Adult African penguins engage in mutual preening on the beach
Left to right: Peter Barham (Physics), Tilo Burghardt (Computer Science) and Innes Cuthill (Biological Sciences) at Bristol Zoo Gardens
The computer captures penguins on their way to the sea
Spot the penguin
13 November 2008
Every year the Royal Society holds its Summer Science Exhibition, the premier showcase for scientific excellence in the UK. Universities and other institutions are invited to submit proposals and a lucky few are accepted. The University of Bristol has been chosen for the past four years. This year it was the turn of the Penguin Recognition Project, a collaborative venture between the departments of Computer Science, Physics and Biological Sciences.
One of the problems facing animal conservationists is that in order to find out what is happening to an individual or a group of endangered animals over time, it is often necessary to identify them in some way as individuals. Currently, tagging is the method of choice. This may cause some distress to the animal, both while it is being tagged and afterwards, and they are not always reliable, since batteries run out and tags may be lost or misread. A far better solution would be to devise an intelligent, visual surveillance system that can be integrated into wildlife habitats as a non-intrusive means of capturing detailed and reliable data on the population. The aim of the Penguin Recognition Project is to do just that. By developing a system capable of the automatic monitoring of species, the team may have found the solution to a multitude of problems facing ecologists around the world who are hoping to conserve anything from butterflies to buffalo. Specifically, they are developing software and hardware to permit remote monitoring and identification of large populations, using techniques that originated in computer vision and human biometrics.
We tend to think of penguins as creatures of the ice, so it comes as something of a surprise to discover that Robben Island, a kilometre-wide piece of rock off the arid coast of South Africa where Nelson Mandela was incarcerated for 25 years, is home to a colony of African (or Jackass) penguins. The island’s original penguin colony was exterminated in the 1800s when the hunting of whales and seals was at its peak (‘robbe’ meaning ‘seal’ in Dutch), but in 1983 the penguins started to come back, since when the colony has grown to about 20,000, despite near-disaster following a major oil spill in 2000. But worldwide their numbers have declined from well over a million penguins in the 1930s to less than 35,000 breeding pairs today, with the loss accelerating in recent years. Today the African penguin is listed as ‘vulnerable to extinction’.
The aim of the Penguin Recognition Project is to develop a system capable of monitoring animal species.
To understand the impact of any conservation measures taken to improve the lot of the African (or other) penguins, it is important to follow the life-cycle of individual birds, thus it becomes necessary to identify individuals and to record where and when they are seen. By collecting large data sets it is then possible to estimate survival times and breeding success rates. However, the methods currently used to identify penguins (transponders placed under the skin or steel bands clipped to their flippers) induce stress in both the bird and the handler. The trauma can lead to nest abandonment and possibly reduced breeding rates, and there have been reports of tags getting snared in undergrowth.
African penguins are quite small – about 40 centimetres in height – and carry a pattern of black spots on their chests that does not change from season to season during their adult life. To African penguins, these patterns are as unique as our fingerprints or the stripes on a zebra. In collaboration with Professor Les Underhill, Head of the Animal Demography Unit at the University of Cape Town, a real-time system has been developed by the Bristol team that can locate individual penguins by ‘recognising’ the patterns made by the spots on their chests. When finalised, the new technology will enable biologists to identify and monitor large numbers of diverse species cheaply, quickly and automatically. This in turn will revolutionise the precision, quantity and quality of population data available to ecologists and conservationists in a wide range of scenarios.
In order to record the movements of African penguins, camera systems, hidden on the penguins’ path between the sea and the nesting area, capture images and send a time-stamped version of them in a live stream to locally connected computers. The relevant areas of interest in each image – in this case the penguin’s chest patterns – are then transferred to local hotspots using a wireless network. From there, directional Yagi antennae submit the data to the central server network. The server extracts the biometrical patterns from the observed penguin snapshots and compares these to the population database. If the pattern is recognised, the penguin can be identified. If it is not recognised, the biologist decides whether a new penguin has been identified and, if so, adds it to the database for future comparison. As a result, the presence of particular penguins can be regularly confirmed at a certain time and location. An essential part of the research is focused on developing the ‘intelligent’ software that allows the system to make sense of complex camera images and interpret animals and their patterns as individual entities.
African penguins are quite small and carry a pattern of black spots on their chests that does not change from season to season during their adult life
Provided that a good image of a penguin can be extracted, the system can correctly identify the individual to within 98 per cent reliability. The main limitations at the moment are that passing penguins may be hidden behind others, or the lighting is poor. The team is currently working to overcome these problems both by combining images from intelligent pan-tilt-zoom cameras, and by using infrared imaging to acquire data both day and night. The basic image-recognition system has already been trialled with zebras and, in principle, could be extended to any species with complex surface patterns that remain constant over life.
This project will bring two remarkable changes to the study of animal demography and behaviours. The first is that it will be possible to obtain data remotely without ever having to handle the animals being studied; the second is that the sheer volume of data obtained will allow the testing of complex hypotheses where only small differences might be expected, thereby providing greater insight into the world inhabited by these endearing creatures and enabling more complete conservation strategies to be put in place.
Resources and copyright: Bristol University @