BICT 2016: Gist+RatSLAM

This video illustrates the strengths of our proposed place recognition approach [1] compared to RatSLAM's place recognition front-end. The experiment is performed on the St. Lucia loop test dataset. The video demonstrates:
(1) A growing self-organizing network of neurons incrementally adapts itself to the topological representation of the input space (i.e. GIST features).
(2) The mapping is performed by fusing the place recognition front-end with the RatSLAM system. It can be observed that the proposed approach utilized less number of neurons to represent the environment compared to RatSLAM's place recognition module. Furthermore, our approach outperformed the RatSLAM's place recognition front-end on F1-scores and thus achieved the faster convergence (in term of loop-closure detection) compared to RatSLAM.
[1] S. M. A. M. Kazmi and B. Mertsching, “Gist+RatSLAM: An In-
cremental Bio-inspired Place Recognition Front-End for RatSLAM,”
Proc. of 9th EAI Int. Conf. on BICT (formerly BIONETICS), 2016.
(dl.acm.org/cita...)