A bag of relevant regions model for visual place recognition in coral reefs

Vision-based place recognition in underwater environments is a key component for autonomous robotic exploration. However, this task can be very challenging due to the inherent properties of this kind of places such as: color distortion, poor visibility, perceptual aliasing and dynamic illumination. In this paper, we present a method for vision-based place recognition in coral reefs. Our method relies on using the Bag-of-Words (BoW) approach to derive a descriptor, for the whole image, from a set of relevant regions, which are extracted by utilizing a visual attention algorithm. The descriptor for each relevant region is built by using an histogram of the chromatic channels of the CIE-Lab color space. We present results of our method for a place recognition task in real life videos as well as comparisons of our method against other popular techniques. It can be seen that our approach performs better in most of the cases.