SeqSLAM with Bag of Visual Words for Appearance Based Loop Closure Detection

The detection of pre-visited areas in robots’ traversed path, widely known as loop closure detection, is vital for drift and position correction in robotic applications, such as simultaneous localization and mapping. In this paper, we present a sequence based approach for pose estimation, by advancing the well known SeqSLAM algorithm with the usage of Bag of Words (BoW) model. A visual vocabulary is produced in an offline procedure resulting in the system ’s ability to describe the incoming image stream by visual words, at the online process. Image similarity is achieved through BoW histogram comparisons instead of sum of absolute differences metric. Comparative results on several publicly-available datasets show the benefits of the proposed method offering high recall scores at 100% precision against the original one.