Mug Shots for Fish

Alfonso Pérez-Escudero

Mug Shots for Fish

Sid is a freelance science journalist.

People usually have distinctive noses, ears, and other features by which they can be easily recognized. Ants and fruit flies, not so much—at least to our eyes. But now, researchers have developed computer software that can scan video images of groups of animals, identify individual creatures within those crowds, and then track their movements—even if humans can’t tell one individual from another. Rather than focusing on specific features, the software compiles a “visual fingerprint” that’s based on the brightness and distinct arrangement of pixels in still images of each animal taken from the video. From a well-lit video, the software can develop visual fingerprints of up to 20 individuals in just 5 to 10 minutes—an interval typically sufficient for each individual to break free from the surrounding crowd and be seen in a wide variety of postures. In analyses of 23 videos of five different types of animals (mice, fruit flies, ants, zebrafish [shown], and small tropical fish called medaka), the software was able to identify and correctly track individuals (color-coded trails) within those groups 99.7% of the time, researchers report online today in Nature Methods. The program can recognize creatures even if they leave and then reenter the video camera’s field of view. In some cases, the researchers say, it was also able to identify individuals within the same group in separate tests a week or more later. The program, which tracks creatures automatically once trained, could be a boon for researchers studying animal behavior, especially for those trying to analyze the movements of substantial numbers of active, nearly identical creatures. One example, the researchers say, might be to study how a school of fish dumped suddenly into an unfamiliar environment explore their new surroundings together and then gradually begin to jostle and compete for territory.

Posted in Plants & Animals, Technology