A fast and effective algorithm to track fish school

To investigate collective motion of groups of animals, it is important to track multiple individual moving animals and acquire their positions over time and space. A few studies have tried to solve this problem aiming for automated data acquisition. But none have solved the problem adequately, since automated tracking
is difficult to achieve due to complexities in individual shape, sophisticated motion pattern sand frequent occlusion.Several algorithms on this problem have been published, usually for one special species, the zebra fish, for instance. Such algorithms tended to be very demanding regarding the video quality (high frame rates, high image resolution and steady back ground), and often are very time-consuming. Here we have developed an
integrated approach based on artificial neural networks which enables us to automatically extract individual trajectories from both high and low quality videos. We applied our method to track different fish videos, it was found that our method has a high efficiency and accuracy in most situations.