Forming a sparse representation for visual place recognition using a neurorobotic approach

This paper introduces a novel unsupervised neural network model for visual\ninformation encoding which aims to address the problem of large-scale visual\nlocalization. Inspired by the structure of the visual cortex, the model (namely\nHSD) alternates layers of topologic sparse coding and pooling to build a more\ncompact code of visual information. Intended for visual place recognition (VPR)\nsystems that use local descriptors, the impact of its integration in a\nbio-inpired model for self-localization (LPMP) is evaluated. Our experimental\nresults on the KITTI dataset show that HSD improves the runtime speed of LPMP\nby a factor of at least 2 and its localization accuracy by 10%. A comparison\nwith CoHog, a state-of-the-art VPR approach, showed that our method achieves\nslightly better results.\n