An Appearance and Viewpoint Invariant Visual Place Recognition for Seasonal Changes

Place recognition has typically been addressed as a problem of recognizing the location of a given query image as a previously visited place while comparing it with the geotagged database images. Despite a lot of research in this area, vision-based place recognition is still an open challenge because of the changing environmental conditions which cause drastic appearance changes, making it difficult for a robot to recognize the place. This research addresses the above-mentioned problem and proposes the solution for place recognition at a low memory footprint. The proposed place recognition system focuses on identifying the combination of different feature detectors and descriptors that are invariant to the viewpoint and seasonal changes and can efficiently recognize a place at high accuracy. Through experimental results, it is shown that combination of CenSure based STAR detector and BRISK achieves high detection accuracy.