Leveraging Semantics in Appearance based Loop Closure Detection for Long- Term Visual SLAM

In simultaneous localization and mapping, the detection of true loop closure benefits in relocalization and increased map accuracy. However, its performance is largely affected by variation in light conditions, viewpoints, seasons, and presence of dynamic objects. Focusing the advantages of visual semantics, this research proposes a semantics aided bag-of-wordsmodel overcoming the limitations of conventional BoW model which performs appearance-based loop detection and has been gold standard for many years. The proposed method reduces the computation time while matching current frame with the database image by segmenting the search space. Further it enhances the true loop detection rate.