"...It's orange and small, and white stripes..." Augmented-reality system for fish species identification in aquariums

Abstract

This paper presents an original Augmented-Reality system to automatically identify aquarium fish species, providing a rich multimedia experience to customers. Our goal is to replace the signs placed near tanks in aquariums with a smartphone application based on image-processing. Our system is grounded on the Active Appearance Model for fish texture sampling. This paper also introduces a novel AAM matching function that measures the superimposition degree of the AAM instance edges and the targets' edges. The newly defined function significantly improves the AAM matching performance on textureless targets without modifying the computational cost. We evaluate our identification algorithm quantitatively on a comprehensive synthetic data set of static images, whereas we evaluate the usability of our AR system in real conditions qualitatively. It yields a 94% correct-identification rate on 15 species and runs up to 15 frames per second on an iPod Touch 4G, ensuring a satisfying user experience. © 2012 Springer-Verlag.