Tracking Penguins from Afar
Tilo - What we do is we identify individual animals, in particular African penguins, by their coat pattern. Instead of banding or otherwise tagging animals we use their natural coat better. In the case of African penguins their plumage in order to identify them. African penguins carry a spot pattern on their chest and like humans have unique fingerprints. African penguins have unique chest prints. We've built a computer system that could identify first of all an African penguin in video frame and then extract the chest print and compare it in a database population so we can tell which penguin is where, when.
Meera - You have me wearing this apron onto which a spot pattern has been printed. What do I now do?
Tilo - What you do is walk towards the camera. You'd better waddle actually and then the camera will pick up. It's a simplified system, rather than the real world that will pick up on this spot pattern and will extract it and will assign a name to it. It can identify you as an individual penguin.
Meera - I'm walking up now and it's saying that I'm Pandora.
Tilo - Yes, you're basically one of the penguins. Actually in real life we don't give names to them it's rather numbers that identify them because you couldn't have thousands of names remembered. We then assign the unique identifier like humans have names they get numbers.
Meera - So you're able to spot these penguins, but what's the benefit of being able to do this?
Tilo - The main problem we try to solve here is finding out about the species. The reason for that is the species is in decline. In the last few years about 50% loss in the population count has been found. Just to give you some numbers: about 100 years ago there were more than 1.5 million African penguins while last year's centres gave only 35,000 breeding pairs which is a really sharp decline. We need to find out why. With this technology we can basically play Big Brother over the colony without being invasive so you don't need to catch the animals or tag the animals any more, which is basically interfering with their life. With this technology you then get out information about it which mainly is where, when. We can talk about foraging duration, how long are they out of the colony to get the fish. We can talk about which penguins walk together in groups. Do they have friendships? You can start asking entire new questions that haven't been asked before in the first place.
Meera - How does the actual technology work to recognise the patterns?
Tilo - It is a computer vision system which does the following: the images are transmitted from the camera to the computer and the first thing the computer has then to decide is where in the frame/where in the image are African penguins in near-frontal poses? So basically penguins that show their chest pattern. Once the computer has figured out where they are in the image then the computer has to extract the chest area and the spot pattern within. Once the chest pattern is then available there's a final step to do and that is the matching. Matching of this chest pattern which is, as I said, very similar to human fingerprints to a database of individuals that have already been identified. Basically re-sighting the animal.
Meera - If it doesn't come up in the database it's obviously a new penguin and you would just assign it a new number.
Tilo - Exactly.
Meera - Where are you going to take this next? Are you going to expand it to other animals?
Tilo - Absolutely, I mean so far we are monitoring a colony on Robben Island but first of all there are three other species of penguin you can apply it to. Then on top of that the technology has also been trialled on zebras. Basically anything stripy and spotty, or technically carrying Turing light patterns, can be identified with the technology. For example with the zebra you have a stripe pattern. What you do is you can turn this stripe pattern into spot pattern by simply taking the bifurcation points and ending points of lines. Then you have a spot pattern again and you do what you do with penguins. If you have a look at different images of different animals, different zebras you will find that although they look very similar the details of the stripe pattern are vastly different. They again can be used to identify them as individuals without tagging them.