Automated recognition of penguins

Many animals carry unique patterns. We have our fingerprints, every zebra has a different pattern of stripes, no two leopards carry exactly the same spots, even ants can be distinguished by the pattern of pits in the cuticles on their backs. So, in principle, biologists can identify all the individuals of a particular species simply by remembering these patterns. This is fine for a small number of individuals, but to identify many thousand individuals an automated process is required. However, it is no simple matter to find and extract patterns and turn them into numbers that can be used as unique identifiers.

A working system would analyse images from remote cameras in real time. First, it would decide whether or not an animal of the species of interest is actually in the image. Next it would find and extract the pattern of spots or lines. Then it must check whether or not the pattern has been recorded previously and if so which individual it represents and if not record it as a new individual.

A University of Bristol team, led by Physicist and penguin enthusiast, Professor Peter Barham; together with Computer Vision expert, Dr Neill Campbell; Biologist, Professor Innes Cuthill and Tilo Burghardt, the PhD student who makes it all work; have managed to show all this is possible.

Now, we have secured funding from the Leverhulme Foundation to develop our work, using African penguins (which carry a unique pattern of spots on their chests) to create a fully automatic recognition system that will continuously monitor all the 20,000 or so individual penguins that breed on Robben Island. The project starts on 1st October 2006 and will run for 3 years with Tilo Burghardt employed as a post-doctoral researcher and a PhD student, Richard Sherley.

The system will allow us not only to monitor the daily comings and goings of the birds, discovering for example how much time they spend at sea foraging for food to feed their chicks, but also (using 3D modelling to calculate the overall volume of the individual penguins) tell us how much food they actually bring back and feed to their young. Over a prolonged period we will gain useful insights into population movements, longevity of the individuals, relationships between local fish stocks and breeding success and even be able to find out the extent to which global warming is affecting the penguins.

Importantly all this data will be gathered using remote cameras so that no penguins need be disturbed, unlike current research where to identify penguins they must first be caught and fitted with a numbered steel band, a process that is stressful both for the penguin and the researcher. Indeed, it has been argued that these flipper bands can adversely affect the penguins so that the reliability of the data gathered from their use has been called into doubt.

The way the Computer Vision programmes are written will make it relatively straightforward to extend the system to other species that carry unique patterns and we certainly hope that we will start to do so during the course of this award. Such full population monitoring systems that can recognize and track all the individuals within a group have the potential to revolutionise behavioural and conservation biology.