line generated visual directory to simplify the similarity comparison between the images. For the evaluation criteria, on the basis of tf-idf scores of visual words, the inverse depth of words are also introduced into the process of loop closure detection. Considering the temporal consistency of image sequences, the co-visibility graphs are also applied to verify the real loop from all candidates. At last, some experiment results are showed and analyzed to illustrate the feasibility and performance of our algorithm in different situations and environments.