AUTHOR=Yu Shumei , Wu Junyi , Xu Haidong , Sun Rongchuan , Sun Lining 

TITLE=Robustness Improvement of Visual Templates Matching Based on Frequency-Tuned Model in RatSLAM

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 14 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.568091

DOI=10.3389/fnbot.2020.568091

ISSN=1662-5218

ABSTRACT=This paper describes an improved brain-inspired simultaneous localization and mapping system (RatSLAM system) that extracts visual features from salient maps using frequency-tuned (FT) model. In original RatSLAM, the visual template feature is in the form of one-dimensional vector that only depends on pixel intensity, therefore the feature is prone to be affected by changes of illumination intensity. Different from the RatSLAM system that directly generates visual templates from raw RGB images, we propose to adopt the frequency-tuned model to convert RGB images into salient maps to obtain the visual template. The visual template extracted from salient maps contains more feature information of images. Experimental results demonstrate that the accuracy of loop closure detection was improved by comparing the number of loop closure detected by our method and the original RatSLAM. It is also validated that the proposed FT model-based visual templates has improved the robustness of RatSLAM in identifying familiar visual scenes.