Ensemble of Bayesian filter with active and passive nodes for loop closure detection

In this paper, we describe a novel extension of the real-time appearance-based mapping (RTAB-Map), called the Ensemble of Real-Time Appearance-Based Mapping (ERTAB-Map). The original RTAB-Map calculates the probabilities of multiple beliefs for loop closure detection based on a single descriptor model. However, the ERTAB-Map can use an arbitrary number of descriptor models, in which a set of probability belief models are evaluated using an ensemble learning approach. The probability values are extracted from the active working memory and the passive long term memory of RTAB-Map. We have performed experiments on 388 images from the Lib6Indoor and 1063 images from Lib6Outdoor datasets. The results show that our ensemble of active and passive outperforms the original RTAB-Map. Furthermore, the ensemble achieves a recall of 91.59% and 98.65% on the Lib6Indoor and Lib6Outdoor respectively, with a corresponding precision of 100%.